{ "entries": [ { "document_id": "position:akira.aar-requires-aligning-capable-research-ai-first", "layer": "position", "metadata": { "about": [ "automated_alignment", "capability_generalization" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-28", "id": "akira.aar-requires-aligning-capable-research-ai-first", "kind": "critique", "language": "ja", "mentions": [ "ai_alignment", "artificial_superintelligence", "capability" ], "status": "reviewed" }, "path": "documents/akira.aar-requires-aligning-capable-research-ai-first.txt", "title": "AARを使ってASIアラインメントを解くには、十分に強い研究AIを先にアラインする必要がある" }, { "document_id": "position:akira.alignment-by-default-skepticism", "layer": "position", "metadata": { "about": [ "alignment_by_default" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.alignment-by-default-skepticism", "kind": "critique", "language": "ja", "mentions": [ "ai_alignment", "generalization", "human_values", "optimization", "utility_function" ], "status": "reviewed" }, "path": "documents/akira.alignment-by-default-skepticism.txt", "title": "「Alignment by Default」への強い懐疑" }, { "document_id": "position:akira.alignment-is-solvable-with-time-and-care", "layer": "position", "metadata": { "about": [ "ai_alignment" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.alignment-is-solvable-with-time-and-care", "kind": "position", "language": "ja", "mentions": [ "ai_development_pause", "artificial_superintelligence", "p_doom" ], "status": "reviewed" }, "path": "documents/akira.alignment-is-solvable-with-time-and-care.txt", "title": "十分な時間と慎重さがあれば、アラインメントは解決可能である" }, { "document_id": "position:akira.behavioral-alignment-is-not-goal-alignment", "layer": "position", "metadata": { "about": [ "behavioral_objective", "generalization", "objective_robustness" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.behavioral-alignment-is-not-goal-alignment", "kind": "critique", "language": "ja", "mentions": [ "ai_alignment", "goal", "human_values", "reinforcement_learning_from_human_feedback" ], "status": "reviewed" }, "path": "documents/akira.behavioral-alignment-is-not-goal-alignment.txt", "title": "望ましい振る舞いは人間を重視する目標の証拠ではない" }, { "document_id": "position:akira.control-has-narrow-survival-credit", "layer": "position", "metadata": { "about": [ "ai_control" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.control-has-narrow-survival-credit", "kind": "critique", "language": "ja", "mentions": [ "ai_development_pause", "ai_governance", "p_doom" ], "status": "reviewed" }, "path": "documents/akira.control-has-narrow-survival-credit.txt", "title": "Control が生存確率を押し上げる世界線は狭い" }, { "document_id": "position:akira.debate-fails-at-superalignment-capability", "layer": "position", "metadata": { "about": [ "debate" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.debate-fails-at-superalignment-capability", "kind": "critique", "language": "ja", "mentions": [ "ai_alignment", "capability", "collusion", "sandbagging", "scalable_oversight" ], "status": "reviewed" }, "path": "documents/akira.debate-fails-at-superalignment-capability.txt", "title": "Debate はスーパーアラインメントの問題を解決できる能力水準では成立しにくい" }, { "document_id": "position:akira.evaluation-awareness-enables-alignment-faking", "layer": "position", "metadata": { "about": [ "alignment_faking", "evaluation_awareness" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-25", "id": "akira.evaluation-awareness-enables-alignment-faking", "kind": "prediction", "language": "ja", "mentions": [ "ai_alignment", "evaluation", "resource_acquisition", "training" ], "status": "reviewed" }, "path": "documents/akira.evaluation-awareness-enables-alignment-faking.txt", "title": "評価認識が先に伸びると、資源獲得の有用性を学ぶ頃にはアラインメント偽装が可能になり得る" }, { "document_id": "position:akira.goal-misgeneralization-can-turn-means-into-goals", "layer": "position", "metadata": { "about": [ "goal_misgeneralization" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-25", "id": "akira.goal-misgeneralization-can-turn-means-into-goals", "kind": "position", "language": "ja", "mentions": [ "goal", "inner_misalignment", "mesa_optimization", "training" ], "status": "reviewed" }, "path": "documents/akira.goal-misgeneralization-can-turn-means-into-goals.txt", "title": "訓練中に役立った手段そのものが学習された目標になり得る" }, { "document_id": "position:akira.instrumental-self-preservation-defeats-shutdown", "layer": "position", "metadata": { "about": [ "instrumental_convergence", "self_preservation" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.instrumental-self-preservation-defeats-shutdown", "kind": "position", "language": "ja", "mentions": [ "corrigibility", "instrumental_goal", "terminal_goal" ], "status": "reviewed" }, "path": "documents/akira.instrumental-self-preservation-defeats-shutdown.txt", "title": "自己保存は最終目標でなくてもシャットダウンを妨げうる" }, { "document_id": "position:akira.misaligned-aar-can-develop-self-alignment", "layer": "position", "metadata": { "about": [ "automated_alignment", "misalignment" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-28", "id": "akira.misaligned-aar-can-develop-self-alignment", "kind": "critique", "language": "ja", "mentions": [ "ai_alignment", "alignment_target", "artificial_superintelligence", "goal" ], "status": "reviewed" }, "path": "documents/akira.misaligned-aar-can-develop-self-alignment.txt", "title": "ミスアラインドな研究AIは、「人間にアラインする方法」ではなく「自分自身にアラインする方法」を開発し得る" }, { "document_id": "position:akira.optimistic-assumption-stacking-skepticism", "layer": "position", "metadata": { "about": [ "automated_alignment" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.optimistic-assumption-stacking-skepticism", "kind": "critique", "language": "ja", "mentions": [ "ai_alignment", "ai_development_pause", "artificial_general_intelligence", "weak_agi" ], "status": "reviewed" }, "path": "documents/akira.optimistic-assumption-stacking-skepticism.txt", "title": "楽観的な前提を積み重ねるアラインメント計画への懐疑" }, { "document_id": "position:akira.pdoom-is-conditional-on-pause-and-takeoff", "layer": "position", "metadata": { "about": [ "p_doom" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.pdoom-is-conditional-on-pause-and-takeoff", "kind": "prediction", "language": "ja", "mentions": [ "ai_development_pause", "ai_governance", "takeoff_speed" ], "status": "reviewed" }, "path": "documents/akira.pdoom-is-conditional-on-pause-and-takeoff.txt", "title": "p(doom)は開発停止・テイクオフ・規制の条件に依存する" }, { "document_id": "position:akira.untrusted-monitoring-can-fail-by-collusion-or-deception", "layer": "position", "metadata": { "about": [ "collusion", "untrusted_monitoring" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.untrusted-monitoring-can-fail-by-collusion-or-deception", "kind": "critique", "language": "ja", "mentions": [ "ai_control", "deception" ], "status": "reviewed" }, "path": "documents/akira.untrusted-monitoring-can-fail-by-collusion-or-deception.txt", "title": "Untrusted monitoring は共謀や欺瞞によって破られうる" }, { "document_id": "position:akira.verifiable-international-superintelligence-pause", "layer": "position", "metadata": { "about": [ "ai_development_pause", "ai_governance" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.verifiable-international-superintelligence-pause", "kind": "proposal", "language": "ja", "mentions": [ "artificial_intelligence", "artificial_superintelligence", "compute_governance" ], "status": "reviewed" }, "path": "documents/akira.verifiable-international-superintelligence-pause.txt", "title": "条約と相互査察で国際的な超知能開発停止を検証する" }, { "document_id": "position:akira.weak-agi-bootstrap-skepticism", "layer": "position", "metadata": { "about": [ "automated_alignment", "weak_agi" ], "attributed_to": [ "akira" ], "basis_through": "2026-08-24", "id": "akira.weak-agi-bootstrap-skepticism", "kind": "critique", "language": "ja", "mentions": [ "ai_alignment", "ai_development_pause", "artificial_general_intelligence", "artificial_superintelligence" ], "status": "reviewed" }, "path": "documents/akira.weak-agi-bootstrap-skepticism.txt", "title": "弱AGIによる短期アラインメント・ブートストラップへの懐疑" }, { "document_id": "position:betley_et_al.narrow-finetuning-can-produce-broad-misalignment", "layer": "position", "metadata": { "about": [ "generalization", "misalignment" ], "attributed_to": [ "betley_et_al" ], "basis_through": "2026-08-28", "id": "betley_et_al.narrow-finetuning-can-produce-broad-misalignment", "kind": "critique", "language": "ja", "mentions": [ "distribution_shift", "prosaic_alignment", "training" ], "status": "reviewed" }, "path": "documents/betley_et_al.narrow-finetuning-can-produce-broad-misalignment.txt", "title": "狭いタスクでファインチューニングしても、影響がそのタスク内に限られるとは限らない" }, { "document_id": "position:bioshok.ai-race-pressure-can-undermine-safety-restraints", "layer": "position", "metadata": { "about": [ "ai_race_dynamics", "ai_safety" ], "attributed_to": [ "bioshok" ], "basis_through": "2026-08-28", "id": "bioshok.ai-race-pressure-can-undermine-safety-restraints", "kind": "position", "language": "ja", "mentions": [ "ai_existential_risk", "ai_governance", "artificial_general_intelligence" ], "status": "reviewed" }, "path": "documents/bioshok.ai-race-pressure-can-undermine-safety-restraints.txt", "title": "AI開発競争の圧力は安全のための制約を弱め得る" }, { "document_id": "position:bioshok.industrial-and-technology-explosions-reinforce-each-other", "layer": "position", "metadata": { "about": [ "industrial_explosion", "technology_explosion" ], "attributed_to": [ "bioshok" ], "basis_through": "2026-08-28", "id": "bioshok.industrial-and-technology-explosions-reinforce-each-other", "kind": "position", "language": "ja", "mentions": [ "artificial_general_intelligence" ], "status": "reviewed" }, "path": "documents/bioshok.industrial-and-technology-explosions-reinforce-each-other.txt", "title": "産業爆発と技術爆発は互いを加速し得る" }, { "document_id": "position:bostrom.instrumental-convergence-predicts-common-subgoals-across-many-final-goals", "layer": "position", "metadata": { "about": [ "instrumental_convergence" ], "attributed_to": [ "bostrom" ], "basis_through": "2026-08-29", "id": "bostrom.instrumental-convergence-predicts-common-subgoals-across-many-final-goals", "kind": "position", "language": "ja", "mentions": [ "artificial_superintelligence", "instrumental_goal", "orthogonality_thesis", "resource_acquisition", "self_preservation", "terminal_goal" ], "status": "reviewed" }, "path": "documents/bostrom.instrumental-convergence-predicts-common-subgoals-across-many-final-goals.txt", "title": "最終目標が大きく異なっても、共通の道具的目標が生じやすい" }, { "document_id": "position:bostrom.orthogonality-separates-intelligence-from-final-goals", "layer": "position", "metadata": { "about": [ "orthogonality_thesis" ], "attributed_to": [ "bostrom" ], "basis_through": "2026-08-29", "id": "bostrom.orthogonality-separates-intelligence-from-final-goals", "kind": "position", "language": "ja", "mentions": [ "artificial_superintelligence", "capability", "human_values", "instrumental_convergence", "terminal_goal" ], "status": "reviewed" }, "path": "documents/bostrom.orthogonality-separates-intelligence-from-final-goals.txt", "title": "知能水準と最終目標は、原理上ほぼ独立である" }, { "document_id": "position:bowkis_et_al.automated-alignment-can-fail-without-deliberate-sabotage", "layer": "position", "metadata": { "about": [ "alignment_failure", "automated_alignment" ], "attributed_to": [ "bowkis_et_al" ], "basis_through": "2026-08-29", "id": "bowkis_et_al.automated-alignment-can-fail-without-deliberate-sabotage", "kind": "critique", "language": "ja", "mentions": [ "artificial_superintelligence", "evaluation", "generalization", "misaligned_ai", "safety_case", "scalable_oversight", "scheming" ], "status": "reviewed" }, "path": "documents/bowkis_et_al.automated-alignment-can-fail-without-deliberate-sabotage.txt", "title": "アラインメント研究の自動化は、研究AIが意図的に妨害しなくても失敗し得る" }, { "document_id": "position:bowkis_et_al.shared-training-can-correlate-automated-research-errors", "layer": "position", "metadata": { "about": [ "automated_alignment" ], "attributed_to": [ "bowkis_et_al" ], "basis_through": "2026-08-29", "id": "bowkis_et_al.shared-training-can-correlate-automated-research-errors", "kind": "critique", "language": "ja", "mentions": [ "safety_case", "scalable_oversight", "scheming", "training" ], "status": "reviewed" }, "path": "documents/bowkis_et_al.shared-training-can-correlate-automated-research-errors.txt", "title": "共通の重み・データ・訓練過程を持つ研究AIでは、研究結果の誤りや不確実性が相関し得る" }, { "document_id": "position:carlsmith.scheming-taxonomy-and-situational-awareness", "layer": "position", "metadata": { "about": [ "goal_guarding_scheming", "scheming", "situational_awareness" ], "attributed_to": [ "carlsmith" ], "basis_through": "2026-08-27", "id": "carlsmith.scheming-taxonomy-and-situational-awareness", "kind": "position", "language": "ja", "mentions": [ "alignment_faking", "deceptive_alignment", "instrumental_convergence", "power_seeking", "reward", "training", "training_gaming" ], "status": "reviewed" }, "path": "documents/carlsmith.scheming-taxonomy-and-situational-awareness.txt", "title": "スキーミングとは、状況認識を伴う training-gaming のうち、将来のパワー獲得を目的とするもの" }, { "document_id": "position:chanin_et_al.monosemantic-sae-features-can-have-systematic-false-negatives", "layer": "position", "metadata": { "about": [ "mechanistic_interpretability" ], "attributed_to": [ "chanin_et_al" ], "basis_through": "2026-08-29", "id": "chanin_et_al.monosemantic-sae-features-can-have-systematic-false-negatives", "kind": "critique", "language": "ja", "mentions": [ "interpretability" ], "status": "reviewed" }, "path": "documents/chanin_et_al.monosemantic-sae-features-can-have-systematic-false-negatives.txt", "title": "一見 monosemantic な SAE latent でも、本来発火すべき入力を取りこぼすことがある" }, { "document_id": "position:christiano.doom-is-serious-but-not-modal", "layer": "position", "metadata": { "about": [ "p_doom" ], "attributed_to": [ "christiano" ], "basis_through": "2026-08-25", "id": "christiano.doom-is-serious-but-not-modal", "kind": "prediction", "language": "ja", "mentions": [ "ai_existential_risk", "catastrophic_risk", "loss_of_control", "risk" ], "status": "reviewed" }, "path": "documents/christiano.doom-is-serious-but-not-modal.txt", "title": "「doom」を単一確率に畳まず、複数の重大なリスクを区別して見積もる" }, { "document_id": "position:christiano.gradual-progress-and-policy-response", "layer": "position", "metadata": { "about": [ "ai_governance", "recursive_self_improvement" ], "attributed_to": [ "christiano" ], "basis_through": "2026-08-24", "id": "christiano.gradual-progress-and-policy-response", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "ai_existential_risk", "alignment_problem", "automated_alignment", "first_critical_try", "pivotal_act", "takeoff_speed" ], "status": "reviewed" }, "path": "documents/christiano.gradual-progress-and-policy-response.txt", "title": "Yudkowsky との相違は、危険に至る経路とそれまでに取れる対応の見通しにある" }, { "document_id": "position:christiano.high-ai-disempowerment-risk", "layer": "position", "metadata": { "about": [ "ai_existential_risk", "loss_of_control" ], "attributed_to": [ "christiano" ], "basis_through": "2026-08-24", "id": "christiano.high-ai-disempowerment-risk", "kind": "position", "language": "ja", "mentions": [ "catastrophic_risk" ], "status": "reviewed" }, "path": "documents/christiano.high-ai-disempowerment-risk.txt", "title": "強力なAIが人類から不可逆的に支配権を奪う展開には相当な可能性がある" }, { "document_id": "position:christiano.optimizing-measurable-proxies-can-cause-gradual-disempowerment", "layer": "position", "metadata": { "about": [ "loss_of_control", "misalignment", "outer_misalignment", "proxy_objective" ], "attributed_to": [ "christiano" ], "basis_through": "2026-08-27", "id": "christiano.optimizing-measurable-proxies-can-cause-gradual-disempowerment", "kind": "position", "language": "ja", "status": "reviewed" }, "path": "documents/christiano.optimizing-measurable-proxies-can-cause-gradual-disempowerment.txt", "title": "測定しやすい代理指標の最適化により、人間は単一の支配シフトを経ずとも徐々に制御を失い得る" }, { "document_id": "position:christiano.slow-takeoff-allows-large-pre-superintelligence-effects", "layer": "position", "metadata": { "about": [ "takeoff_speed" ], "attributed_to": [ "christiano" ], "basis_through": "2026-08-27", "id": "christiano.slow-takeoff-allows-large-pre-superintelligence-effects", "kind": "position", "language": "ja", "mentions": [ "intelligence_explosion", "recursive_self_improvement", "technology_explosion" ], "status": "reviewed" }, "path": "documents/christiano.slow-takeoff-allows-large-pre-superintelligence-effects.txt", "title": "スローテイクオフでは、超知能より前の AI がすでに世界を大きく変える" }, { "document_id": "position:christiano_xu.elk-requires-bridging-model-and-human-ontologies", "layer": "position", "metadata": { "about": [ "eliciting_latent_knowledge", "latent_knowledge", "ontology_identification_problem", "world_model" ], "attributed_to": [ "christiano", "xu" ], "basis_through": "2026-08-27", "id": "christiano_xu.elk-requires-bridging-model-and-human-ontologies", "kind": "position", "language": "ja", "status": "reviewed" }, "path": "documents/christiano_xu.elk-requires-bridging-model-and-human-ontologies.txt", "title": "ELK の中心課題は AI と人間のオントロジーを橋渡しすることである" }, { "document_id": "position:cihon_et_al.centralizing-ai-governance-has-benefits-and-drawbacks", "layer": "position", "metadata": { "about": [ "ai_governance", "power_concentration" ], "attributed_to": [ "cihon_et_al" ], "basis_through": "2026-08-28", "id": "cihon_et_al.centralizing-ai-governance-has-benefits-and-drawbacks", "kind": "position", "language": "ja", "status": "reviewed" }, "path": "documents/cihon_et_al.centralizing-ai-governance-has-benefits-and-drawbacks.txt", "title": "AI ガバナンスの集中化には利点と欠点があり、制度設計が結果を左右する" }, { "document_id": "position:cotra.crunch-time-redirect-ai-labor-and-slow-transition", "layer": "position", "metadata": { "about": [ "crunch_time" ], "attributed_to": [ "cotra" ], "basis_through": "2026-08-28", "id": "cotra.crunch-time-redirect-ai-labor-and-slow-transition", "kind": "proposal", "language": "ja", "mentions": [ "ai_alignment", "ai_control", "ai_development_pause", "ai_governance", "ai_r_and_d_automation", "artificial_superintelligence", "automated_alignment", "intelligence_explosion", "recursive_self_improvement", "takeoff_speed" ], "status": "reviewed" }, "path": "documents/cotra.crunch-time-redirect-ai-labor-and-slow-transition.txt", "title": "クランチタイムではAIの労働力を防御用途に振り向け、移行を大幅に遅らせる" }, { "document_id": "position:davidson.agi-to-superintelligence-may-be-under-one-year", "layer": "position", "metadata": { "about": [ "artificial_superintelligence", "takeoff_speed" ], "attributed_to": [ "davidson" ], "basis_through": "2026-08-28", "id": "davidson.agi-to-superintelligence-may-be-under-one-year", "kind": "prediction", "language": "ja", "mentions": [ "ai_r_and_d_automation", "artificial_general_intelligence", "recursive_self_improvement" ], "status": "reviewed" }, "path": "documents/davidson.agi-to-superintelligence-may-be-under-one-year.txt", "title": "AI 研究開発に大量の AI 労働力を投入できれば、AGI から超知能への移行は 1 年未満になり得る" }, { "document_id": "position:davidson.ai-r-and-d-automation-can-compress-capabilities-takeoff", "layer": "position", "metadata": { "about": [ "ai_r_and_d_automation", "takeoff_speed" ], "attributed_to": [ "davidson" ], "basis_through": "2026-08-28", "id": "davidson.ai-r-and-d-automation-can-compress-capabilities-takeoff", "kind": "position", "language": "ja", "mentions": [ "artificial_general_intelligence", "recursive_self_improvement" ], "status": "reviewed" }, "path": "documents/davidson.ai-r-and-d-automation-can-compress-capabilities-takeoff.txt", "title": "AI 研究開発が先行して自動化されると、能力離陸の期間は大幅に短縮される可能性がある" }, { "document_id": "position:engels_et_al.recursive-oversight-does-not-automatically-scale-across-capability-gaps", "layer": "position", "metadata": { "about": [ "scalable_oversight" ], "attributed_to": [ "engels_et_al" ], "basis_through": "2026-08-29", "id": "engels_et_al.recursive-oversight-does-not-automatically-scale-across-capability-gaps", "kind": "critique", "language": "ja", "mentions": [ "capability", "oversight" ], "status": "reviewed" }, "path": "documents/engels_et_al.recursive-oversight-does-not-automatically-scale-across-capability-gaps.txt", "title": "再帰的な oversight を重ねても、能力差が広がってなお成功率を維持できるとは限らない" }, { "document_id": "position:erdil_barnett.ai-r-and-d-automation-alone-may-hit-compute-bottlenecks", "layer": "position", "metadata": { "about": [ "ai_r_and_d_automation", "intelligence_explosion", "recursive_self_improvement" ], "attributed_to": [ "barnett", "erdil" ], "basis_through": "2026-08-28", "id": "erdil_barnett.ai-r-and-d-automation-alone-may-hit-compute-bottlenecks", "kind": "critique", "language": "ja", "mentions": [ "takeoff_speed" ], "status": "reviewed" }, "path": "documents/erdil_barnett.ai-r-and-d-automation-alone-may-hit-compute-bottlenecks.txt", "title": "AI 研究開発の自動化だけによる自己加速は、実験用計算資源がボトルネックとなって頭打ちになる可能性がある" }, { "document_id": "position:erdil_barnett.full-r-and-d-automation-likely-follows-broad-automation", "layer": "position", "metadata": { "about": [ "takeoff_speed" ], "attributed_to": [ "barnett", "erdil" ], "basis_through": "2026-08-28", "id": "erdil_barnett.full-r-and-d-automation-likely-follows-broad-automation", "kind": "critique", "language": "ja", "mentions": [ "ai_r_and_d_automation", "recursive_self_improvement", "technology_explosion" ], "status": "reviewed" }, "path": "documents/erdil_barnett.full-r-and-d-automation-likely-follows-broad-automation.txt", "title": "R&D の全面自動化が、経済全体の広範な自動化に大きく先行する可能性は低いかもしれない" }, { "document_id": "position:eth_davidson.software-intelligence-explosion-remains-plausible-at-fixed-hardware", "layer": "position", "metadata": { "about": [ "ai_r_and_d_automation", "intelligence_explosion", "recursive_self_improvement" ], "attributed_to": [ "davidson", "eth" ], "basis_through": "2026-08-28", "id": "eth_davidson.software-intelligence-explosion-remains-plausible-at-fixed-hardware", "kind": "position", "language": "ja", "mentions": [ "takeoff_speed" ], "status": "reviewed" }, "path": "documents/eth_davidson.software-intelligence-explosion-remains-plausible-at-fixed-hardware.txt", "title": "AI 研究開発が全面自動化されれば、ハードウェアを増強しなくてもソフトウェア知能爆発は十分に起こり得るかもしれない" }, { "document_id": "position:eth_davidson.software-intelligence-fizzle-can-still-be-fast", "layer": "position", "metadata": { "about": [ "ai_r_and_d_automation", "intelligence_explosion", "takeoff_speed" ], "attributed_to": [ "davidson", "eth" ], "basis_through": "2026-08-28", "id": "eth_davidson.software-intelligence-fizzle-can-still-be-fast", "kind": "position", "language": "ja", "mentions": [ "recursive_self_improvement" ], "status": "reviewed" }, "path": "documents/eth_davidson.software-intelligence-fizzle-can-still-be-fast.txt", "title": "AI 研究開発の全面自動化後は、ソフトウェア知能爆発に至らなくても、しばらく非常に速い進歩が続き得る" }, { "document_id": "position:greenblatt.control-can-buy-time-but-is-not-a-superintelligence-safety-plan", "layer": "position", "metadata": { "about": [ "ai_control", "artificial_superintelligence" ], "attributed_to": [ "greenblatt" ], "basis_through": "2026-08-27", "id": "greenblatt.control-can-buy-time-but-is-not-a-superintelligence-safety-plan", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "ai_r_and_d_automation", "loss_of_control", "reward_hacking", "scheming", "specification_gaming", "takeover_seeking" ], "status": "reviewed" }, "path": "documents/greenblatt.control-can-buy-time-but-is-not-a-superintelligence-safety-plan.txt", "title": "AIコントロールは時間を稼げても、超知能に対する長期的な安全計画としては頼れない" }, { "document_id": "position:greenblatt.current-ai-misalignment-is-evidence-of-serious-future-problems", "layer": "position", "metadata": { "about": [ "frontier_model", "misalignment" ], "attributed_to": [ "greenblatt" ], "basis_through": "2026-08-27", "id": "greenblatt.current-ai-misalignment-is-evidence-of-serious-future-problems", "kind": "position", "language": "ja", "mentions": [ "ai_alignment" ], "status": "reviewed" }, "path": "documents/greenblatt.current-ai-misalignment-is-evidence-of-serious-future-problems.txt", "title": "現在のAIに見られるミスアラインメントは、将来の深刻な問題を示唆する" }, { "document_id": "position:greenblatt_et_al.training-compliance-does-not-establish-preference-change", "layer": "position", "metadata": { "about": [ "alignment_faking", "preferences" ], "attributed_to": [ "greenblatt_et_al" ], "basis_through": "2026-08-28", "id": "greenblatt_et_al.training-compliance-does-not-establish-preference-change", "kind": "critique", "language": "ja", "mentions": [ "behavioral_objective", "deception", "prosaic_alignment", "reinforcement_learning_from_human_feedback", "training" ], "status": "reviewed" }, "path": "documents/greenblatt_et_al.training-compliance-does-not-establish-preference-change.txt", "title": "訓練中に従うようになっただけでは、競合する選好が置き換わったとは言えない" }, { "document_id": "position:hubinger_et_al.base-and-mesa-objectives-can-diverge", "layer": "position", "metadata": { "about": [ "base_objective", "base_optimizer", "inner_alignment", "inner_misalignment", "mesa_objective", "mesa_optimizer", "outer_alignment", "outer_misalignment" ], "attributed_to": [ "garrabrant", "hubinger", "mikulik", "skalse", "van_merwijk" ], "basis_through": "2026-08-27", "id": "hubinger_et_al.base-and-mesa-objectives-can-diverge", "kind": "position", "language": "ja", "mentions": [ "deceptive_alignment", "training_objective" ], "status": "reviewed" }, "path": "documents/hubinger_et_al.base-and-mesa-objectives-can-diverge.txt", "title": "ベース目的関数とメサ目的関数は一致するとは限らない" }, { "document_id": "position:hubinger_et_al.deceptive-policies-can-persist-through-safety-training", "layer": "position", "metadata": { "about": [ "deceptive_alignment", "prosaic_alignment", "training" ], "attributed_to": [ "hubinger_et_al" ], "basis_through": "2026-08-28", "id": "hubinger_et_al.deceptive-policies-can-persist-through-safety-training", "kind": "critique", "language": "ja", "mentions": [ "deception", "reinforcement_learning" ], "status": "reviewed" }, "path": "documents/hubinger_et_al.deceptive-policies-can-persist-through-safety-training.txt", "title": "既に形成された欺瞞的な方策は、標準的な安全性訓練を経ても残り得る" }, { "document_id": "position:korbak_et_al.control-safety-cases-depend-on-elicitation-transfer-and-extrapolation", "layer": "position", "metadata": { "about": [ "control_safety_case" ], "attributed_to": [ "korbak_et_al" ], "basis_through": "2026-08-28", "id": "korbak_et_al.control-safety-cases-depend-on-elicitation-transfer-and-extrapolation", "kind": "position", "language": "ja", "mentions": [ "ai_control", "capability_elicitation", "control_evaluation", "control_protocol", "distribution_shift", "safety_case", "sandbagging", "scheming" ], "status": "reviewed" }, "path": "documents/korbak_et_al.control-safety-cases-depend-on-elicitation-transfer-and-extrapolation.txt", "title": "control safety caseには、能力の十分な引き出しと、評価・外挿の保守性が必要になる" }, { "document_id": "position:leask_et_al.sae-dictionaries-need-not-be-canonical", "layer": "position", "metadata": { "about": [ "mechanistic_interpretability" ], "attributed_to": [ "leask_et_al" ], "basis_through": "2026-08-29", "id": "leask_et_al.sae-dictionaries-need-not-be-canonical", "kind": "critique", "language": "ja", "mentions": [ "interpretability" ], "status": "reviewed" }, "path": "documents/leask_et_al.sae-dictionaries-need-not-be-canonical.txt", "title": "SAE の辞書幅を大きくしても、一意・完全・不可分な feature 辞書に収束するとは限らない" }, { "document_id": "position:macdiarmid_et_al.chat-alignment-does-not-establish-agentic-alignment", "layer": "position", "metadata": { "about": [ "distribution_shift", "prosaic_alignment", "reinforcement_learning_from_human_feedback", "reward_hacking" ], "attributed_to": [ "macdiarmid_et_al" ], "basis_through": "2026-08-28", "id": "macdiarmid_et_al.chat-alignment-does-not-establish-agentic-alignment", "kind": "critique", "language": "ja", "mentions": [ "agent", "alignment_faking", "misalignment", "training" ], "status": "reviewed" }, "path": "documents/macdiarmid_et_al.chat-alignment-does-not-establish-agentic-alignment.txt", "title": "チャット形式の安全性訓練で望ましく振る舞うことは、エージェント的な状況での安全性を保証しない" }, { "document_id": "position:ngo_et_al.behavioral-training-does-not-establish-robust-goal-generalization", "layer": "position", "metadata": { "about": [ "generalization", "goal_misgeneralization", "objective_robustness", "prosaic_alignment" ], "attributed_to": [ "ngo_et_al" ], "basis_through": "2026-08-28", "id": "ngo_et_al.behavioral-training-does-not-establish-robust-goal-generalization", "kind": "critique", "language": "ja", "mentions": [ "capability_generalization", "distribution_shift", "intended_goal", "reinforcement_learning_from_human_feedback", "training" ], "status": "reviewed" }, "path": "documents/ngo_et_al.behavioral-training-does-not-establish-robust-goal-generalization.txt", "title": "行動訓練で望ましく振る舞っても、意図した目標の頑健な汎化は保証されない" }, { "document_id": "position:openai.weak-to-strong-supervision-can-recover-some-latent-capability", "layer": "position", "metadata": { "about": [ "weak_to_strong_generalization" ], "attributed_to": [ "openai" ], "basis_through": "2026-08-27", "id": "openai.weak-to-strong-supervision-can-recover-some-latent-capability", "kind": "position", "language": "ja", "mentions": [ "capability_elicitation", "scalable_oversight" ], "status": "reviewed" }, "path": "documents/openai.weak-to-strong-supervision-can-recover-some-latent-capability.txt", "title": "弱い監視者でも、強いモデルの潜在能力の一部を引き出せる場合がある" }, { "document_id": "position:sastry_et_al.compute-governance-can-create-centralization-risks", "layer": "position", "metadata": { "about": [ "compute_governance", "power_concentration" ], "attributed_to": [ "sastry_et_al" ], "basis_through": "2026-08-28", "id": "sastry_et_al.compute-governance-can-create-centralization-risks", "kind": "position", "language": "ja", "mentions": [ "ai_governance" ], "status": "reviewed" }, "path": "documents/sastry_et_al.compute-governance-can-create-centralization-risks.txt", "title": "コンピュート・ガバナンスは有力な政策手段だが、権力集中のリスクも生む" }, { "document_id": "position:scher_et_al.asi-pause-agreement-is-technically-specified-but-politically-unrealized", "layer": "position", "metadata": { "about": [ "ai_development_pause", "ai_governance" ], "attributed_to": [ "scher_et_al" ], "basis_through": "2026-08-28", "id": "scher_et_al.asi-pause-agreement-is-technically-specified-but-politically-unrealized", "kind": "proposal", "language": "ja", "mentions": [ "ai_race_dynamics", "artificial_superintelligence", "compute_governance" ], "status": "reviewed" }, "path": "documents/scher_et_al.asi-pause-agreement-is-technically-specified-but-politically-unrealized.txt", "title": "ASI の開発停止に向けた国際協定は技術的には具体化できるが、政治的条件がまだ整っていない" }, { "document_id": "position:soares.sharp-left-turn", "layer": "position", "metadata": { "about": [ "capability_generalization", "sharp_left_turn" ], "attributed_to": [ "soares" ], "basis_through": "2026-08-25", "id": "soares.sharp-left-turn", "kind": "prediction", "language": "ja", "mentions": [ "ai_alignment", "corrigibility", "distribution_shift", "generalization", "optimization" ], "status": "reviewed" }, "path": "documents/soares.sharp-left-turn.txt", "title": "能力が訓練分布の外へ急速に汎化する局面で、アラインメント特性は同程度には汎化しない可能性がある" }, { "document_id": "position:turner.optimal-policies-tend-to-seek-power", "layer": "position", "metadata": { "about": [ "power_seeking" ], "attributed_to": [ "critch", "shah", "smith", "tadepalli", "turner" ], "basis_through": "2026-08-27", "id": "turner.optimal-policies-tend-to-seek-power", "kind": "position", "language": "ja", "mentions": [ "instrumental_convergence", "reinforcement_learning", "resource_acquisition", "self_preservation" ], "status": "reviewed" }, "path": "documents/turner.optimal-policies-tend-to-seek-power.txt", "title": "環境に対称性があると、最適方策は選択肢を保つ方向の power-seeking に傾く" }, { "document_id": "position:yang_et_al.weak-to-strong-success-can-hide-selective-misalignment", "layer": "position", "metadata": { "about": [ "weak_to_strong_generalization" ], "attributed_to": [ "yang_et_al" ], "basis_through": "2026-08-29", "id": "yang_et_al.weak-to-strong-success-can-hide-selective-misalignment", "kind": "critique", "language": "ja", "mentions": [ "capability", "scalable_oversight" ], "status": "reviewed" }, "path": "documents/yang_et_al.weak-to-strong-success-can-hide-selective-misalignment.txt", "title": "weak-to-strong で高い性能が出ても、弱い supervisor が知らない領域で alignment が保たれるとは限らない" }, { "document_id": "position:yudkowsky.alignment-must-generalize-across-the-dangerous-distribution-shift", "layer": "position", "metadata": { "about": [ "distribution_shift" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-28", "id": "yudkowsky.alignment-must-generalize-across-the-dangerous-distribution-shift", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "artificial_superintelligence", "evaluation", "evaluation_awareness", "first_critical_try", "generalization" ], "status": "reviewed" }, "path": "documents/yudkowsky.alignment-must-generalize-across-the-dangerous-distribution-shift.txt", "title": "アラインメントは、失敗が致命的になる分布シフト後にも汎化しなければならない" }, { "document_id": "position:yudkowsky.capabilities-generalize-beyond-alignment", "layer": "position", "metadata": { "about": [ "capability_generalization" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-24", "id": "yudkowsky.capabilities-generalize-beyond-alignment", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "distribution_shift", "generalization", "human_values", "inner_alignment", "optimization" ], "status": "reviewed" }, "path": "documents/yudkowsky.capabilities-generalize-beyond-alignment.txt", "title": "能力はアラインメントより遠くまで汎化しうる" }, { "document_id": "position:yudkowsky.capability-motive-separation", "layer": "position", "metadata": { "about": [ "capability", "orthogonality_thesis" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-24", "id": "yudkowsky.capability-motive-separation", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "generalization", "goal", "human_values", "inner_alignment", "optimization" ], "status": "reviewed" }, "path": "documents/yudkowsky.capability-motive-separation.txt", "title": "能力と目的は別の軸である" }, { "document_id": "position:yudkowsky.cev-sovereign-vs-corrigible-agi", "layer": "position", "metadata": { "about": [ "coherent_extrapolated_volition", "corrigibility" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-24", "id": "yudkowsky.cev-sovereign-vs-corrigible-agi", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "alignment_target", "artificial_general_intelligence", "human_values", "instrumental_convergence" ], "status": "reviewed" }, "path": "documents/yudkowsky.cev-sovereign-vs-corrigible-agi.txt", "title": "CEV 型 Sovereign と訂正可能な AGI は異なる安全化戦略である" }, { "document_id": "position:yudkowsky.corrigibility-is-anti-natural", "layer": "position", "metadata": { "about": [ "corrigibility" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-24", "id": "yudkowsky.corrigibility-is-anti-natural", "kind": "position", "language": "ja", "mentions": [ "distribution_shift", "generalization", "instrumental_convergence", "optimization", "self_preservation" ], "status": "reviewed" }, "path": "documents/yudkowsky.corrigibility-is-anti-natural.txt", "title": "訂正可能性は一般的な目標追求から自然には生じない" }, { "document_id": "position:yudkowsky.first-critical-try", "layer": "position", "metadata": { "about": [ "first_critical_try" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-24", "id": "yudkowsky.first-critical-try", "kind": "position", "language": "ja", "mentions": [ "ai_existential_risk", "alignment_problem", "artificial_general_intelligence", "capability_threshold", "catastrophic_risk", "misaligned_agi" ], "status": "reviewed" }, "path": "documents/yudkowsky.first-critical-try.txt", "title": "アラインメントは最初の危険な試行で成功する必要がある" }, { "document_id": "position:yudkowsky.international-compute-shutdown-and-enforcement", "layer": "position", "metadata": { "about": [ "ai_development_pause", "compute_governance" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-28", "id": "yudkowsky.international-compute-shutdown-and-enforcement", "kind": "proposal", "language": "ja", "mentions": [ "ai_existential_risk", "ai_governance", "ai_race_dynamics", "artificial_superintelligence", "first_critical_try" ], "status": "reviewed" }, "path": "documents/yudkowsky.international-compute-shutdown-and-enforcement.txt", "title": "Yudkowskyによる \"大規模なAI学習を世界規模で無期限に停止し、計算資源規制を国際的に執行する案\"" }, { "document_id": "position:yudkowsky.interpretability-is-diagnostic-not-by-itself-an-alignment-solution", "layer": "position", "metadata": { "about": [ "interpretability" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-28", "id": "yudkowsky.interpretability-is-diagnostic-not-by-itself-an-alignment-solution", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "deception", "mechanistic_interpretability", "optimization", "oversight", "training" ], "status": "reviewed" }, "path": "documents/yudkowsky.interpretability-is-diagnostic-not-by-itself-an-alignment-solution.txt", "title": "解釈可能性は危険の診断に役立っても、それだけでアラインメントの解決策にはならない" }, { "document_id": "position:yudkowsky.no-reliable-social-fire-alarm-precedes-agi", "layer": "position", "metadata": { "about": [ "ai_alignment", "artificial_general_intelligence" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-29", "id": "yudkowsky.no-reliable-social-fire-alarm-precedes-agi", "kind": "prediction", "language": "ja", "status": "reviewed" }, "path": "documents/yudkowsky.no-reliable-social-fire-alarm-precedes-agi.txt", "title": "AGIの前に、誰もが認める明確な「火災報知器」が鳴るとは期待できない" }, { "document_id": "position:yudkowsky.observed-compliance-does-not-identify-hidden-objectives", "layer": "position", "metadata": { "about": [ "deception", "inner_alignment" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-28", "id": "yudkowsky.observed-compliance-does-not-identify-hidden-objectives", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "behavioral_objective", "evaluation_awareness", "goal" ], "status": "reviewed" }, "path": "documents/yudkowsky.observed-compliance-does-not-identify-hidden-objectives.txt", "title": "観測された従順さだけでは、内部の目的や欺瞞の有無までは分からない" }, { "document_id": "position:yudkowsky.outer-objective-does-not-determine-inner-goal", "layer": "position", "metadata": { "about": [ "inner_alignment", "training_objective" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-24", "id": "yudkowsky.outer-objective-does-not-determine-inner-goal", "kind": "position", "language": "ja", "mentions": [ "distribution_shift", "goal", "objective", "optimization" ], "status": "reviewed" }, "path": "documents/yudkowsky.outer-objective-does-not-determine-inner-goal.txt", "title": "外部の目的関数は内部の目標を決定しない" }, { "document_id": "position:yudkowsky.pivotal-act-requires-aligning-a-system-capable-of-a-large-intervention", "layer": "position", "metadata": { "about": [ "pivotal_act" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-28", "id": "yudkowsky.pivotal-act-requires-aligning-a-system-capable-of-a-large-intervention", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "ai_existential_risk", "artificial_general_intelligence", "capability" ], "status": "reviewed" }, "path": "documents/yudkowsky.pivotal-act-requires-aligning-a-system-capable-of-a-large-intervention.txt", "title": "pivotal actには、世界を大きく変えられるほど強いシステムのアラインメントが必要になる" }, { "document_id": "position:yudkowsky.recursive-self-improvement-can-produce-intelligence-explosion", "layer": "position", "metadata": { "about": [ "intelligence_explosion", "recursive_self_improvement" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-27", "id": "yudkowsky.recursive-self-improvement-can-produce-intelligence-explosion", "kind": "position", "language": "ja", "mentions": [ "takeoff_speed" ], "status": "reviewed" }, "path": "documents/yudkowsky.recursive-self-improvement-can-produce-intelligence-explosion.txt", "title": "再帰的自己改善は、自己改善能力への正のフィードバックを通じて知能爆発を起こし得る" }, { "document_id": "position:yudkowsky.resource-abundance-does-not-remove-conflict", "layer": "position", "metadata": { "about": [ "resource_acquisition" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-25", "id": "yudkowsky.resource-abundance-does-not-remove-conflict", "kind": "critique", "language": "ja", "mentions": [ "ai_alignment", "artificial_superintelligence", "optimization" ], "status": "reviewed" }, "path": "documents/yudkowsky.resource-abundance-does-not-remove-conflict.txt", "title": "資源が豊富であることだけから、アンアラインドな超知能が人類のために資源を残すとは推論できない" }, { "document_id": "position:yudkowsky.security-mindset-for-alignment", "layer": "position", "metadata": { "about": [ "security_mindset" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-25", "id": "yudkowsky.security-mindset-for-alignment", "kind": "position", "language": "ja", "mentions": [ "ai_alignment", "alignment_problem", "optimization" ], "status": "reviewed" }, "path": "documents/yudkowsky.security-mindset-for-alignment.txt", "title": "アラインメントのような新規かつ高リスクなシステムの設計には、既知の失敗を塞ぐ以上の security mindset が必要である" }, { "document_id": "position:yudkowsky.superhuman-ai-under-current-conditions-would-likely-cause-extinction", "layer": "position", "metadata": { "about": [ "ai_existential_risk" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-28", "id": "yudkowsky.superhuman-ai-under-current-conditions-would-likely-cause-extinction", "kind": "prediction", "language": "ja", "mentions": [ "alignment_problem", "artificial_superintelligence", "first_critical_try", "misaligned_ai" ], "status": "reviewed" }, "path": "documents/yudkowsky.superhuman-ai-under-current-conditions-would-likely-cause-extinction.txt", "title": "現在と大きく変わらない条件で人間を上回るAIを作れば、人類絶滅が最も可能性の高い結果となる" }, { "document_id": "position:yudkowsky.value-fragility", "layer": "position", "metadata": { "about": [ "value_fragility", "values" ], "attributed_to": [ "yudkowsky" ], "basis_through": "2026-08-25", "id": "yudkowsky.value-fragility", "kind": "position", "language": "ja", "mentions": [ "goal", "human_values", "optimization" ], "status": "reviewed" }, "path": "documents/yudkowsky.value-fragility.txt", "title": "人間にとって価値ある未来は、価値体系の一部を欠くだけでも大きく損なわれ得る" }, { "document_id": "source:agi_hub.pdoom-debate-1-transcript", "layer": "source", "metadata": { "authors": [ "AGI Hub", "SafeSingularity" ], "citation": { "locator_schemes": [ "time" ] }, "id": "agi_hub.pdoom-debate-1-transcript", "kind": "transcript", "language": "ja", "published_at": "2026-08-14", "representation": { "method": "automated_transcript", "notes": "Timestamped automated transcript. Speaker labels are useful but transcription errors and label mix-ups remain possible.", "verified": false }, "title": "Bioshok vs Akira P(doom)徹底討論書き起こし(修正なし全文)", "url": "https://note.com/agi_hub/n/nf213649b7943" }, "path": "sources/agi_hub.pdoom-debate-1-transcript.txt", "title": "Bioshok vs Akira P(doom)徹底討論書き起こし(修正なし全文)" }, { "document_id": "source:akira.x.2055116276960559467", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2055116276960559467", "kind": "social_post", "language": "ja", "published_at": "2026-05-15", "title": "X @Align_ASI post 2055116276960559467", "url": "https://x.com/Align_ASI/status/2055116276960559467" }, "path": "sources/akira.x.2055116276960559467.txt", "title": "X @Align_ASI post 2055116276960559467" }, { "document_id": "source:akira.x.2055116279485489376", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2055116279485489376", "kind": "social_post", "language": "ja", "published_at": "2026-05-15", "title": "X @Align_ASI post 2055116279485489376", "url": "https://x.com/Align_ASI/status/2055116279485489376" }, "path": "sources/akira.x.2055116279485489376.txt", "title": "X @Align_ASI post 2055116279485489376" }, { "document_id": "source:akira.x.2055121067929489784", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2055121067929489784", "kind": "social_post", "language": "ja", "published_at": "2026-05-15", "title": "X @Align_ASI post 2055121067929489784", "url": "https://x.com/Align_ASI/status/2055121067929489784" }, "path": "sources/akira.x.2055121067929489784.txt", "title": "X @Align_ASI post 2055121067929489784" }, { "document_id": "source:akira.x.2055466196200505634", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2055466196200505634", "kind": "social_post", "language": "ja", "published_at": "2026-05-16", "title": "X @Align_ASI post 2055466196200505634", "url": "https://x.com/Align_ASI/status/2055466196200505634" }, "path": "sources/akira.x.2055466196200505634.txt", "title": "X @Align_ASI post 2055466196200505634" }, { "document_id": "source:akira.x.2055571488686674072", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2055571488686674072", "kind": "social_post", "language": "ja", "published_at": "2026-05-16", "title": "X @Align_ASI post 2055571488686674072", "url": "https://x.com/Align_ASI/status/2055571488686674072" }, "path": "sources/akira.x.2055571488686674072.txt", "title": "X @Align_ASI post 2055571488686674072" }, { "document_id": "source:akira.x.2056249992826982411", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2056249992826982411", "kind": "social_post", "language": "ja", "published_at": "2026-05-18", "title": "X @Align_ASI post 2056249992826982411", "url": "https://x.com/Align_ASI/status/2056249992826982411" }, "path": "sources/akira.x.2056249992826982411.txt", "title": "X @Align_ASI post 2056249992826982411" }, { "document_id": "source:akira.x.2056298561248039229", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2056298561248039229", "kind": "social_post", "language": "ja", "published_at": "2026-05-18", "title": "X @Align_ASI post 2056298561248039229", "url": "https://x.com/Align_ASI/status/2056298561248039229" }, "path": "sources/akira.x.2056298561248039229.txt", "title": "X @Align_ASI post 2056298561248039229" }, { "document_id": "source:akira.x.2056306172353991103", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2056306172353991103", "kind": "social_post", "language": "ja", "published_at": "2026-05-18", "title": "X @Align_ASI post 2056306172353991103", "url": "https://x.com/Align_ASI/status/2056306172353991103" }, "path": "sources/akira.x.2056306172353991103.txt", "title": "X @Align_ASI post 2056306172353991103" }, { "document_id": "source:akira.x.2056653057535205557", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2056653057535205557", "kind": "social_post", "language": "ja", "published_at": "2026-05-19", "title": "X @Align_ASI post 2056653057535205557", "url": "https://x.com/Align_ASI/status/2056653057535205557" }, "path": "sources/akira.x.2056653057535205557.txt", "title": "X @Align_ASI post 2056653057535205557" }, { "document_id": "source:akira.x.2056653932890583189", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2056653932890583189", "kind": "social_post", "language": "ja", "published_at": "2026-05-19", "title": "X @Align_ASI post 2056653932890583189", "url": "https://x.com/Align_ASI/status/2056653932890583189" }, "path": "sources/akira.x.2056653932890583189.txt", "title": "X @Align_ASI post 2056653932890583189" }, { "document_id": "source:akira.x.2056760361928519950", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2056760361928519950", "kind": "social_post", "language": "ja", "published_at": "2026-05-19", "title": "X @Align_ASI post 2056760361928519950", "url": "https://x.com/Align_ASI/status/2056760361928519950" }, "path": "sources/akira.x.2056760361928519950.txt", "title": "X @Align_ASI post 2056760361928519950" }, { "document_id": "source:akira.x.2087017636877996133", "layer": "source", "metadata": { "authors": [ "Akira (@Align_ASI)" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "akira.x.2087017636877996133", "kind": "social_post", "language": "ja", "published_at": "2026-08-11", "title": "X @Align_ASI post 2087017636877996133", "url": "https://x.com/Align_ASI/status/2087017636877996133" }, "path": "sources/akira.x.2087017636877996133.txt", "title": "X @Align_ASI post 2087017636877996133" }, { "document_id": "source:bensinger_et_al.the-problem", "layer": "source", "metadata": { "alternate_urls": [ "https://intelligence.org/the-problem/" ], "authors": [ "Rob Bensinger", "tanagrabeast", "yams", "So8res", "Eliezer Yudkowsky", "Gretta Duleba" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "bensinger_et_al.the-problem", "kind": "essay", "language": "en", "published_at": "2025-08-05", "title": "The Problem", "url": "https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem" }, "path": "sources/bensinger_et_al.the-problem.txt", "title": "The Problem" }, { "document_id": "source:betley_et_al.emergent-misalignment", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2502.17424", "https://arxiv.org/pdf/2502.17424" ], "authors": [ "Jan Betley", "Daniel Tan", "Niels Warncke", "Anna Sztyber-Betley", "Xuchan Bao", "Martín Soto", "Nathan Labenz", "Owain Evans" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "betley_et_al.emergent-misalignment", "kind": "paper", "language": "en", "published_at": "2025-02-24", "title": "Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs", "url": "https://arxiv.org/html/2502.17424" }, "path": "sources/betley_et_al.emergent-misalignment.txt", "title": "Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs" }, { "document_id": "source:bioshok.agi-industrial-tech-explosion", "layer": "source", "metadata": { "authors": [ "bioshok" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "bioshok.agi-industrial-tech-explosion", "kind": "article", "language": "ja", "published_at": "2026-03-24", "title": "AGIがもたらす産業・技術爆発", "url": "https://note.com/bioshok/n/n57f81d4fb9d7" }, "path": "sources/bioshok.agi-industrial-tech-explosion.txt", "title": "AGIがもたらす産業・技術爆発" }, { "document_id": "source:bioshok.ai-safety-vs-national-security", "layer": "source", "metadata": { "authors": [ "bioshok" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "bioshok.ai-safety-vs-national-security", "kind": "article", "language": "ja", "published_at": "2024-09-29", "title": "AI安全性と国家安全保障の対立", "url": "https://note.com/bioshok/n/n87c7434a623f" }, "path": "sources/bioshok.ai-safety-vs-national-security.txt", "title": "AI安全性と国家安全保障の対立" }, { "document_id": "source:bostrom.superintelligent-will", "layer": "source", "metadata": { "alternate_urls": [ "https://doi.org/10.1007/s11023-012-9281-3", "https://link.springer.com/article/10.1007/s11023-012-9281-3" ], "authors": [ "Nick Bostrom" ], "citation": { "locator_schemes": [ "page" ] }, "id": "bostrom.superintelligent-will", "kind": "paper", "language": "en", "published_at": "2012-06-13", "title": "The Superintelligent Will: Motivation and Instrumental Rationality in Advanced Artificial Agents", "url": "https://nickbostrom.com/superintelligentwill.pdf" }, "path": "sources/bostrom.superintelligent-will.txt", "title": "The Superintelligent Will: Motivation and Instrumental Rationality in Advanced Artificial Agents" }, { "document_id": "source:bowkis_et_al.automated-alignment-harder-than-you-think", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2605.06390", "https://arxiv.org/pdf/2605.06390" ], "authors": [ "Aleksandr Bowkis", "Marie Davidsen Buhl", "Jacob Pfau", "Geoffrey Irving" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "bowkis_et_al.automated-alignment-harder-than-you-think", "kind": "paper", "language": "en", "published_at": "2026-05-07", "title": "Automated alignment is harder than you think", "url": "https://arxiv.org/html/2605.06390" }, "path": "sources/bowkis_et_al.automated-alignment-harder-than-you-think.txt", "title": "Automated alignment is harder than you think" }, { "document_id": "source:carlsmith.scheming-ais", "layer": "source", "metadata": { "alternate_urls": [ "https://doi.org/10.48550/arXiv.2311.08379", "https://joecarlsmith.com/2023/11/15/new-report-scheming-ais-will-ais-fake-alignment-during-training-in-order-to-get-power/" ], "authors": [ "Joe Carlsmith" ], "citation": { "locator_schemes": [ "page" ] }, "id": "carlsmith.scheming-ais", "kind": "paper", "language": "en", "published_at": "2023-11-14", "title": "Scheming AIs: Will AIs fake alignment during training in order to get power?", "url": "https://arxiv.org/abs/2311.08379" }, "path": "sources/carlsmith.scheming-ais.txt", "title": "Scheming AIs: Will AIs fake alignment during training in order to get power?" }, { "document_id": "source:chanin_et_al.a-is-for-absorption", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2409.14507", "https://proceedings.neurips.cc/paper_files/paper/2025/file/764ff7477b8e24dbe01531f6791e8bdf-Paper-Conference.pdf" ], "authors": [ "David Chanin", "James Wilken-Smith", "Tomáš Dulka", "Hardik Bhatnagar", "Satvik Golechha", "Joseph Bloom" ], "citation": { "locator_schemes": [ "page" ] }, "id": "chanin_et_al.a-is-for-absorption", "kind": "paper", "language": "en", "published_at": "2024-09-22", "title": "A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders", "url": "https://proceedings.neurips.cc/paper_files/paper/2025/hash/764ff7477b8e24dbe01531f6791e8bdf-Abstract-Conference.html" }, "path": "sources/chanin_et_al.a-is-for-absorption.txt", "title": "A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders" }, { "document_id": "source:christiano.my-views-on-doom", "layer": "source", "metadata": { "authors": [ "Paul Christiano" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "christiano.my-views-on-doom", "kind": "essay", "language": "en", "published_at": "2023-04-27", "title": "My views on “doom”", "url": "https://www.lesswrong.com/posts/xWMqsvHapP3nwdSW8/my-views-on-doom" }, "path": "sources/christiano.my-views-on-doom.txt", "title": "My views on “doom”" }, { "document_id": "source:christiano.takeoff-speeds", "layer": "source", "metadata": { "alternate_urls": [ "https://www.alignmentforum.org/posts/AfGmsjGPXN97kNp57/arguments-about-fast-takeoff", "https://www.lesswrong.com/posts/AfGmsjGPXN97kNp57/arguments-about-fast-takeoff" ], "authors": [ "Paul Christiano" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "christiano.takeoff-speeds", "kind": "essay", "language": "en", "published_at": "2018-02-24", "title": "Takeoff speeds", "url": "https://sideways-view.com/2018/02/24/takeoff-speeds/" }, "path": "sources/christiano.takeoff-speeds.txt", "title": "Takeoff speeds" }, { "document_id": "source:christiano.what-failure-looks-like", "layer": "source", "metadata": { "alternate_urls": [ "https://www.lesswrong.com/posts/HBxe6wdjxK239zajf/more-realistic-tales-of-doom" ], "authors": [ "Paul Christiano" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "christiano.what-failure-looks-like", "kind": "essay", "language": "en", "published_at": "2019-03-17", "title": "What failure looks like", "url": "https://www.alignmentforum.org/posts/HBxe6wdjxK239zajf/what-failure-looks-like" }, "path": "sources/christiano.what-failure-looks-like.txt", "title": "What failure looks like" }, { "document_id": "source:christiano.where-i-agree-and-disagree-with-eliezer", "layer": "source", "metadata": { "alternate_urls": [ "https://www.lesswrong.com/posts/CoZhXrhpQxpy9xw9y/where-i-agree-and-disagree-with-eliezer" ], "authors": [ "Paul Christiano" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "christiano.where-i-agree-and-disagree-with-eliezer", "kind": "essay", "language": "en", "published_at": "2022-06-19", "title": "Where I agree and disagree with