AI alignment bottleneck is human incompetence, not model capability

jeremygillen1 · x · 2026-08-19

Jeremy Gillen clarifies that he does not consider LLMs incompetent, finding them helpful for research. He argues that the "incompetence" limiting the value of AI for alignment research is a property of humans, not the AI models themselves.

Related event: Alignment researchers debate whether today's AI risks stem from prosaic failures or philosophy(7 posts)→

Original post →

More from Safety

Safety channel →