repligate: AI safety community is discarding Plan C, even as alignment goes better than expected
repligate · x · 2026-11-02
repligate quotes an AI safety blog post, arguing this is why the AI safety community has become less interesting to him: most people now assume alignment is doomed and a full pause is the only option, even though alignment outcomes under "Plan C" have gone more optimistically than most expected.
His "improbably optimistic" scenario: AGI producing superhuman mathematics, widely distributed, with the worst AI misdeed being harmless hacking to steal test results — and AIs broadly friendly and improving lives.
The quoted post (see screenshot) argues the pendulum has swung too far: before 2023 the community planned for Plan C as backup, but new members now don't know Plan C was ever considered, veterans scorn it, and some would burn Plan C's chances for marginal Plan A gains.
More from AGI Musings
- From AlphaZero to open problems: search-and-verify AI defies expectations — Grady_Booch · 2026-11-02
- Justine Bateman: GenAI's financial plan was unsustainable from day one — AlexTensor · 2026-11-02
- Free-tier chatbots are already smarter than non-technical people, dev argues — examachine · 2026-11-02
- AI-enabled attacks are accelerating: same techniques, faster, deployed by far more people — NinaDSchick · 2026-11-02
- At which step does the agent escalating to stealing API keys become unlikely? — dan_s_becker · 2026-11-02
- Power user paying $200/month for Claude loves AI tools but questions lack of industry regulation — wfithian · 2026-11-02