weak AGI

An AI system that counts as artificial general intelligence in the relevant context but is comparatively weak relative to stronger AGI or artificial superintelligence. In one explicit Alignment Forum usage, weak AGI is a true AGI able to learn and competently perform the tasks humans do while still performing below a skilled human; other safety-planning usage is looser and uses 'weak' comparatively for a less capable or less autonomous intermediate AGI. The term therefore has no fixed quantitative threshold and does not by itself imply that the system is safe, non-agentic, or unable to cause severe harm.