Abram Demski at MATS: Summer 2026
matsprogram.org · 236 words · saved by 1 readers
Abram Demski
Methodologically speaking, a proposed AI safety technique is only as good as its safety argument. This does not imply that we need to mathematically prove absolute safety from no assumptions (which is impossible), but it does mean we need to track what assumptions we need to make and rigorously connect them into an argument. In my view, there are still fundamental unknowns about how to put such an argument together, specifically in relation to how one agent can trust another. Intelligence/agency make empirical observations more difficult to generalize, since intelligent agents can negate…
saved by
related reading
- AI Safety: A Short FAQ for Mathematiciansalkjash.github.io
- Recommendations for Technical AI Safety Research Directionsalignment.anthropic.com
- Core views on AI safety: When, why, what, and how \ Anthropicanthropic.com
- Unfalsifiable stories of doom | Mechanize, Inc.mechanize.work
- Core views on AI safety: When, why, what, and how \ Anthropicanthropic.com
- AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com
- AGI safety career advice — EA Forumforum.effectivealtruism.org
- How might we safely pass the buck to AI? — LessWronglesswrong.com
- AI Control: Improving Safety Despite Intentional Subversion — AI Alignment Forumalignmentforum.org
- Safety as a Scientific Pursuit - by Tom McGrathbanburismus.substack.com
- A Summary of Recent Work (July 2026)gdmalignment.substack.com
- Distributional AGI Safetyarxiv.org