AI safety researcher: interpretability and alignment methods work but not reliably, and we don't know how to fix that
DavidSKrueger · x · 2026-09-22
David Krueger, AI safety researcher, responded to skeptics by noting he has published work in all four relevant areas — interpretability, testing, alignment, and control — at top AI venues. His point: the methods exist and are not useless, but they don't work reliably, and the field doesn't know how to fix that. We can't understand how AI systems work, predict their behavior, prevent misbehavior, or stay in control when it happens — foundational open problems despite years of effort.
Related event: Cambridge Researcher: Four Unresolved Problems Make AI Safety No Quick Fix(2 posts)→
More from AGI Musings
- OpenAI, Anthropic and DeepMind job posts signal PMs must write evals and code — aakashgupta · 2026-11-12
- Architecture study sparks monitoring concerns over a multi-hop planning spike — ChowdhuryNeil · 2026-11-12
- PowerShell creator Jeffrey Snover at Ignite: AI revolution is an extinction event for those who don't adapt — dfinke · 2026-11-12
- Worst AI use: signaling you don't care about things you should — Afinetheorem · 2026-11-12
- Stanford HAI blueprint: 56 US agencies report 3,611 AI use cases, evaluation lags — StanfordHAI · 2026-11-12
- Fleuret amplifies debate: AI can't slash wet-lab experiments in unmapped biology — francoisfleuret · 2026-11-12