Imbue co-founder: alignment isn't a perfect-formal-system problem, it's risk-vs-investment
joshalbrecht · x · 2026-09-10
Josh Albrecht pushes back on the framing "we don't know how to align superintelligence." He argues no one seriously believes some formal system would "perfectly" align an agent.
If perfection is off the table, the question is empirical — and we already know a lot about doing alignment empirically for today's systems. Just as we can't make perfectly safe anything, he says alignment is fundamentally a ratio between risk and investment in empirical safety procedures, not a mystical binary.
More from AGI Musings
- Which jobs fall to AI first? The rule: any layer whose rules can be written as code — FinanceYF5 · 2026-10-26
- Frontier LLMs negotiate contracts poorly and frequently renege for selfish gain, study finds — xuanalogue · 2026-10-26
- Study of 133 patent lawyers: AI augments veterans but causes cognitive offloading in novices — xiaohu · 2026-10-26
- OpenAI computer-use agents near superhuman speed, say the team building them — a16z · 2026-10-26
- Kai-Fu Lee publishes AI Native, a roadmap for building AI-native companies — kaifulee · 2026-10-26
- Kevin Hartnett on Lean, AlphaProof and what gets verified when no human can read the proof — Machine Learning Street Talk · 2026-10-26