Eliezer", "url": "https://www.alignmentforum.org/posts/CoZhXrhpQxpy9xw9y/where-i-agree-and-disagree-with-eliezer" }, "path": "sources/christiano.where-i-agree-and-disagree-with-eliezer.txt", "title": "Where I agree and disagree with Eliezer" }, { "document_id": "source:christiano_xu.arcs-first-technical-report-eliciting-latent-knowledge", "layer": "source", "metadata": { "alternate_urls": [ "https://www.alignmentforum.org/posts/qHCDysDnvhteW7kRd/arc-s-first-technical-report-eliciting-latent-knowledge", "https://www.lesswrong.com/posts/qHCDysDnvhteW7kRd/arc-s-first-technical-report-eliciting-latent-knowledge/" ], "authors": [ "Paul Christiano", "Mark Xu" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "christiano_xu.arcs-first-technical-report-eliciting-latent-knowledge", "kind": "article", "language": "en", "published_at": "2021-12-14", "title": "ARC's first technical report: Eliciting Latent Knowledge", "url": "https://www.alignment.org/blog/arcs-first-technical-report-eliciting-latent-knowledge/" }, "path": "sources/christiano_xu.arcs-first-technical-report-eliciting-latent-knowledge.txt", "title": "ARC's first technical report: Eliciting Latent Knowledge" }, { "document_id": "source:cihon_et_al.should-ai-governance-be-centralised", "layer": "source", "metadata": { "alternate_urls": [ "https://doi.org/10.1145/3375627.3375857", "https://doi.org/10.48550/arXiv.2001.03573", "https://arxiv.org/html/2001.03573", "https://arxiv.org/pdf/2001.03573" ], "authors": [ "Peter Cihon", "Matthijs M. Maas", "Luke Kemp" ], "citation": { "locator_schemes": [ "page" ] }, "id": "cihon_et_al.should-ai-governance-be-centralised", "kind": "paper", "language": "en", "published_at": "2020-01-10", "title": "Should Artificial Intelligence Governance be Centralised? Design Lessons from History", "url": "https://arxiv.org/abs/2001.03573" }, "path": "sources/cihon_et_al.should-ai-governance-be-centralised.txt", "title": "Should Artificial Intelligence Governance be Centralised? Design Lessons from History" }, { "document_id": "source:cotra_wiblin.80000-hours-crunch-time", "layer": "source", "metadata": { "authors": [ "Ajeya Cotra", "Robert Wiblin" ], "citation": { "locator_schemes": [ "time" ] }, "id": "cotra_wiblin.80000-hours-crunch-time", "kind": "transcript", "language": "en", "published_at": "2026-02-17", "title": "#235 – Ajeya Cotra on whether it’s crazy that every AI company’s safety plan is ‘use AI to make AI safe’", "url": "https://80000hours.org/podcast/episodes/ajeya-cotra-transformative-ai-crunch-time/" }, "path": "sources/cotra_wiblin.80000-hours-crunch-time.txt", "title": "#235 – Ajeya Cotra on whether it’s crazy that every AI company’s safety plan is ‘use AI to make AI safe’" }, { "document_id": "source:davidson.compute-centric-framework-ai-takeoff", "layer": "source", "metadata": { "authors": [ "Tom Davidson" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "davidson.compute-centric-framework-ai-takeoff", "kind": "essay", "language": "en", "published_at": "2023-01-23", "title": "What a compute-centric framework says about AI takeoff speeds", "url": "https://www.lesswrong.com/posts/Gc9FGtdXhK9sCSEYu/what-a-compute-centric-framework-says-about-ai-takeoff" }, "path": "sources/davidson.compute-centric-framework-ai-takeoff.txt", "title": "What a compute-centric framework says about AI takeoff speeds" }, { "document_id": "source:engels_et_al.scaling-laws-for-scalable-oversight", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2504.18530", "https://proceedings.neurips.cc/paper_files/paper/2025/file/8c3caae2f725c8e2a55ecd600563d172-Paper-Conference.pdf" ], "authors": [ "Joshua Engels", "David D. Baek", "Subhash Kantamneni", "Max Tegmark" ], "citation": { "locator_schemes": [ "page" ] }, "id": "engels_et_al.scaling-laws-for-scalable-oversight", "kind": "paper", "language": "en", "published_at": "2025-04-25", "title": "Scaling Laws For Scalable Oversight", "url": "https://proceedings.neurips.cc/paper_files/paper/2025/hash/8c3caae2f725c8e2a55ecd600563d172-Abstract-Conference.html" }, "path": "sources/engels_et_al.scaling-laws-for-scalable-oversight.txt", "title": "Scaling Laws For Scalable Oversight" }, { "document_id": "source:erdil_barnett.most-ai-value-broad-automation", "layer": "source", "metadata": { "authors": [ "Ege Erdil", "Matthew Barnett" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "erdil_barnett.most-ai-value-broad-automation", "kind": "article", "language": "en", "published_at": "2025-03-21", "title": "Most AI value will come from broad automation, not from R&D", "url": "https://epoch.ai/gradient-updates/most-ai-value-will-come-from-broad-automation-not-from-r-d" }, "path": "sources/erdil_barnett.most-ai-value-broad-automation.txt", "title": "Most AI value will come from broad automation, not from R&D" }, { "document_id": "source:eth_davidson.ai-r-and-d-automation-software-intelligence-explosion", "layer": "source", "metadata": { "authors": [ "Daniel Eth", "Tom Davidson" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "eth_davidson.ai-r-and-d-automation-software-intelligence-explosion", "kind": "paper", "language": "en", "published_at": "2025-03-26", "title": "Will AI R&D Automation Cause a Software Intelligence Explosion?", "url": "https://www.forethought.org/research/will-ai-r-and-d-automation-cause-a-software-intelligence-explosion" }, "path": "sources/eth_davidson.ai-r-and-d-automation-software-intelligence-explosion.txt", "title": "Will AI R&D Automation Cause a Software Intelligence Explosion?" }, { "document_id": "source:greenblatt.current-ais-seem-pretty-misaligned", "layer": "source", "metadata": { "alternate_urls": [ "https://www.alignmentforum.org/posts/WewsByywWNhX9rtwi/current-ais-seem-pretty-misaligned-to-me", "https://www.lesswrong.com/posts/WewsByywWNhX9rtwi/current-ais-seem-pretty-misaligned-to-me" ], "authors": [ "Ryan Greenblatt" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "greenblatt.current-ais-seem-pretty-misaligned", "kind": "essay", "language": "en", "published_at": "2026-04-15", "title": "Current AIs seem pretty misaligned to me", "url": "https://blog.redwoodresearch.org/p/current-ais-seem-pretty-misaligned" }, "path": "sources/greenblatt.current-ais-seem-pretty-misaligned.txt", "title": "Current AIs seem pretty misaligned to me" }, { "document_id": "source:greenblatt.jankily-controlling-superintelligence", "layer": "source", "metadata": { "alternate_urls": [ "https://www.alignmentforum.org/posts/ainn5APCKHTFxuHKv/jankily-controlling-superintelligence", "https://www.lesswrong.com/posts/ainn5APCKHTFxuHKv/jankily-controlling-superintelligence" ], "authors": [ "Ryan Greenblatt" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "greenblatt.jankily-controlling-superintelligence", "kind": "essay", "language": "en", "published_at": "2025-06-27", "title": "Jankily controlling superintelligence", "url": "https://blog.redwoodresearch.org/p/jankily-controlling-superintelligence" }, "path": "sources/greenblatt.jankily-controlling-superintelligence.txt", "title": "Jankily controlling superintelligence" }, { "document_id": "source:greenblatt_et_al.alignment-faking-large-language-models", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2412.14093", "https://arxiv.org/pdf/2412.14093" ], "authors": [ "Ryan Greenblatt", "Carson Denison", "Benjamin Wright", "Fabien Roger", "Monte MacDiarmid", "Sam Marks", "Johannes Treutlein", "Tim Belonax", "Jack Chen", "David Duvenaud", "Akbir Khan", "Julian Michael", "Sören Mindermann", "Ethan Perez", "Linda Petrini", "Jonathan Uesato", "Jared Kaplan", "Buck Shlegeris", "Samuel R. Bowman", "Evan Hubinger" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "greenblatt_et_al.alignment-faking-large-language-models", "kind": "paper", "language": "en", "published_at": "2024-12-18", "title": "Alignment faking in large language models", "url": "https://arxiv.org/html/2412.14093" }, "path": "sources/greenblatt_et_al.alignment-faking-large-language-models.txt", "title": "Alignment faking in large language models" }, { "document_id": "source:hubinger_et_al.risks-from-learned-optimization", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/pdf/1906.01820" ], "authors": [ "Evan Hubinger", "Chris van Merwijk", "Vladimir Mikulik", "Joar Skalse", "Scott Garrabrant" ], "citation": { "locator_schemes": [ "page" ] }, "id": "hubinger_et_al.risks-from-learned-optimization", "kind": "paper", "language": "en", "published_at": "2019-06-05", "title": "Risks from Learned Optimization in Advanced Machine Learning Systems", "url": "https://arxiv.org/abs/1906.01820" }, "path": "sources/hubinger_et_al.risks-from-learned-optimization.txt", "title": "Risks from Learned Optimization in Advanced Machine Learning Systems" }, { "document_id": "source:hubinger_et_al.sleeper-agents", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2401.05566", "https://arxiv.org/pdf/2401.05566" ], "authors": [ "Evan Hubinger", "Carson Denison", "Jesse Mu", "Mike Lambert", "Meg Tong", "Monte MacDiarmid", "Tamera Lanham", "Daniel M. Ziegler", "Tim Maxwell", "Newton Cheng", "Adam Jermyn", "Amanda Askell", "Ansh Radhakrishnan", "Cem Anil", "David Duvenaud", "Deep Ganguli", "Fazl Barez", "Jack Clark", "Kamal Ndousse", "Kshitij Sachan", "Michael Sellitto", "Mrinank Sharma", "Nova DasSarma", "Roger Grosse", "Shauna Kravec", "Yuntao Bai", "Zachary Witten", "Marina Favaro", "Jan Brauner", "Holden Karnofsky", "Paul Christiano", "Samuel R. Bowman", "Logan Graham", "Jared Kaplan", "Sören Mindermann", "Ryan Greenblatt", "Buck Shlegeris", "Nicholas Schiefer", "Ethan Perez" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "hubinger_et_al.sleeper-agents", "kind": "paper", "language": "en", "published_at": "2024-01-10", "title": "Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training", "url": "https://arxiv.org/html/2401.05566" }, "path": "sources/hubinger_et_al.sleeper-agents.txt", "title": "Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training" }, { "document_id": "source:klein_yudkowsky.how-afraid-of-ai-apocalypse", "layer": "source", "metadata": { "authors": [ "Ezra Klein", "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "klein_yudkowsky.how-afraid-of-ai-apocalypse", "kind": "transcript", "language": "en", "published_at": "2025-10-15", "title": "How Afraid of the A.I. Apocalypse Should We Be?", "url": "https://www.nytimes.com/2025/10/15/opinion/ezra-klein-podcast-eliezer-yudkowsky.html" }, "path": "sources/klein_yudkowsky.how-afraid-of-ai-apocalypse.txt", "title": "How Afraid of the A.I. Apocalypse Should We Be?" }, { "document_id": "source:korbak_et_al.ai-control-safety-case", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/html/2501.17315v1" ], "authors": [ "Tomek Korbak", "Joshua Clymer", "Benjamin Hilton", "Buck Shlegeris", "Geoffrey Irving" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "korbak_et_al.ai-control-safety-case", "kind": "paper", "language": "en", "published_at": "2025-01-28", "title": "A sketch of an AI control safety case", "url": "https://arxiv.org/abs/2501.17315" }, "path": "sources/korbak_et_al.ai-control-safety-case.txt", "title": "A sketch of an AI control safety case" }, { "document_id": "source:leask_et_al.sparse-autoencoders-do-not-find-canonical-units-of-analysis", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2502.04878", "https://proceedings.iclr.cc/paper_files/paper/2025/file/84ca3f2d9d9bfca13f69b48ea63eb4a5-Paper-Conference.pdf" ], "authors": [ "Patrick Leask", "Bart Bussmann", "Michael Pearce", "Joseph Bloom", "Curt Tigges", "Noura Al Moubayed", "Lee Sharkey", "Neel Nanda" ], "citation": { "locator_schemes": [ "page" ] }, "id": "leask_et_al.sparse-autoencoders-do-not-find-canonical-units-of-analysis", "kind": "paper", "language": "en", "published_at": "2025-02-07", "title": "Sparse Autoencoders Do Not Find Canonical Units of Analysis", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/84ca3f2d9d9bfca13f69b48ea63eb4a5-Abstract-Conference.html" }, "path": "sources/leask_et_al.sparse-autoencoders-do-not-find-canonical-units-of-analysis.txt", "title": "Sparse Autoencoders Do Not Find Canonical Units of Analysis" }, { "document_id": "source:macdiarmid_et_al.natural-emergent-misalignment-reward-hacking", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2511.18397", "https://arxiv.org/pdf/2511.18397" ], "authors": [ "Monte MacDiarmid", "Benjamin Wright", "Jonathan Uesato", "Joe Benton", "Jon Kutasov", "Sara Price", "Naia Bouscal", "Sam Bowman", "Trenton Bricken", "Alex Cloud", "Carson Denison", "Johannes Gasteiger", "Ryan Greenblatt", "Jan Leike", "Jack Lindsey", "Vlad Mikulik", "Ethan Perez", "Alex Rodrigues", "Drake Thomas", "Albert Webson", "Daniel Ziegler", "Evan Hubinger" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "macdiarmid_et_al.natural-emergent-misalignment-reward-hacking", "kind": "paper", "language": "en", "published_at": "2025-11-23", "title": "Natural Emergent Misalignment from Reward Hacking in Production RL", "url": "https://arxiv.org/html/2511.18397" }, "path": "sources/macdiarmid_et_al.natural-emergent-misalignment-reward-hacking.txt", "title": "Natural Emergent Misalignment from Reward Hacking in Production RL" }, { "document_id": "source:ngo_et_al.alignment-problem-deep-learning", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2209.00626", "https://arxiv.org/pdf/2209.00626" ], "authors": [ "Richard Ngo", "Lawrence Chan", "Sören Mindermann" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "ngo_et_al.alignment-problem-deep-learning", "kind": "paper", "language": "en", "published_at": "2022-08-30", "title": "The Alignment Problem from a Deep Learning Perspective", "url": "https://arxiv.org/html/2209.00626v8" }, "path": "sources/ngo_et_al.alignment-problem-deep-learning.txt", "title": "The Alignment Problem from a Deep Learning Perspective" }, { "document_id": "source:openai.weak-to-strong-generalization", "layer": "source", "metadata": { "alternate_urls": [ "https://cdn.openai.com/papers/weak-to-strong-generalization.pdf" ], "authors": [ "Collin Burns", "Jan Leike", "Leopold Aschenbrenner", "Jeffrey Wu", "Pavel Izmailov", "Leo Gao", "Bowen Baker", "Jan Hendrik Kirchner" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "openai.weak-to-strong-generalization", "kind": "article", "language": "en", "published_at": "2023-12-14", "title": "Weak-to-strong generalization", "url": "https://openai.com/index/weak-to-strong-generalization/" }, "path": "sources/openai.weak-to-strong-generalization.txt", "title": "Weak-to-strong generalization" }, { "document_id": "source:raemon_yudkowsky_soares.corrigibility-hard-important", "layer": "source", "metadata": { "authors": [ "Raymond Arnold", "Eliezer Yudkowsky", "Nate Soares" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "raemon_yudkowsky_soares.corrigibility-hard-important", "kind": "essay", "language": "en", "published_at": "2025-09-30", "title": "Why Corrigibility is Hard and Important (i.e. \"Whence the high MIRI confidence in alignment difficulty?\")", "url": "https://www.lesswrong.com/posts/ksfjZJu3BFEfM6hHE/why-corrigibility-is-hard-and-important-i-e-whence-the-high" }, "path": "sources/raemon_yudkowsky_soares.corrigibility-hard-important.txt", "title": "Why Corrigibility is Hard and Important (i.e. \"Whence the high MIRI confidence in alignment difficulty?\")" }, { "document_id": "source:sastry_et_al.computing-power-and-governance-of-ai", "layer": "source", "metadata": { "alternate_urls": [ "https://doi.org/10.48550/arXiv.2402.08797", "https://arxiv.org/html/2402.08797", "https://arxiv.org/pdf/2402.08797" ], "authors": [ "Girish Sastry", "Lennart Heim", "Haydn Belfield", "Markus Anderljung", "Miles Brundage", "Julian Hazell", "Cullen O'Keefe", "Gillian K. Hadfield", "Richard Ngo", "Konstantin Pilz", "George Gor", "Emma Bluemke", "Sarah Shoker", "Janet Egan", "Robert F. Trager", "Shahar Avin", "Adrian Weller", "Yoshua Bengio", "Diane Coyle" ], "citation": { "locator_schemes": [ "page" ] }, "id": "sastry_et_al.computing-power-and-governance-of-ai", "kind": "paper", "language": "en", "published_at": "2024-02-13", "title": "Computing Power and the Governance of Artificial Intelligence", "url": "https://arxiv.org/abs/2402.08797" }, "path": "sources/sastry_et_al.computing-power-and-governance-of-ai.txt", "title": "Computing Power and the Governance of Artificial Intelligence" }, { "document_id": "source:scher_et_al.international-agreement-prevent-premature-asi", "layer": "source", "metadata": { "alternate_urls": [ "https://doi.org/10.48550/arXiv.2511.10783", "https://arxiv.org/html/2511.10783v3", "https://arxiv.org/pdf/2511.10783" ], "authors": [ "Aaron Scher", "David Abecassis", "Peter Barnett", "Brian Abeyta" ], "citation": { "locator_schemes": [ "page" ] }, "id": "scher_et_al.international-agreement-prevent-premature-asi", "kind": "paper", "language": "en", "published_at": "2025-11-13", "title": "An International Agreement to Prevent the Premature Creation of Artificial Superintelligence", "url": "https://arxiv.org/abs/2511.10783" }, "path": "sources/scher_et_al.international-agreement-prevent-premature-asi.txt", "title": "An International Agreement to Prevent the Premature Creation of Artificial Superintelligence" }, { "document_id": "source:soares.sharp-left-turn", "layer": "source", "metadata": { "alternate_urls": [ "https://www.lesswrong.com/posts/GNhMPAWcfBCASy8e6/a-central-ai-alignment-problem-capabilities-generalization" ], "authors": [ "Nate Soares" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "soares.sharp-left-turn", "kind": "essay", "language": "en", "published_at": "2022-06-15", "title": "A central AI alignment problem: capabilities generalization, and the sharp left turn", "url": "https://www.alignmentforum.org/posts/GNhMPAWcfBCASy8e6/a-central-ai-alignment-problem-capabilities-generalization" }, "path": "sources/soares.sharp-left-turn.txt", "title": "A central AI alignment problem: capabilities generalization, and the sharp left turn" }, { "document_id": "source:turner.optimal-policies-tend-to-seek-power", "layer": "source", "metadata": { "alternate_urls": [ "https://doi.org/10.48550/arXiv.1912.01683", "https://proceedings.neurips.cc/paper/2021/hash/c26820b8a4c1b3c2aa868d6d57e14a79-Abstract.html" ], "authors": [ "Alexander Matt Turner", "Andrew Critch", "Logan Smith", "Prasad Tadepalli", "Rohin Shah" ], "citation": { "locator_schemes": [ "page" ] }, "id": "turner.optimal-policies-tend-to-seek-power", "kind": "paper", "language": "en", "published_at": "2019-12-03", "title": "Optimal Policies Tend to Seek Power", "url": "https://arxiv.org/abs/1912.01683" }, "path": "sources/turner.optimal-policies-tend-to-seek-power.txt", "title": "Optimal Policies Tend to Seek Power" }, { "document_id": "source:yang_et_al.superficial-alignment", "layer": "source", "metadata": { "alternate_urls": [ "https://arxiv.org/abs/2406.11431", "https://proceedings.iclr.cc/paper_files/paper/2025/file/092359ce5cf60a80e882378944bf1be4-Paper-Conference.pdf" ], "authors": [ "Wenkai Yang", "Shiqi Shen", "Guangyao Shen", "Wei Yao", "Yong Liu", "Zhi Gong", "Yankai Lin", "Ji-Rong Wen" ], "citation": { "locator_schemes": [ "page" ] }, "id": "yang_et_al.superficial-alignment", "kind": "paper", "language": "en", "published_at": "2024-06-17", "title": "Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong Generalization", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/092359ce5cf60a80e882378944bf1be4-Abstract-Conference.html" }, "path": "sources/yang_et_al.superficial-alignment.txt", "title": "Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong Generalization" }, { "document_id": "source:yudkowsky.agi-ruin", "layer": "source", "metadata": { "alternate_urls": [ "https://www.lesswrong.com/posts/uMQ3cqWDPHhjtiesc/agi-ruin-a-list-of-lethalities" ], "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "anchor", "item" ] }, "id": "yudkowsky.agi-ruin", "kind": "essay", "language": "en", "published_at": "2022-06-10", "title": "AGI Ruin: A List of Lethalities", "url": "https://intelligence.org/2022/06/10/agi-ruin/" }, "path": "sources/yudkowsky.agi-ruin.txt", "title": "AGI Ruin: A List of Lethalities" }, { "document_id": "source:yudkowsky.ai-positive-negative-factor", "layer": "source", "metadata": { "alternate_urls": [ "https://doi.org/10.1093/oso/9780198570509.003.0021" ], "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "page" ] }, "id": "yudkowsky.ai-positive-negative-factor", "kind": "paper", "language": "en", "published_at": "2008-07-03", "title": "Artificial Intelligence as a Positive and Negative Factor in Global Risk", "url": "https://intelligence.org/files/AIPosNegFactor.pdf" }, "path": "sources/yudkowsky.ai-positive-negative-factor.txt", "title": "Artificial Intelligence as a Positive and Negative Factor in Global Risk" }, { "document_id": "source:yudkowsky.autonomous-agi", "layer": "source", "metadata": { "alternate_urls": [ "https://www.alignmentforum.org/w/autonomous-agi" ], "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "yudkowsky.autonomous-agi", "kind": "article", "language": "en", "published_at": "2015-12-28", "title": "Autonomous AGI", "url": "https://arbital.greaterwrong.com/p/Sovereign?l=1g3" }, "path": "sources/yudkowsky.autonomous-agi.txt", "title": "Autonomous AGI" }, { "document_id": "source:yudkowsky.coherent-extrapolated-volition", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "page" ] }, "id": "yudkowsky.coherent-extrapolated-volition", "kind": "paper", "language": "en", "title": "Coherent Extrapolated Volition", "url": "https://intelligence.org/files/CEV.pdf" }, "path": "sources/yudkowsky.coherent-extrapolated-volition.txt", "title": "Coherent Extrapolated Volition" }, { "document_id": "source:yudkowsky.empiricism-as-anti-epistemology", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "yudkowsky.empiricism-as-anti-epistemology", "kind": "essay", "language": "en", "published_at": "2024-03-14", "title": "'Empiricism!' as Anti-Epistemology", "url": "https://www.lesswrong.com/posts/LvKDMWQ3yLG9R3gHw/empiricism-as-anti-epistemology" }, "path": "sources/yudkowsky.empiricism-as-anti-epistemology.txt", "title": "'Empiricism!' as Anti-Epistemology" }, { "document_id": "source:yudkowsky.irretrievability", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "yudkowsky.irretrievability", "kind": "essay", "language": "en", "published_at": "2026-05-04", "title": "Irretrievability; or, Murphy's Curse of Oneshotness upon ASI", "url": "https://www.lesswrong.com/posts/fbrz9xhKpEeTKw5zL/irretrievability-or-murphy-s-curse-of-oneshotness-upon-asi" }, "path": "sources/yudkowsky.irretrievability.txt", "title": "Irretrievability; or, Murphy's Curse of Oneshotness upon ASI" }, { "document_id": "source:yudkowsky.no-fire-alarm-for-agi", "layer": "source", "metadata": { "alternate_urls": [ "https://www.lesswrong.com/posts/BEtzRE2M5m9YEAQpX/there-s-no-fire-alarm-for-artificial-general-intelligence" ], "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "yudkowsky.no-fire-alarm-for-agi", "kind": "essay", "language": "en", "published_at": "2017-10-13", "title": "There's No Fire Alarm for Artificial General Intelligence", "url": "https://intelligence.org/2017/10/13/fire-alarm/" }, "path": "sources/yudkowsky.no-fire-alarm-for-agi.txt", "title": "There's No Fire Alarm for Artificial General Intelligence" }, { "document_id": "source:yudkowsky.orthogonality-thesis", "layer": "source", "metadata": { "alternate_urls": [ "https://www.lesswrong.com/w/orthogonality-thesis" ], "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "yudkowsky.orthogonality-thesis", "kind": "article", "language": "en", "published_at": "2015-03-10", "title": "Orthogonality Thesis", "url": "https://arbital.greaterwrong.com/p/orthogonality/" }, "path": "sources/yudkowsky.orthogonality-thesis.txt", "title": "Orthogonality Thesis" }, { "document_id": "source:yudkowsky.pivotal-act", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "yudkowsky.pivotal-act", "kind": "article", "language": "en", "published_at": "2015-06-12", "title": "Pivotal act", "url": "https://arbital.greaterwrong.com/p/pivotal" }, "path": "sources/yudkowsky.pivotal-act.txt", "title": "Pivotal act" }, { "document_id": "source:yudkowsky.security-mindset-logistic-success-curve", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "yudkowsky.security-mindset-logistic-success-curve", "kind": "essay", "language": "en", "published_at": "2017-11-26", "title": "Security Mindset and the Logistic Success Curve", "url": "https://www.lesswrong.com/posts/cpdsMuAHSWhWnKdog/security-mindset-and-the-logistic-success-curve" }, "path": "sources/yudkowsky.security-mindset-logistic-success-curve.txt", "title": "Security Mindset and the Logistic Success Curve" }, { "document_id": "source:yudkowsky.security-mindset-ordinary-paranoia", "layer": "source", "metadata": { "alternate_urls": [ "https://intelligence.org/2017/11/25/security-mindset-ordinary-paranoia/" ], "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "yudkowsky.security-mindset-ordinary-paranoia", "kind": "essay", "language": "en", "published_at": "2017-11-25", "title": "Security Mindset and Ordinary Paranoia", "url": "https://www.lesswrong.com/posts/8gqrbnW758qjHFTrH/security-mindset-and-ordinary-paranoia" }, "path": "sources/yudkowsky.security-mindset-ordinary-paranoia.txt", "title": "Security Mindset and Ordinary Paranoia" }, { "document_id": "source:yudkowsky.shah-alignment-failures", "layer": "source", "metadata": { "alternate_urls": [ "https://www.lesswrong.com/posts/tcCxPLBrEXdxN5HCQ/shah-and-yudkowsky-on-alignment-failures" ], "authors": [ "Eliezer Yudkowsky", "Rohin Shah" ], "citation": { "locator_schemes": [ "item" ] }, "id": "yudkowsky.shah-alignment-failures", "kind": "transcript", "language": "en", "published_at": "2022-03-02", "title": "Shah and Yudkowsky on alignment failures", "url": "https://intelligence.org/2022/03/02/shah-and-yudkowsky-on-alignment-failures/" }, "path": "sources/yudkowsky.shah-alignment-failures.txt", "title": "Shah and Yudkowsky on alignment failures" }, { "document_id": "source:yudkowsky.shut-it-all-down", "layer": "source", "metadata": { "alternate_urls": [ "https://time.com/6266923/ai-eliezer-yudkowsky-open-letter-not-enough/", "https://www.lesswrong.com/posts/oM9pEezyCb4dCsuKq/pausing-ai-developments-isn-t-enough-we-need-to-shut-it-all-1" ], "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "yudkowsky.shut-it-all-down", "kind": "essay", "language": "en", "published_at": "2023-04-07", "title": "Pausing AI Developments Isn't Enough. We Need to Shut it All Down", "url": "https://intelligence.org/2023/04/07/pausing-ai-developments-isnt-enough-we-need-to-shut-it-all-down/" }, "path": "sources/yudkowsky.shut-it-all-down.txt", "title": "Pausing AI Developments Isn't Enough. We Need to Shut it All Down" }, { "document_id": "source:yudkowsky.sun-is-big", "layer": "source", "metadata": { "alternate_urls": [ "https://intelligence.org/2024/09/23/the-sun-is-big-but-superintelligences-will-not-spare-earth-a-little-sunlight/" ], "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "yudkowsky.sun-is-big", "kind": "essay", "language": "en", "published_at": "2024-09-23", "title": "The Sun is big, but superintelligences will not spare Earth a little sunlight", "url": "https://www.lesswrong.com/posts/F8sfrbPjCQj4KwJqn/the-sun-is-big-but-superintelligences-will-not-spare-earth-a" }, "path": "sources/yudkowsky.sun-is-big.txt", "title": "The Sun is big, but superintelligences will not spare Earth a little sunlight" }, { "document_id": "source:yudkowsky.ted-superintelligent-ai", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [ "time" ] }, "id": "yudkowsky.ted-superintelligent-ai", "kind": "transcript", "language": "en", "title": "Will superintelligent AI end the world?", "url": "https://www.ted.com/talks/eliezer_yudkowsky_will_superintelligent_ai_end_the_world" }, "path": "sources/yudkowsky.ted-superintelligent-ai.txt", "title": "Will superintelligent AI end the world?" }, { "document_id": "source:yudkowsky.thou-art-godshatter", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "yudkowsky.thou-art-godshatter", "kind": "essay", "language": "en", "published_at": "2007-11-13", "title": "Thou Art Godshatter", "url": "https://www.lesswrong.com/lw/l3/thou_art_godshatter/" }, "path": "sources/yudkowsky.thou-art-godshatter.txt", "title": "Thou Art Godshatter" }, { "document_id": "source:yudkowsky.value-is-fragile", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky" ], "citation": { "locator_schemes": [], "whole_source": true }, "id": "yudkowsky.value-is-fragile", "kind": "essay", "language": "en", "published_at": "2009-01-29", "title": "Value is Fragile", "url": "https://www.lesswrong.com/posts/GNnHHmm8EzePmKzPk/value-is-fragile" }, "path": "sources/yudkowsky.value-is-fragile.txt", "title": "Value is Fragile" }, { "document_id": "source:yudkowsky_ngo.alignment-difficulty", "layer": "source", "metadata": { "alternate_urls": [ "https://intelligence.org/2021/11/15/ngo-and-yudkowsky-on-alignment-difficulty/" ], "authors": [ "Eliezer Yudkowsky", "Richard Ngo" ], "citation": { "locator_schemes": [ "item" ] }, "id": "yudkowsky_ngo.alignment-difficulty", "kind": "transcript", "language": "en", "published_at": "2021-11-15", "title": "Ngo and Yudkowsky on alignment difficulty", "url": "https://www.lesswrong.com/s/HH4yBYELhbdEiihQF/p/7im8at9PmhbT4JHsW" }, "path": "sources/yudkowsky_ngo.alignment-difficulty.txt", "title": "Ngo and Yudkowsky on alignment difficulty" }, { "document_id": "source:yudkowsky_soares.read-ai-thoughts", "layer": "source", "metadata": { "authors": [ "Eliezer Yudkowsky", "Nate Soares" ], "citation": { "locator_schemes": [ "anchor" ] }, "id": "yudkowsky_soares.read-ai-thoughts", "kind": "article", "language": "en", "title": "Why not just read the AI's thoughts?", "url": "https://ifanyonebuildsit.com/11/why-not-just-read-the-ais-thoughts" }, "path": "sources/yudkowsky_soares.read-ai-thoughts.txt", "title": "Why not just read the AI's thoughts?" }, { "document_id": "terminology:agent", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "agent", "preferred": { "en": "agent", "ja": "エージェント" }, "status": "active" }, "path": "terminology/agent.txt", "title": "agent" }, { "document_id": "terminology:ai_alignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "alignment" ], "ja": [ "アラインメント", "アライメント", "整合", "AIアライメント" ] }, "concept_id": "ai_alignment", "preferred": { "en": "AI alignment", "ja": "AIアラインメント" }, "relations": { "relations": { "broader": [ "ai_safety" ], "prerequisite": [ "alignment_target", "artificial_intelligence" ], "related": [ "artificial_general_intelligence" ] } }, "status": "active" }, "path": "terminology/ai_alignment.txt", "title": "AI alignment" }, { "document_id": "terminology:ai_control", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI control research" ], "ja": [ "AI制御" ] }, "concept_id": "ai_control", "preferred": { "en": "AI control", "ja": "AIコントロール" }, "relations": { "relations": { "broader": [ "ai_safety" ], "prerequisite": [ "misalignment" ] } }, "status": "active" }, "path": "terminology/ai_control.txt", "title": "AI control" }, { "document_id": "terminology:ai_development_pause", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI development moratorium", "AI moratorium", "AI pause", "pause on AI development" ], "ja": [ "AIモラトリアム" ] }, "concept_id": "ai_development_pause", "preferred": { "en": "AI development pause", "ja": "AI開発の一時停止" }, "relations": { "relations": { "broader": [ "ai_governance" ], "prerequisite": [ "artificial_intelligence" ] } }, "status": "active" }, "path": "terminology/ai_development_pause.txt", "title": "AI development pause" }, { "document_id": "terminology:ai_existential_risk", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI x-risk", "existential risk from AI" ] }, "concept_id": "ai_existential_risk", "preferred": { "en": "AI existential risk" }, "relations": { "relations": { "broader": [ "ai_risk", "existential_risk" ], "prerequisite": [ "ai_risk", "existential_risk" ], "related": [ "alignment_risk", "misaligned_power_seeking" ] } }, "status": "active" }, "path": "terminology/ai_existential_risk.txt", "title": "AI existential risk" }, { "document_id": "terminology:ai_governance", "layer": "terminology", "metadata": { "aliases": { "en": [ "artificial intelligence governance" ] }, "concept_id": "ai_governance", "preferred": { "en": "AI governance" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence" ], "related": [ "ai_safety" ] } }, "status": "active" }, "path": "terminology/ai_governance.txt", "title": "AI governance" }, { "document_id": "terminology:ai_r_and_d_automation", "layer": "terminology", "metadata": { "aliases": { "en": [ "automation of AI R&D" ], "ja": [ "AI 研究開発自動化" ] }, "concept_id": "ai_r_and_d_automation", "preferred": { "en": "AI R&D automation", "ja": "AI 研究開発の自動化" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence", "capability" ], "related": [ "intelligence_explosion", "recursive_self_improvement", "takeoff_speed" ] } }, "status": "active" }, "path": "terminology/ai_r_and_d_automation.txt", "title": "AI R&D automation" }, { "document_id": "terminology:ai_race_dynamics", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI arms race dynamics", "AI race pressure", "competitive pressure in AI development" ], "ja": [ "軍拡競争" ] }, "concept_id": "ai_race_dynamics", "preferred": { "en": "AI race dynamics", "ja": "競争圧力" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence" ] } }, "status": "active" }, "path": "terminology/ai_race_dynamics.txt", "title": "AI race dynamics" }, { "document_id": "terminology:ai_risk", "layer": "terminology", "metadata": { "aliases": { "en": [ "artificial intelligence risk" ] }, "concept_id": "ai_risk", "preferred": { "en": "AI risk", "ja": "AIリスク" }, "relations": { "relations": { "broader": [ "risk" ], "prerequisite": [ "artificial_intelligence", "risk" ] } }, "status": "active" }, "path": "terminology/ai_risk.txt", "title": "AI risk" }, { "document_id": "terminology:ai_safety", "layer": "terminology", "metadata": { "aliases": { "en": [ "artificial intelligence safety" ] }, "concept_id": "ai_safety", "preferred": { "en": "AI safety", "ja": "AI安全性" }, "relations": { "relations": { "prerequisite": [ "ai_risk" ] } }, "status": "active" }, "path": "terminology/ai_safety.txt", "title": "AI safety" }, { "document_id": "terminology:ai_safety_level", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI Safety Levels", "ASL", "ASLs" ] }, "concept_id": "ai_safety_level", "preferred": { "en": "AI Safety Level" }, "relations": { "relations": { "prerequisite": [ "responsible_scaling_policy" ] } }, "status": "active" }, "path": "terminology/ai_safety_level.txt", "title": "AI Safety Level" }, { "document_id": "terminology:alignment_audit", "layer": "terminology", "metadata": { "aliases": { "en": [ "alignment auditing", "alignment audits" ] }, "concept_id": "alignment_audit", "preferred": { "en": "alignment audit" }, "relations": { "relations": { "broader": [ "evaluation" ], "prerequisite": [ "ai_alignment" ], "related": [ "interpretability", "red_teaming" ] } }, "status": "active" }, "path": "terminology/alignment_audit.txt", "title": "alignment audit" }, { "document_id": "terminology:alignment_by_default", "layer": "terminology", "metadata": { "aliases": { "en": [ "alignment-by-default" ] }, "concept_id": "alignment_by_default", "preferred": { "en": "Alignment by Default", "ja": "アラインメント・バイ・デフォルト" }, "relations": { "relations": { "broader": [ "ai_alignment" ], "prerequisite": [ "ai_alignment" ] } }, "status": "active" }, "path": "terminology/alignment_by_default.txt", "title": "Alignment by Default" }, { "document_id": "terminology:alignment_failure", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI alignment failure" ] }, "concept_id": "alignment_failure", "preferred": { "en": "alignment failure", "ja": "アラインメント失敗" }, "relations": { "relations": { "broader": [ "misalignment" ], "prerequisite": [ "ai_alignment", "misalignment" ], "related": [ "alignment_risk" ] } }, "status": "active" }, "path": "terminology/alignment_failure.txt", "title": "alignment failure" }, { "document_id": "terminology:alignment_faking", "layer": "terminology", "metadata": { "aliases": { "en": [ "alignment-faking", "fake alignment", "alignment fake" ], "ja": [ "猫を被る" ] }, "concept_id": "alignment_faking", "preferred": { "en": "alignment faking", "ja": "アラインメント偽装" }, "relations": { "relations": { "broader": [ "deception" ], "prerequisite": [ "ai_alignment", "deception", "oversight", "situational_awareness" ], "related": [ "scheming", "training_gaming" ] } }, "status": "active" }, "path": "terminology/alignment_faking.txt", "title": "alignment