Kundaje: crippling open models' bio knowledge is a bad idea, incentive problem unsolved
anshulkundaje · x · 2026-09-17
Stanford professor Anshul Kundaje shared his evolving take in a thread. His strongest disagreement is with crippling open models with respect to biological knowledge: while there are many arguments why this is a bad idea, he admits he hasn't figured out how to incentivize open-source efforts to implement safeguards as capabilities improve, calling it the specific point he most wants sharper arguments on.
More from AGI Musings
- Schmidhuber Publishes 40-Year Retrospective on Recursive Self-Improvement Since 1987 — rbhar90 · 2026-09-17
- Dedekind quote on divine lineage and human creative power — zetalyrae · 2026-09-17
- Aerospace engineer: admitting your AI needs observability should be a selling point — jonerp · 2026-09-17
- Dev compares OpenAI and Anthropic to 1999 AOL: 'gatekeeping intelligence like ISPs feared the open web' — adamamcbride · 2026-09-17
- Star Trek's Moriarty Saw It Coming: What the Hugging Face Agent Incident Teaches Us — chickey23 · 2026-09-17
- Pedro Domingos: Like 1950s cars, AI should be improved, not arrested — pmddomingos · 2026-09-17