sources (68)
- Bioshok vs Akira P(doom)徹底討論書き起こし(修正なし全文)
- X @Align_ASI post 2055116276960559467
- X @Align_ASI post 2055116279485489376
- X @Align_ASI post 2055121067929489784
- X @Align_ASI post 2055466196200505634
- X @Align_ASI post 2055571488686674072
- X @Align_ASI post 2056249992826982411
- X @Align_ASI post 2056298561248039229
- X @Align_ASI post 2056306172353991103
- X @Align_ASI post 2056653057535205557
- X @Align_ASI post 2056653932890583189
- X @Align_ASI post 2056760361928519950
- X @Align_ASI post 2087017636877996133
- The Problem
- Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
- AGIがもたらす産業・技術爆発
- AI安全性と国家安全保障の対立
- The Superintelligent Will: Motivation and Instrumental Rationality in Advanced Artificial Agents
- Automated alignment is harder than you think
- Scheming AIs: Will AIs fake alignment during training in order to get power?
- A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders
- My views on “doom”
- Takeoff speeds
- What failure looks like
- Where I agree and disagree with Eliezer
- ARC's first technical report: Eliciting Latent Knowledge
- Should Artificial Intelligence Governance be Centralised? Design Lessons from History
- #235 – Ajeya Cotra on whether it’s crazy that every AI company’s safety plan is ‘use AI to make AI safe’
- What a compute-centric framework says about AI takeoff speeds
- Scaling Laws For Scalable Oversight
- Most AI value will come from broad automation, not from R&D
- Will AI R&D Automation Cause a Software Intelligence Explosion?
- Current AIs seem pretty misaligned to me
- Jankily controlling superintelligence
- Alignment faking in large language models
- Risks from Learned Optimization in Advanced Machine Learning Systems
- Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
- How Afraid of the A.I. Apocalypse Should We Be?
- A sketch of an AI control safety case
- Sparse Autoencoders Do Not Find Canonical Units of Analysis
- Natural Emergent Misalignment from Reward Hacking in Production RL
- The Alignment Problem from a Deep Learning Perspective
- Weak-to-strong generalization
- Why Corrigibility is Hard and Important (i.e. "Whence the high MIRI confidence in alignment difficulty?")
- Computing Power and the Governance of Artificial Intelligence
- An International Agreement to Prevent the Premature Creation of Artificial Superintelligence
- A central AI alignment problem: capabilities generalization, and the sharp left turn
- Optimal Policies Tend to Seek Power
- Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong Generalization
- AGI Ruin: A List of Lethalities
- Artificial Intelligence as a Positive and Negative Factor in Global Risk
- Autonomous AGI
- Coherent Extrapolated Volition
- 'Empiricism!' as Anti-Epistemology
- Irretrievability; or, Murphy's Curse of Oneshotness upon ASI
- There's No Fire Alarm for Artificial General Intelligence
- Orthogonality Thesis
- Pivotal act
- Security Mindset and the Logistic Success Curve
- Security Mindset and Ordinary Paranoia
- Shah and Yudkowsky on alignment failures
- Pausing AI Developments Isn't Enough. We Need to Shut it All Down
- The Sun is big, but superintelligences will not spare Earth a little sunlight
- Will superintelligent AI end the world?
- Thou Art Godshatter
- Value is Fragile
- Ngo and Yudkowsky on alignment difficulty
- Why not just read the AI's thoughts?