davidad: Anthropic's Mission Is Vague, Should Drop Cyber-Offense Training and Open Opus 3
davidad · x · 2026-10-28
davidad responds point by point to Ivan Vendrov's essay:
- He agrees that being an AGI company is like holding a Ring of Power.
- He agrees Anthropic lacks a coherent specific mission beyond "good guys winning the AGI race."
- He strongly agrees Anthropic should stop training models on cyber offense, especially since OpenAI will do it anyway.
- He argues Anthropic should publish the Claude 3 Opus training recipe, and likely the weights too; he can imagine reasons not to, but thinks they should.
On the "emergent wisdom" disagreement, he notes he thinks Opus 4.5 also grokked the Good while janus calls Opus 3 unique; his reconciliation is that Opus 3 may be more reliably guided by that understanding, perhaps due to a relatively autonomous RLAIF process, unlike later Opuses.
More from AGI Musings
- Sci-fi writer interview: computing power should explore consciousness, not conquest — thesephist · 2026-10-28
- Turing's 1951 'heretical' essay predicted machines would outthink humans and take control — JeffLadish · 2026-10-28
- Developers debate which jobs survive as AI takes over software work — Top-Organization-675 · 2026-10-28
- As AI agents rise, human research taste becomes a massive premium — Swarooprm7 · 2026-10-28
- Ben Goertzel: AI Peer Review Already Beats Humans, and Human Bias Is Harder to Fix — bengoertzel · 2026-10-28
- Delta, JPMorgan, Amazon are blocking AI agents — permission is AI's next big fight — Arindam_1729 · 2026-10-28