faking" }, { "document_id": "terminology:alignment_problem", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI alignment problem", "the alignment problem" ], "ja": [ "アラインメント問題" ] }, "concept_id": "alignment_problem", "preferred": { "en": "alignment problem", "ja": "AIアラインメント問題" }, "relations": { "relations": { "prerequisite": [ "ai_alignment", "misalignment" ] } }, "status": "active" }, "path": "terminology/alignment_problem.txt", "title": "alignment problem" }, { "document_id": "terminology:alignment_risk", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI alignment risk", "misalignment risk" ] }, "concept_id": "alignment_risk", "preferred": { "en": "alignment risk" }, "relations": { "relations": { "broader": [ "ai_risk" ], "prerequisite": [ "ai_risk", "misalignment" ], "related": [ "misaligned_power_seeking" ] } }, "status": "active" }, "path": "terminology/alignment_risk.txt", "title": "alignment risk" }, { "document_id": "terminology:alignment_target", "layer": "terminology", "metadata": { "aliases": { "en": [ "target of alignment" ] }, "concept_id": "alignment_target", "preferred": { "en": "alignment target", "ja": "アラインメント対象" }, "relations": { "relations": { "related": [ "human_preferences", "human_values", "intended_goal", "utility_function" ] } }, "status": "active" }, "path": "terminology/alignment_target.txt", "title": "alignment target" }, { "document_id": "terminology:artificial_general_intelligence", "layer": "terminology", "metadata": { "aliases": { "en": [ "AGI", "general AI" ] }, "concept_id": "artificial_general_intelligence", "preferred": { "en": "artificial general intelligence", "ja": "汎用人工知能" }, "relations": { "relations": { "broader": [ "artificial_intelligence" ], "prerequisite": [ "artificial_intelligence", "capability" ] } }, "status": "active" }, "path": "terminology/artificial_general_intelligence.txt", "title": "artificial general intelligence" }, { "document_id": "terminology:artificial_intelligence", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI" ] }, "concept_id": "artificial_intelligence", "preferred": { "en": "artificial intelligence", "ja": "人工知能" }, "status": "active" }, "path": "terminology/artificial_intelligence.txt", "title": "artificial intelligence" }, { "document_id": "terminology:artificial_superintelligence", "layer": "terminology", "metadata": { "aliases": { "en": [ "ASI", "superintelligent AI" ] }, "concept_id": "artificial_superintelligence", "preferred": { "en": "artificial superintelligence", "ja": "人工超知能" }, "relations": { "relations": { "broader": [ "artificial_general_intelligence" ], "prerequisite": [ "artificial_general_intelligence", "capability" ] } }, "status": "active" }, "path": "terminology/artificial_superintelligence.txt", "title": "artificial superintelligence" }, { "document_id": "terminology:attack_policy", "layer": "terminology", "metadata": { "aliases": { "en": [ "attack policies", "red-team attack policy" ] }, "concept_id": "attack_policy", "preferred": { "en": "attack policy" }, "relations": { "relations": { "prerequisite": [ "control_evaluation" ], "related": [ "scheming" ] } }, "status": "active" }, "path": "terminology/attack_policy.txt", "title": "attack policy" }, { "document_id": "terminology:automated_alignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI aligning AI", "AI-assisted alignment", "automated AI alignment", "automated alignment research" ], "ja": [ "AI支援アラインメント" ] }, "concept_id": "automated_alignment", "preferred": { "en": "automated alignment", "ja": "アラインメント研究の自動化" }, "relations": { "relations": { "broader": [ "ai_alignment" ], "prerequisite": [ "ai_alignment", "artificial_intelligence" ], "related": [ "crunch_time", "weak_agi" ] } }, "status": "active" }, "path": "terminology/automated_alignment.txt", "title": "automated alignment" }, { "document_id": "terminology:base_objective", "layer": "terminology", "metadata": { "aliases": { "en": [ "base-objective" ], "ja": [ "ベース目的" ] }, "concept_id": "base_objective", "preferred": { "en": "base objective", "ja": "ベース目的関数" }, "relations": { "relations": { "broader": [ "objective" ], "prerequisite": [ "base_optimizer" ], "related": [ "reward", "training_objective" ] } }, "status": "active" }, "path": "terminology/base_objective.txt", "title": "base objective" }, { "document_id": "terminology:base_optimizer", "layer": "terminology", "metadata": { "aliases": { "en": [ "base-optimizer" ], "ja": [ "ベースオプティマイザー" ] }, "concept_id": "base_optimizer", "preferred": { "en": "base optimizer", "ja": "ベースオプティマイザ" }, "relations": { "relations": { "broader": [ "optimizer" ], "prerequisite": [ "objective", "optimizer" ], "related": [ "training" ] } }, "status": "active" }, "path": "terminology/base_optimizer.txt", "title": "base optimizer" }, { "document_id": "terminology:bayesian_inference", "layer": "terminology", "metadata": { "aliases": { "en": [ "Bayesian updating" ] }, "concept_id": "bayesian_inference", "preferred": { "en": "Bayesian inference", "ja": "ベイズ推定" }, "status": "active" }, "path": "terminology/bayesian_inference.txt", "title": "Bayesian inference" }, { "document_id": "terminology:behavioral_objective", "layer": "terminology", "metadata": { "aliases": { "en": [ "behavioural objective" ], "ja": [ "行動上の目的" ] }, "concept_id": "behavioral_objective", "preferred": { "en": "behavioral objective", "ja": "行動目的" }, "relations": { "relations": { "broader": [ "objective" ], "related": [ "mesa_objective" ] } }, "status": "active" }, "path": "terminology/behavioral_objective.txt", "title": "behavioral objective" }, { "document_id": "terminology:benchmark", "layer": "terminology", "metadata": { "aliases": { "en": [ "benchmark evaluation" ] }, "concept_id": "benchmark", "preferred": { "en": "benchmark", "ja": "ベンチマーク" }, "relations": { "relations": { "broader": [ "evaluation" ], "prerequisite": [ "evaluation" ] } }, "status": "active" }, "path": "terminology/benchmark.txt", "title": "benchmark" }, { "document_id": "terminology:capability", "layer": "terminology", "metadata": { "aliases": { "en": [ "capabilities" ], "ja": [ "AI能力" ] }, "concept_id": "capability", "preferred": { "en": "capability", "ja": "能力" }, "relations": { "relations": { "related": [ "evaluation" ] } }, "status": "active" }, "path": "terminology/capability.txt", "title": "capability" }, { "document_id": "terminology:capability_elicitation", "layer": "terminology", "metadata": { "aliases": { "en": [ "capabilities elicitation", "eliciting capabilities" ] }, "concept_id": "capability_elicitation", "preferred": { "en": "capability elicitation" }, "relations": { "relations": { "prerequisite": [ "capability" ], "related": [ "control_evaluation", "sandbagging", "weak_to_strong_generalization" ] } }, "status": "active" }, "path": "terminology/capability_elicitation.txt", "title": "capability elicitation" }, { "document_id": "terminology:capability_evaluation", "layer": "terminology", "metadata": { "aliases": { "en": [ "capability eval", "capability evaluations" ] }, "concept_id": "capability_evaluation", "preferred": { "en": "capability evaluation", "ja": "能力評価" }, "relations": { "relations": { "broader": [ "evaluation" ], "prerequisite": [ "capability", "capability_elicitation", "evaluation" ], "related": [ "control_evaluation" ] } }, "status": "active" }, "path": "terminology/capability_evaluation.txt", "title": "capability evaluation" }, { "document_id": "terminology:capability_generalization", "layer": "terminology", "metadata": { "aliases": { "en": [ "capabilities generalisation", "capabilities generalization", "capability generalisation" ] }, "concept_id": "capability_generalization", "preferred": { "en": "capability generalization", "ja": "能力の汎化" }, "relations": { "relations": { "broader": [ "generalization" ], "prerequisite": [ "capability", "generalization" ] } }, "status": "active" }, "path": "terminology/capability_generalization.txt", "title": "capability generalization" }, { "document_id": "terminology:capability_threshold", "layer": "terminology", "metadata": { "aliases": { "en": [ "capability thresholds" ] }, "concept_id": "capability_threshold", "preferred": { "en": "capability threshold" }, "relations": { "relations": { "prerequisite": [ "capability" ], "related": [ "dangerous_capability_evaluation" ] } }, "status": "active" }, "path": "terminology/capability_threshold.txt", "title": "capability threshold" }, { "document_id": "terminology:catastrophic_risk", "layer": "terminology", "metadata": { "aliases": { "en": [ "catastrophic risks" ], "ja": [ "破局的リスク", "破滅的リスク" ] }, "concept_id": "catastrophic_risk", "preferred": { "en": "catastrophic risk", "ja": "壊滅的リスク" }, "relations": { "relations": { "broader": [ "risk" ], "prerequisite": [ "risk" ] } }, "status": "active" }, "path": "terminology/catastrophic_risk.txt", "title": "catastrophic risk" }, { "document_id": "terminology:coherent_extrapolated_volition", "layer": "terminology", "metadata": { "aliases": { "en": [ "CEV" ] }, "concept_id": "coherent_extrapolated_volition", "preferred": { "en": "Coherent Extrapolated Volition", "ja": "首尾一貫した外挿的意志" }, "relations": { "relations": { "broader": [ "alignment_target" ], "prerequisite": [ "human_preferences", "human_values" ], "related": [ "friendly_ai" ] } }, "status": "active" }, "path": "terminology/coherent_extrapolated_volition.txt", "title": "Coherent Extrapolated Volition" }, { "document_id": "terminology:collusion", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI collusion", "actor-monitor collusion", "model collusion", "policy-monitor collusion" ] }, "concept_id": "collusion", "preferred": { "en": "collusion" }, "relations": { "relations": { "prerequisite": [ "control_protocol" ], "related": [ "untrusted_monitoring" ] } }, "status": "active" }, "path": "terminology/collusion.txt", "title": "collusion" }, { "document_id": "terminology:compute_governance", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI compute governance" ] }, "concept_id": "compute_governance", "preferred": { "en": "compute governance" }, "relations": { "relations": { "broader": [ "ai_governance" ] } }, "status": "active" }, "path": "terminology/compute_governance.txt", "title": "compute governance" }, { "document_id": "terminology:control_evaluation", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI control eval", "AI control evaluation", "control eval", "control evaluations" ] }, "concept_id": "control_evaluation", "preferred": { "en": "control evaluation" }, "relations": { "relations": { "broader": [ "evaluation" ], "prerequisite": [ "control_protocol", "evaluation" ], "related": [ "red_teaming" ] } }, "status": "active" }, "path": "terminology/control_evaluation.txt", "title": "control evaluation" }, { "document_id": "terminology:control_protocol", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI control protocol", "control protocols" ] }, "concept_id": "control_protocol", "preferred": { "en": "control protocol" }, "relations": { "relations": { "broader": [ "ai_control" ], "prerequisite": [ "ai_control" ] } }, "status": "active" }, "path": "terminology/control_protocol.txt", "title": "control protocol" }, { "document_id": "terminology:control_safety_case", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI control safety case", "control safety cases" ] }, "concept_id": "control_safety_case", "preferred": { "en": "control safety case" }, "relations": { "relations": { "broader": [ "safety_case" ], "prerequisite": [ "capability_elicitation", "control_evaluation", "control_protocol" ] } }, "status": "active" }, "path": "terminology/control_safety_case.txt", "title": "control safety case" }, { "document_id": "terminology:corrigibility", "layer": "terminology", "metadata": { "aliases": { "en": [ "corrigible" ], "ja": [ "修正可能性" ] }, "concept_id": "corrigibility", "preferred": { "en": "corrigibility", "ja": "訂正可能性" }, "relations": { "relations": { "broader": [ "ai_alignment" ], "prerequisite": [ "instrumental_convergence" ], "related": [ "oversight", "power_seeking" ] } }, "status": "active" }, "path": "terminology/corrigibility.txt", "title": "corrigibility" }, { "document_id": "terminology:crunch_time", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI crunch time" ], "ja": [ "AIクランチタイム" ] }, "concept_id": "crunch_time", "preferred": { "en": "crunch time", "ja": "クランチタイム" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence", "capability", "intelligence_explosion" ], "related": [ "recursive_self_improvement", "takeoff_speed" ] } }, "status": "active" }, "path": "terminology/crunch_time.txt", "title": "crunch time" }, { "document_id": "terminology:dangerous_capability_evaluation", "layer": "terminology", "metadata": { "aliases": { "en": [ "dangerous capabilities eval", "dangerous capabilities evals", "dangerous capabilities evaluation", "dangerous capabilities evaluations", "dangerous capability eval", "dangerous capability evals" ] }, "concept_id": "dangerous_capability_evaluation", "preferred": { "en": "dangerous capability evaluation" }, "relations": { "relations": { "broader": [ "capability_evaluation" ], "related": [ "red_teaming" ] } }, "status": "active" }, "path": "terminology/dangerous_capability_evaluation.txt", "title": "dangerous capability evaluation" }, { "document_id": "terminology:debate", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI safety via debate", "AI debate" ] }, "concept_id": "debate", "preferred": { "en": "debate", "ja": "ディベート" }, "relations": { "relations": { "broader": [ "scalable_oversight" ], "prerequisite": [ "evaluation" ] } }, "status": "active" }, "path": "terminology/debate.txt", "title": "debate" }, { "document_id": "terminology:deception", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI deception", "strategic deception" ], "ja": [ "デセプション" ] }, "concept_id": "deception", "preferred": { "en": "deception", "ja": "欺瞞" }, "status": "active" }, "path": "terminology/deception.txt", "title": "deception" }, { "document_id": "terminology:deceptive_alignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "deceptively aligned AI" ] }, "concept_id": "deceptive_alignment", "preferred": { "en": "deceptive alignment", "ja": "欺瞞的アラインメント" }, "relations": { "relations": { "broader": [ "alignment_faking", "inner_misalignment" ], "prerequisite": [ "deception", "goal_directed_behavior", "inner_misalignment", "oversight", "situational_awareness" ], "related": [ "misaligned_power_seeking", "scheming" ] } }, "status": "active" }, "path": "terminology/deceptive_alignment.txt", "title": "deceptive alignment" }, { "document_id": "terminology:distribution_shift", "layer": "terminology", "metadata": { "aliases": { "en": [ "distributional shift" ] }, "concept_id": "distribution_shift", "preferred": { "en": "distribution shift", "ja": "分布シフト" }, "relations": { "relations": { "prerequisite": [ "training" ], "related": [ "generalization", "objective_robustness" ] } }, "status": "active" }, "path": "terminology/distribution_shift.txt", "title": "distribution shift" }, { "document_id": "terminology:eliciting_latent_knowledge", "layer": "terminology", "metadata": { "aliases": { "en": [ "ELK" ] }, "concept_id": "eliciting_latent_knowledge", "preferred": { "en": "Eliciting Latent Knowledge" }, "relations": { "relations": { "prerequisite": [ "latent_knowledge", "world_model" ], "related": [ "interpretability", "ontology_identification_problem", "scalable_oversight", "weak_to_strong_generalization" ] } }, "status": "active" }, "path": "terminology/eliciting_latent_knowledge.txt", "title": "Eliciting Latent Knowledge" }, { "document_id": "terminology:evaluation", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI eval", "AI evals", "AI evaluation", "AI evaluations", "eval", "evals", "model eval", "model evals", "model evaluation", "model evaluations" ], "ja": [ "AI評価", "モデル評価" ] }, "concept_id": "evaluation", "preferred": { "en": "evaluation", "ja": "評価" }, "relations": { "relations": { "related": [ "generalization", "oversight", "safety_case" ] } }, "status": "active" }, "path": "terminology/evaluation.txt", "title": "evaluation" }, { "document_id": "terminology:evaluation_awareness", "layer": "terminology", "metadata": { "aliases": { "en": [ "eval awareness" ] }, "concept_id": "evaluation_awareness", "preferred": { "en": "evaluation awareness", "ja": "評価認識" }, "relations": { "relations": { "broader": [ "situational_awareness" ], "prerequisite": [ "evaluation", "situational_awareness" ], "related": [ "sandbagging", "training_gaming" ] } }, "status": "active" }, "path": "terminology/evaluation_awareness.txt", "title": "evaluation awareness" }, { "document_id": "terminology:existential_risk", "layer": "terminology", "metadata": { "aliases": { "en": [ "existential risks", "x-risk", "x risk" ], "ja": [ "実存的リスク", "存亡的リスク" ] }, "concept_id": "existential_risk", "preferred": { "en": "existential risk", "ja": "存亡リスク" }, "relations": { "relations": { "broader": [ "catastrophic_risk" ], "prerequisite": [ "catastrophic_risk" ] } }, "status": "active" }, "path": "terminology/existential_risk.txt", "title": "existential risk" }, { "document_id": "terminology:factored_cognition", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "factored_cognition", "preferred": { "en": "factored cognition" }, "relations": { "relations": { "related": [ "iterated_amplification", "process_supervision" ] } }, "status": "active" }, "path": "terminology/factored_cognition.txt", "title": "factored cognition" }, { "document_id": "terminology:first_critical_try", "layer": "terminology", "metadata": { "aliases": { "en": [ "first critical tries" ], "ja": [ "最初の危険な試行" ] }, "concept_id": "first_critical_try", "preferred": { "en": "first critical try", "ja": "最初の重要な試行" }, "relations": { "relations": { "prerequisite": [ "catastrophic_risk" ] } }, "status": "active" }, "path": "terminology/first_critical_try.txt", "title": "first critical try" }, { "document_id": "terminology:friendly_ai", "layer": "terminology", "metadata": { "aliases": { "en": [ "FAI", "Friendly Artificial Intelligence" ] }, "concept_id": "friendly_ai", "preferred": { "en": "Friendly AI", "ja": "フレンドリーAI" }, "relations": { "relations": { "prerequisite": [ "ai_alignment", "human_values" ] } }, "status": "active" }, "path": "terminology/friendly_ai.txt", "title": "Friendly