Richard Ngo: would-be alignment researchers at Anthropic drift into capabilities and PR work
RichardMCNgo · x · 2026-09-21
Richard Ngo argues normal science could have helped alignment if a proper paradigm had been established, but most people suited for that now do "thinly-disguised capabilities work (or heavily PR-optimized work)" at Anthropic. He also debates better terms for pre-paradigmatic problems — wicked, category II, etc. A notable inside-baseball debate on alignment research culture.
Related event: Richard Ngo calls AI safety a wicked problem, criticizes Anthropic(2 posts)→
More from AGI Musings
- Ex-Copilot engineer: tools will change, but humans are wired to keep building — rakyll · 2026-09-21
- Ex-GitHub Copilot engineer: coding agents speed everything up, yet leave us more work than ever — rakyll · 2026-09-21
- X exec: seasoned software execs ill-suited for probabilistic AI products, young builders hold the edge — nikitabier · 2026-09-21
- Gwern's 2016 'Tool AIs Want to Be Agent AIs' thesis vindicated, says founder — tszzl · 2026-09-21
- An 8-minute argument that the process-of-life is morally primary over consciousness — danfaggella · 2026-09-21
- Investor Stewart Alsop: the real alignment problem is smart people's collective delusions — StewartalsopIII · 2026-09-21