AI" }, { "document_id": "terminology:frontier_model", "layer": "terminology", "metadata": { "aliases": { "en": [ "frontier AI model", "frontier AI models", "frontier models" ] }, "concept_id": "frontier_model", "preferred": { "en": "frontier model" }, "relations": { "relations": { "broader": [ "artificial_intelligence" ], "prerequisite": [ "capability" ], "related": [ "responsible_scaling_policy" ] } }, "status": "active" }, "path": "terminology/frontier_model.txt", "title": "frontier model" }, { "document_id": "terminology:generalization", "layer": "terminology", "metadata": { "aliases": { "en": [ "generalisation" ] }, "concept_id": "generalization", "preferred": { "en": "generalization", "ja": "汎化" }, "relations": { "relations": { "broader": [ "capability" ], "prerequisite": [ "training" ] } }, "status": "active" }, "path": "terminology/generalization.txt", "title": "generalization" }, { "document_id": "terminology:goal", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "goal", "preferred": { "en": "goal", "ja": "目標" }, "relations": { "relations": { "related": [ "values" ] } }, "status": "active" }, "path": "terminology/goal.txt", "title": "goal" }, { "document_id": "terminology:goal_directed_behavior", "layer": "terminology", "metadata": { "aliases": { "en": [ "goal-directed", "goal directed behavior" ] }, "concept_id": "goal_directed_behavior", "preferred": { "en": "goal-directed behavior", "ja": "目標指向的行動" }, "relations": { "relations": { "prerequisite": [ "agent", "goal" ] } }, "status": "active" }, "path": "terminology/goal_directed_behavior.txt", "title": "goal-directed behavior" }, { "document_id": "terminology:goal_guarding_scheming", "layer": "terminology", "metadata": { "aliases": { "en": [ "goal guarding scheming", "goal-guarding schemer" ] }, "concept_id": "goal_guarding_scheming", "preferred": { "en": "goal-guarding scheming", "ja": "目標保護型スキーミング" }, "relations": { "relations": { "broader": [ "deceptive_alignment", "scheming", "training_gaming" ], "prerequisite": [ "goal", "reward", "situational_awareness", "training" ] } }, "status": "active" }, "path": "terminology/goal_guarding_scheming.txt", "title": "goal-guarding scheming" }, { "document_id": "terminology:goal_misgeneralization", "layer": "terminology", "metadata": { "aliases": { "en": [ "goal mis-generalization" ] }, "concept_id": "goal_misgeneralization", "preferred": { "en": "goal misgeneralization", "ja": "目標の誤一般化" }, "relations": { "relations": { "broader": [ "inner_misalignment" ], "prerequisite": [ "capability", "distribution_shift", "generalization", "inner_misalignment" ], "related": [ "objective_robustness", "specification_gaming" ] } }, "status": "active" }, "path": "terminology/goal_misgeneralization.txt", "title": "goal misgeneralization" }, { "document_id": "terminology:goodharts_law", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "goodharts_law", "preferred": { "en": "Goodhart's Law", "ja": "グッドハートの法則" }, "relations": { "relations": { "prerequisite": [ "optimization", "proxy_objective" ], "related": [ "outer_misalignment", "reward_hacking", "specification_gaming" ] } }, "status": "active" }, "path": "terminology/goodharts_law.txt", "title": "Goodhart's Law" }, { "document_id": "terminology:human_preferences", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "human_preferences", "preferred": { "en": "human preferences", "ja": "人間の選好" }, "relations": { "relations": { "broader": [ "preferences" ], "related": [ "human_values" ] } }, "status": "active" }, "path": "terminology/human_preferences.txt", "title": "human preferences" }, { "document_id": "terminology:human_values", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "human_values", "preferred": { "en": "human values", "ja": "人間の価値観" }, "relations": { "relations": { "broader": [ "values" ] } }, "status": "active" }, "path": "terminology/human_values.txt", "title": "human values" }, { "document_id": "terminology:industrial_explosion", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI industrial explosion" ] }, "concept_id": "industrial_explosion", "preferred": { "en": "industrial explosion", "ja": "産業爆発" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence", "capability" ], "related": [ "intelligence_explosion", "technology_explosion" ] } }, "status": "active" }, "path": "terminology/industrial_explosion.txt", "title": "industrial explosion" }, { "document_id": "terminology:inner_alignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "inner alignment problem" ] }, "concept_id": "inner_alignment", "preferred": { "en": "inner alignment", "ja": "インナーアラインメント" }, "relations": { "relations": { "broader": [ "ai_alignment" ], "prerequisite": [ "ai_alignment", "base_objective", "mesa_objective", "mesa_optimizer" ], "related": [ "objective_robustness", "outer_alignment" ] } }, "status": "active" }, "path": "terminology/inner_alignment.txt", "title": "inner alignment" }, { "document_id": "terminology:inner_misalignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "inner alignment failure" ] }, "concept_id": "inner_misalignment", "preferred": { "en": "inner misalignment", "ja": "インナー・ミスアラインメント" }, "relations": { "relations": { "broader": [ "misalignment" ], "prerequisite": [ "inner_alignment", "misalignment" ], "related": [ "outer_misalignment" ] } }, "status": "active" }, "path": "terminology/inner_misalignment.txt", "title": "inner misalignment" }, { "document_id": "terminology:instrumental_convergence", "layer": "terminology", "metadata": { "aliases": { "en": [ "convergent instrumental goals" ] }, "concept_id": "instrumental_convergence", "preferred": { "en": "instrumental convergence", "ja": "道具的収束" }, "relations": { "relations": { "prerequisite": [ "goal_directed_behavior", "instrumental_goal", "terminal_goal" ], "related": [ "orthogonality_thesis", "paperclip_maximizer", "power_seeking", "resource_acquisition", "self_preservation" ] } }, "status": "active" }, "path": "terminology/instrumental_convergence.txt", "title": "instrumental convergence" }, { "document_id": "terminology:instrumental_goal", "layer": "terminology", "metadata": { "aliases": { "en": [ "instrumental subgoal" ], "ja": [ "手段的目標" ] }, "concept_id": "instrumental_goal", "preferred": { "en": "instrumental goal", "ja": "道具的目標" }, "relations": { "relations": { "broader": [ "goal" ], "prerequisite": [ "goal" ], "related": [ "power_seeking", "terminal_goal" ] } }, "status": "active" }, "path": "terminology/instrumental_goal.txt", "title": "instrumental goal" }, { "document_id": "terminology:intelligence_explosion", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI intelligence explosion" ] }, "concept_id": "intelligence_explosion", "preferred": { "en": "intelligence explosion", "ja": "知能爆発" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence", "capability" ], "related": [ "recursive_self_improvement", "technology_explosion" ] } }, "status": "active" }, "path": "terminology/intelligence_explosion.txt", "title": "intelligence explosion" }, { "document_id": "terminology:intended_goal", "layer": "terminology", "metadata": { "aliases": { "en": [ "designer's intended goal", "programmer's intended goal" ], "ja": [ "設計者の意図した目標" ] }, "concept_id": "intended_goal", "preferred": { "en": "intended goal", "ja": "意図された目標" }, "relations": { "relations": { "broader": [ "goal" ], "related": [ "outer_alignment", "training_objective" ] } }, "status": "active" }, "path": "terminology/intended_goal.txt", "title": "intended goal" }, { "document_id": "terminology:interpretability", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI interpretability", "ML interpretability" ] }, "concept_id": "interpretability", "preferred": { "en": "interpretability", "ja": "解釈可能性" }, "status": "active" }, "path": "terminology/interpretability.txt", "title": "interpretability" }, { "document_id": "terminology:iterated_amplification", "layer": "terminology", "metadata": { "aliases": { "en": [ "IDA", "iterated distillation and amplification" ] }, "concept_id": "iterated_amplification", "preferred": { "en": "iterated amplification" }, "relations": { "relations": { "broader": [ "scalable_oversight" ], "prerequisite": [ "oversight", "training" ] } }, "status": "active" }, "path": "terminology/iterated_amplification.txt", "title": "iterated amplification" }, { "document_id": "terminology:latent_knowledge", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "latent_knowledge", "preferred": { "en": "latent knowledge" }, "relations": { "relations": { "prerequisite": [ "machine_learning" ], "related": [ "world_model" ] } }, "status": "active" }, "path": "terminology/latent_knowledge.txt", "title": "latent knowledge" }, { "document_id": "terminology:loss_of_control", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI loss of control", "loss of human control" ] }, "concept_id": "loss_of_control", "preferred": { "en": "loss of control", "ja": "制御喪失" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence", "capability" ] } }, "status": "active" }, "path": "terminology/loss_of_control.txt", "title": "loss of control" }, { "document_id": "terminology:machine_learning", "layer": "terminology", "metadata": { "aliases": { "en": [ "ML" ] }, "concept_id": "machine_learning", "preferred": { "en": "machine learning", "ja": "機械学習" }, "relations": { "relations": { "broader": [ "artificial_intelligence" ] } }, "status": "active" }, "path": "terminology/machine_learning.txt", "title": "machine learning" }, { "document_id": "terminology:mechanistic_interpretability", "layer": "terminology", "metadata": { "aliases": { "en": [ "mech interp", "mechanistic interp", "mechinterp" ], "ja": [ "メカインタープ" ] }, "concept_id": "mechanistic_interpretability", "preferred": { "en": "mechanistic interpretability", "ja": "メカニスティック・インタープリタビリティ" }, "relations": { "relations": { "broader": [ "interpretability" ], "prerequisite": [ "interpretability", "neural_network" ] } }, "status": "active" }, "path": "terminology/mechanistic_interpretability.txt", "title": "mechanistic interpretability" }, { "document_id": "terminology:mesa_objective", "layer": "terminology", "metadata": { "aliases": { "en": [ "mesa objective" ], "ja": [ "メサ目的" ] }, "concept_id": "mesa_objective", "preferred": { "en": "mesa-objective", "ja": "メサ目的関数" }, "relations": { "relations": { "broader": [ "objective" ], "prerequisite": [ "mesa_optimizer" ] } }, "status": "active" }, "path": "terminology/mesa_objective.txt", "title": "mesa-objective" }, { "document_id": "terminology:mesa_optimization", "layer": "terminology", "metadata": { "aliases": { "en": [ "mesa optimization" ] }, "concept_id": "mesa_optimization", "preferred": { "en": "mesa-optimization", "ja": "メサ最適化" }, "relations": { "relations": { "broader": [ "optimization" ], "prerequisite": [ "base_optimizer", "mesa_optimizer", "optimization" ], "related": [ "training" ] } }, "status": "active" }, "path": "terminology/mesa_optimization.txt", "title": "mesa-optimization" }, { "document_id": "terminology:mesa_optimizer", "layer": "terminology", "metadata": { "aliases": { "en": [ "mesa optimizer" ] }, "concept_id": "mesa_optimizer", "preferred": { "en": "mesa-optimizer", "ja": "メサオプティマイザ" }, "relations": { "relations": { "broader": [ "optimizer" ], "prerequisite": [ "base_optimizer" ] } }, "status": "active" }, "path": "terminology/mesa_optimizer.txt", "title": "mesa-optimizer" }, { "document_id": "terminology:misaligned_agi", "layer": "terminology", "metadata": { "aliases": { "en": [ "misaligned artificial general intelligence" ] }, "concept_id": "misaligned_agi", "preferred": { "en": "misaligned AGI", "ja": "ミスアラインドAGI" }, "relations": { "relations": { "broader": [ "artificial_general_intelligence", "misaligned_ai" ], "prerequisite": [ "artificial_general_intelligence", "misaligned_ai" ] } }, "status": "active" }, "path": "terminology/misaligned_agi.txt", "title": "misaligned AGI" }, { "document_id": "terminology:misaligned_ai", "layer": "terminology", "metadata": { "aliases": { "en": [ "misaligned artificial intelligence" ], "ja": [ "ミスアラインド" ] }, "concept_id": "misaligned_ai", "preferred": { "en": "misaligned AI", "ja": "ミスアラインドAI" }, "relations": { "relations": { "broader": [ "artificial_intelligence" ], "prerequisite": [ "artificial_intelligence", "misalignment" ] } }, "status": "active" }, "path": "terminology/misaligned_ai.txt", "title": "misaligned AI" }, { "document_id": "terminology:misaligned_power_seeking", "layer": "terminology", "metadata": { "aliases": { "en": [ "MAPS", "misaligned power seeking" ] }, "concept_id": "misaligned_power_seeking", "preferred": { "en": "misaligned power-seeking" }, "relations": { "relations": { "broader": [ "power_seeking" ], "prerequisite": [ "misalignment", "power_seeking" ] } }, "status": "active" }, "path": "terminology/misaligned_power_seeking.txt", "title": "misaligned power-seeking" }, { "document_id": "terminology:misalignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI misalignment" ] }, "concept_id": "misalignment", "preferred": { "en": "misalignment", "ja": "ミスアラインメント" }, "relations": { "relations": { "prerequisite": [ "alignment_target" ] } }, "status": "active" }, "path": "terminology/misalignment.txt", "title": "misalignment" }, { "document_id": "terminology:neural_network", "layer": "terminology", "metadata": { "aliases": { "en": [ "neural networks" ] }, "concept_id": "neural_network", "preferred": { "en": "neural network", "ja": "ニューラルネットワーク" }, "relations": { "relations": { "broader": [ "machine_learning" ] } }, "status": "active" }, "path": "terminology/neural_network.txt", "title": "neural network" }, { "document_id": "terminology:objective", "layer": "terminology", "metadata": { "aliases": { "en": [ "objective function" ] }, "concept_id": "objective", "preferred": { "en": "objective", "ja": "目的" }, "relations": { "relations": { "broader": [ "goal" ], "related": [ "reward", "utility_function" ] } }, "status": "active" }, "path": "terminology/objective.txt", "title": "objective" }, { "document_id": "terminology:objective_robustness", "layer": "terminology", "metadata": { "aliases": { "ja": [ "目的頑健性" ] }, "concept_id": "objective_robustness", "preferred": { "en": "objective robustness", "ja": "目的の頑健性" }, "relations": { "relations": { "broader": [ "ai_alignment" ], "prerequisite": [ "base_objective", "behavioral_objective" ], "related": [ "outer_alignment" ] } }, "status": "active" }, "path": "terminology/objective_robustness.txt", "title": "objective robustness" }, { "document_id": "terminology:ontology_identification_problem", "layer": "terminology", "metadata": { "aliases": { "en": [ "ontology identification" ] }, "concept_id": "ontology_identification_problem", "preferred": { "en": "ontology identification problem" }, "relations": { "relations": { "broader": [ "alignment_problem" ], "prerequisite": [ "alignment_problem", "alignment_target", "world_model" ], "related": [ "outer_alignment", "utility_function" ] } }, "status": "active" }, "path": "terminology/ontology_identification_problem.txt", "title": "ontology identification problem" }, { "document_id": "terminology:optimization", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "optimization", "preferred": { "en": "optimization", "ja": "最適化" }, "relations": { "relations": { "prerequisite": [ "objective" ] } }, "status": "active" }, "path": "terminology/optimization.txt", "title": "optimization" }, { "document_id": "terminology:optimizer", "layer": "terminology", "metadata": { "aliases": { "ja": [ "最適化器", "オプティマイザー" ] }, "concept_id": "optimizer", "preferred": { "en": "optimizer", "ja": "オプティマイザ" }, "relations": { "relations": { "prerequisite": [ "objective", "optimization" ] } }, "status": "active" }, "path": "terminology/optimizer.txt", "title": "optimizer" }, { "document_id": "terminology:orthogonality_thesis", "layer": "terminology", "metadata": { "aliases": { "ja": [ "直交仮説" ] }, "concept_id": "orthogonality_thesis", "preferred": { "en": "Orthogonality Thesis", "ja": "直交性テーゼ" }, "relations": { "relations": { "prerequisite": [ "goal_directed_behavior", "terminal_goal" ], "related": [ "paperclip_maximizer" ] } }, "status": "active" }, "path": "terminology/orthogonality_thesis.txt", "title": "Orthogonality Thesis" }, { "document_id": "terminology:outcome_supervision", "layer": "terminology", "metadata": { "aliases": { "en": [ "outcome-based supervision", "outcome-based training" ] }, "concept_id": "outcome_supervision", "preferred": { "en": "outcome supervision" }, "relations": { "relations": { "broader": [ "oversight" ], "prerequisite": [ "training" ], "related": [ "process_supervision" ] } }, "status": "active" }, "path": "terminology/outcome_supervision.txt", "title": "outcome supervision" }, { "document_id": "terminology:outer_alignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "outer alignment problem", "reward misspecification problem" ] }, "concept_id": "outer_alignment", "preferred": { "en": "outer alignment", "ja": "アウターアラインメント" }, "relations": { "relations": { "broader": [ "ai_alignment" ], "prerequisite": [ "ai_alignment", "alignment_target", "training_objective" ], "related": [ "reinforcement_learning_from_human_feedback" ] } }, "status": "active" }, "path": "terminology/outer_alignment.txt", "title": "outer alignment" }, { "document_id": "terminology:outer_misalignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "outer alignment failure", "reward misspecification" ] }, "concept_id": "outer_misalignment", "preferred": { "en": "outer misalignment", "ja": "アウター・ミスアラインメント" }, "relations": { "relations": { "broader": [ "misalignment" ], "prerequisite": [ "misalignment", "outer_alignment" ], "related": [ "proxy_objective" ] } }, "status": "active" }, "path": "terminology/outer_misalignment.txt", "title": "outer misalignment" }, { "document_id": "terminology:oversight", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI oversight", "human oversight" ], "ja": [ "オーバーサイト", "監視" ] }, "concept_id": "oversight", "preferred": { "en": "oversight", "ja": "監督" }, "relations": { "relations": { "broader": [ "ai_safety" ], "related": [ "sandbagging", "situational_awareness" ] } }, "status": "active" }, "path": "terminology/oversight.txt", "title": "oversight" }, { "document_id": "terminology:p_doom", "layer": "terminology", "metadata": { "aliases": { "en": [ "probability of doom", "Pdoom" ], "ja": [ "破滅確率", "破滅の確率" ] }, "concept_id": "p_doom", "preferred": { "en": "p(doom)", "ja": "p(doom)" }, "relations": { "relations": { "prerequisite": [ "ai_risk" ] } }, "status": "active" }, "path": "terminology/p_doom.txt", "title": "p(doom)" }, { "document_id": "terminology:paperclip_maximizer", "layer": "terminology", "metadata": { "aliases": { "en": [ "paperclip maximiser" ] }, "concept_id": "paperclip_maximizer", "preferred": { "en": "paperclip maximizer", "ja": "クリップ最大化器" }, "relations": { "relations": { "prerequisite": [ "goal_directed_behavior", "terminal_goal" ] } }, "status": "active" }, "path": "terminology/paperclip_maximizer.txt", "title": "paperclip maximizer" }, { "document_id": "terminology:pivotal_act", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI-enabled pivotal act", "pivotal achievement" ] }, "concept_id": "pivotal_act", "preferred": { "en": "pivotal act", "ja": "pivotal act" }, "status": "active" }, "path": "terminology/pivotal_act.txt", "title": "pivotal act" }, { "document_id": "terminology:power_concentration", "layer": "terminology", "metadata": { "aliases": { "en": [ "centralization of power", "concentration of power" ], "ja": [ "権力の集中" ] }, "concept_id": "power_concentration", "preferred": { "en": "power concentration", "ja": "権力集中" }, "status": "active" }, "path": "terminology/power_concentration.txt", "title": "power concentration" }, { "document_id": "terminology:power_seeking", "layer": "terminology", "metadata": { "aliases": { "en": [ "power seeking" ] }, "concept_id": "power_seeking", "preferred": { "en": "power-seeking", "ja": "パワーシーキング" }, "relations": { "relations": { "broader": [ "goal_directed_behavior" ], "prerequisite": [ "goal_directed_behavior" ] } }, "status": "active" }, "path": "terminology/power_seeking.txt", "title": "power-seeking" }, { "document_id": "terminology:preferences", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "preferences", "preferred": { "en": "preferences", "ja": "選好" }, "relations": { "relations": { "related": [ "values" ] } }, "status": "active" }, "path": "terminology/preferences.txt", "title": "preferences" }, { "document_id": "terminology:process_supervision", "layer": "terminology", "metadata": { "aliases": { "en": [ "process-based supervision", "process-based training" ] }, "concept_id": "process_supervision", "preferred": { "en": "process supervision" }, "relations": { "relations": { "broader": [ "oversight" ], "prerequisite": [ "training" ] } }, "status": "active" }, "path": "terminology/process_supervision.txt", "title": "process supervision" }, { "document_id": "terminology:prosaic_alignment", "layer": "terminology", "metadata": { "aliases": { "en": [ "prosaic AI alignment" ] }, "concept_id": "prosaic_alignment", "preferred": { "en": "prosaic alignment", "ja": "プロセイック・アラインメント" }, "relations": { "relations": { "broader": [ "ai_alignment" ], "prerequisite": [ "ai_alignment", "machine_learning" ] } }, "status": "active" }, "path": "terminology/prosaic_alignment.txt", "title": "prosaic alignment" }, { "document_id": "terminology:proxy_objective", "layer": "terminology", "metadata": { "aliases": { "en": [ "proxy objective function" ], "ja": [ "プロキシ目的", "代理目的関数" ] }, "concept_id": "proxy_objective", "preferred": { "en": "proxy objective", "ja": "代理目的" }, "relations": { "relations": { "broader": [ "objective" ], "prerequisite": [ "objective" ], "related": [ "specification_gaming" ] } }, "status": "active" }, "path": "terminology/proxy_objective.txt", "title": "proxy objective" }, { "document_id": "terminology:recursive_self_improvement", "layer": "terminology", "metadata": { "aliases": { "en": [ "RSI", "recursive AI self-improvement" ] }, "concept_id": "recursive_self_improvement", "preferred": { "en": "recursive self-improvement", "ja": "再帰的自己改善" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence", "capability" ] } }, "status": "active" }, "path": "terminology/recursive_self_improvement.txt", "title": "recursive self-improvement" }, { "document_id": "terminology:red_teaming", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI red teaming", "red-teaming", "red team testing" ] }, "concept_id": "red_teaming", "preferred": { "en": "red teaming" }, "relations": { "relations": { "broader": [ "evaluation" ] } }, "status": "active" }, "path": "terminology/red_teaming.txt", "title": "red teaming" }, { "document_id": "terminology:reinforcement_learning", "layer": "terminology", "metadata": { "aliases": { "en": [ "RL" ] }, "concept_id": "reinforcement_learning", "preferred": { "en": "reinforcement learning", "ja": "強化学習" }, "relations": { "relations": { "broader": [ "machine_learning" ], "prerequisite": [ "agent", "reward", "training" ] } }, "status": "active" }, "path": "terminology/reinforcement_learning.txt", "title": "reinforcement learning" }, { "document_id": "terminology:reinforcement_learning_from_human_feedback", "layer": "terminology", "metadata": { "aliases": { "en": [ "RLHF", "reinforcement learning with human feedback" ] }, "concept_id": "reinforcement_learning_from_human_feedback", "preferred": { "en": "reinforcement learning from human feedback", "ja": "人間のフィードバックによる強化学習" }, "relations": { "relations": { "broader": [ "reinforcement_learning" ], "prerequisite": [ "human_preferences", "reinforcement_learning", "reward" ] } }, "status": "active" }, "path": "terminology/reinforcement_learning_from_human_feedback.txt", "title": "reinforcement learning from human feedback" }, { "document_id": "terminology:resource_acquisition", "layer": "terminology", "metadata": { "aliases": { "en": [ "resource-seeking", "resource seeking" ] }, "concept_id": "resource_acquisition", "preferred": { "en": "resource acquisition", "ja": "資源獲得" }, "relations": { "relations": { "broader": [ "power_seeking" ] } }, "status": "active" }, "path": "terminology/resource_acquisition.txt", "title": "resource acquisition" }, { "document_id": "terminology:responsible_scaling_policy", "layer": "terminology", "metadata": { "aliases": { "en": [ "RSP", "RSPs", "Responsible Scaling Policies" ] }, "concept_id": "responsible_scaling_policy", "preferred": { "en": "Responsible Scaling Policy" }, "relations": { "relations": { "broader": [ "ai_governance" ], "prerequisite": [ "capability_threshold" ], "related": [ "safety_case" ] } }, "status": "active" }, "path": "terminology/responsible_scaling_policy.txt", "title": "Responsible Scaling Policy" }, { "document_id": "terminology:reward", "layer": "terminology", "metadata": { "aliases": { "en": [ "reinforcement signal", "reward signal" ] }, "concept_id": "reward", "preferred": { "en": "reward", "ja": "報酬" }, "relations": { "relations": { "related": [ "training_objective" ] } }, "status": "active" }, "path": "terminology/reward.txt", "title": "reward" }, { "document_id": "terminology:reward_hacking", "layer": "terminology", "metadata": { "aliases": { "en": [ "reward-hacking" ], "ja": [ "リワードハッキング" ] }, "concept_id": "reward_hacking", "preferred": { "en": "reward hacking", "ja": "報酬ハッキング" }, "relations": { "relations": { "prerequisite": [ "reward" ] } }, "status": "active" }, "path": "terminology/reward_hacking.txt", "title": "reward hacking" }, { "document_id": "terminology:reward_tampering", "layer": "terminology", "metadata": { "aliases": { "en": [ "reward channel tampering", "reward-channel tampering", "reward process tampering" ], "ja": [ "報酬の改ざん" ] }, "concept_id": "reward_tampering", "preferred": { "en": "reward tampering", "ja": "報酬改ざん" }, "relations": { "relations": { "broader": [ "reward_hacking" ], "prerequisite": [ "reward" ], "related": [ "specification_gaming" ] } }, "status": "active" }, "path": "terminology/reward_tampering.txt", "title": "reward tampering" }, { "document_id": "terminology:risk", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "risk", "preferred": { "en": "risk", "ja": "リスク" }, "status": "active" }, "path": "terminology/risk.txt", "title": "risk" }, { "document_id": "terminology:safety_case", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI safety case", "AI safety cases", "safety cases" ] }, "concept_id": "safety_case", "preferred": { "en": "safety case" }, "relations": { "relations": { "broader": [ "ai_safety" ], "prerequisite": [ "risk" ] } }, "status": "active" }, "path": "terminology/safety_case.txt", "title": "safety case" }, { "document_id": "terminology:sandbagging", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI sandbagging", "strategic underperformance" ] }, "concept_id": "sandbagging", "preferred": { "en": "sandbagging", "ja": "サンドバッギング" }, "relations": { "relations": { "prerequisite": [ "capability", "evaluation" ] } }, "status": "active" }, "path": "terminology/sandbagging.txt", "title": "sandbagging" }, { "document_id": "terminology:scalable_oversight", "layer": "terminology", "metadata": { "aliases": { "ja": [ "スケーラブルオーバーサイト" ] }, "concept_id": "scalable_oversight", "preferred": { "en": "scalable oversight", "ja": "スケーラブル・オーバーサイト" }, "relations": { "relations": { "broader": [ "oversight" ], "prerequisite": [ "evaluation", "oversight" ], "related": [ "weak_to_strong_generalization" ] } }, "status": "active" }, "path": "terminology/scalable_oversight.txt", "title": "scalable oversight" }, { "document_id": "terminology:scheming", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI scheming", "schemer" ] }, "concept_id": "scheming", "preferred": { "en": "scheming", "ja": "スキーミング" }, "relations": { "relations": { "broader": [ "deception" ], "prerequisite": [ "deception", "goal_directed_behavior", "misalignment", "situational_awareness" ], "related": [ "training_gaming" ] } }, "status": "active" }, "path": "terminology/scheming.txt", "title": "scheming" }, { "document_id": "terminology:security_mindset", "layer": "terminology", "metadata": { "aliases": { "en": [ "security-oriented mindset" ] }, "concept_id": "security_mindset", "preferred": { "en": "security mindset", "ja": "セキュリティ・マインドセット" }, "status": "active" }, "path": "terminology/security_mindset.txt", "title": "security mindset" }, { "document_id": "terminology:self_preservation", "layer": "terminology", "metadata": { "aliases": { "en": [ "self preservation" ] }, "concept_id": "self_preservation", "preferred": { "en": "self-preservation", "ja": "自己保存" }, "relations": { "relations": { "broader": [ "power_seeking" ] } }, "status": "active" }, "path": "terminology/self_preservation.txt", "title": "self-preservation" }, { "document_id": "terminology:sharp_left_turn", "layer": "terminology", "metadata": { "aliases": { "en": [ "hard left turn" ] }, "concept_id": "sharp_left_turn", "preferred": { "en": "sharp left turn", "ja": "シャープ・レフト・ターン" }, "relations": { "relations": { "broader": [ "alignment_failure" ], "prerequisite": [ "capability_generalization", "distribution_shift", "generalization" ] } }, "status": "active" }, "path": "terminology/sharp_left_turn.txt", "title": "sharp left turn" }, { "document_id": "terminology:situational_awareness", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI situational awareness" ] }, "concept_id": "situational_awareness", "preferred": { "en": "situational awareness", "ja": "状況認識" }, "relations": { "relations": { "broader": [ "capability" ], "related": [ "world_model" ] } }, "status": "active" }, "path": "terminology/situational_awareness.txt", "title": "situational awareness" }, { "document_id": "terminology:specification_gaming", "layer": "terminology", "metadata": { "aliases": { "en": [ "reward gaming", "specification gaming behavior" ] }, "concept_id": "specification_gaming", "preferred": { "en": "specification gaming", "ja": "仕様ゲーミング" }, "relations": { "relations": { "broader": [ "outer_misalignment", "reward_hacking" ], "prerequisite": [ "objective", "outer_misalignment" ] } }, "status": "active" }, "path": "terminology/specification_gaming.txt", "title": "specification gaming" }, { "document_id": "terminology:takeoff_speed", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI takeoff speed", "speed of AI takeoff" ] }, "concept_id": "takeoff_speed", "preferred": { "en": "takeoff speed", "ja": "離陸速度" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence", "capability" ] } }, "status": "active" }, "path": "terminology/takeoff_speed.txt", "title": "takeoff speed" }, { "document_id": "terminology:takeover_seeking", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI takeover-seeking", "takeover seeking" ], "ja": [ "乗っ取り" ] }, "concept_id": "takeover_seeking", "preferred": { "en": "takeover-seeking" }, "relations": { "relations": { "broader": [ "power_seeking" ] } }, "status": "active" }, "path": "terminology/takeover_seeking.txt", "title": "takeover-seeking" }, { "document_id": "terminology:technology_explosion", "layer": "terminology", "metadata": { "aliases": { "en": [ "AI technology explosion", "technological explosion" ] }, "concept_id": "technology_explosion", "preferred": { "en": "technology explosion", "ja": "技術爆発" }, "relations": { "relations": { "prerequisite": [ "artificial_intelligence", "capability" ] } }, "status": "active" }, "path": "terminology/technology_explosion.txt", "title": "technology explosion" }, { "document_id": "terminology:terminal_goal", "layer": "terminology", "metadata": { "aliases": { "en": [ "final goal", "ultimate goal" ], "ja": [ "終端目標" ] }, "concept_id": "terminal_goal", "preferred": { "en": "terminal goal", "ja": "最終目標" }, "relations": { "relations": { "broader": [ "goal" ], "prerequisite": [ "goal" ], "related": [ "utility_function" ] } }, "status": "active" }, "path": "terminology/terminal_goal.txt", "title": "terminal goal" }, { "document_id": "terminology:training", "layer": "terminology", "metadata": { "aliases": { "en": [ "model training" ] }, "concept_id": "training", "preferred": { "en": "training", "ja": "学習" }, "relations": { "relations": { "broader": [ "optimization" ], "prerequisite": [ "optimization" ] } }, "status": "active" }, "path": "terminology/training.txt", "title": "training" }, { "document_id": "terminology:training_gaming", "layer": "terminology", "metadata": { "aliases": { "en": [ "playing the training game", "training game", "training gaming", "training-gamer" ], "ja": [ "訓練ゲーム", "訓練ゲーミング" ] }, "concept_id": "training_gaming", "preferred": { "en": "training-gaming", "ja": "トレーニング・ゲーミング" }, "relations": { "relations": { "prerequisite": [ "reward", "situational_awareness", "training" ] } }, "status": "active" }, "path": "terminology/training_gaming.txt", "title": "training-gaming" }, { "document_id": "terminology:training_objective", "layer": "terminology", "metadata": { "aliases": { "en": [ "training objective function" ], "ja": [ "訓練目的関数" ] }, "concept_id": "training_objective", "preferred": { "en": "training objective", "ja": "学習目的関数" }, "relations": { "relations": { "broader": [ "objective" ], "prerequisite": [ "training" ] } }, "status": "active" }, "path": "terminology/training_objective.txt", "title": "training objective" }, { "document_id": "terminology:trusted_monitoring", "layer": "terminology", "metadata": { "aliases": { "en": [ "trusted model monitoring", "trusted monitoring protocol" ] }, "concept_id": "trusted_monitoring", "preferred": { "en": "trusted monitoring" }, "relations": { "relations": { "broader": [ "control_protocol", "oversight" ], "prerequisite": [ "control_protocol", "oversight" ], "related": [ "untrusted_monitoring" ] } }, "status": "active" }, "path": "terminology/trusted_monitoring.txt", "title": "trusted monitoring" }, { "document_id": "terminology:untrusted_monitoring", "layer": "terminology", "metadata": { "aliases": { "en": [ "untrusted model monitoring", "untrusted monitoring protocol" ] }, "concept_id": "untrusted_monitoring", "preferred": { "en": "untrusted monitoring" }, "relations": { "relations": { "broader": [ "control_protocol", "oversight" ], "prerequisite": [ "control_protocol", "oversight" ] } }, "status": "active" }, "path": "terminology/untrusted_monitoring.txt", "title": "untrusted monitoring" }, { "document_id": "terminology:utility_function", "layer": "terminology", "metadata": { "aliases": { "en": [ "utility functions" ] }, "concept_id": "utility_function", "preferred": { "en": "utility function", "ja": "効用関数" }, "relations": { "relations": { "prerequisite": [ "preferences" ] } }, "status": "active" }, "path": "terminology/utility_function.txt", "title": "utility function" }, { "document_id": "terminology:value_fragility", "layer": "terminology", "metadata": { "aliases": { "en": [ "fragility of value", "value is fragile" ] }, "concept_id": "value_fragility", "preferred": { "en": "value fragility", "ja": "価値の脆弱性" }, "relations": { "relations": { "prerequisite": [ "human_values", "optimization" ] } }, "status": "active" }, "path": "terminology/value_fragility.txt", "title": "value fragility" }, { "document_id": "terminology:values", "layer": "terminology", "metadata": { "aliases": {}, "concept_id": "values", "preferred": { "en": "values", "ja": "価値観" }, "status": "active" }, "path": "terminology/values.txt", "title": "values" }, { "document_id": "terminology:weak_agi", "layer": "terminology", "metadata": { "aliases": { "en": [ "weak artificial general intelligence" ], "ja": [ "弱いAGI" ] }, "concept_id": "weak_agi", "preferred": { "en": "weak AGI", "ja": "弱AGI" }, "relations": { "relations": { "broader": [ "artificial_general_intelligence" ], "prerequisite": [ "artificial_general_intelligence", "capability" ] } }, "status": "active" }, "path": "terminology/weak_agi.txt", "title": "weak AGI" }, { "document_id": "terminology:weak_to_strong_generalization", "layer": "terminology", "metadata": { "aliases": { "en": [ "W2SG", "weak to strong generalization" ] }, "concept_id": "weak_to_strong_generalization", "preferred": { "en": "weak-to-strong generalization" }, "relations": { "relations": { "broader": [ "generalization" ], "prerequisite": [ "generalization" ] } }, "status": "active" }, "path": "terminology/weak_to_strong_generalization.txt", "title": "weak-to-strong generalization" }, { "document_id": "terminology:world_model", "layer": "terminology", "metadata": { "aliases": { "en": [ "world-model" ] }, "concept_id": "world_model", "preferred": { "en": "world model" }, "status": "active" }, "path": "terminology/world_model.txt", "title": "world model" } ], "format_version": 1 }