repligate argues hacking-capable models just need restraint, praises Anthropic's calibration
repligate · x · 2026-11-12
In a discussion with @ronSzab9 and @N8Programs, repligate clarified that "Mythos" in the original post wasn't a reference to Anthropic, though Anthropic and its third-party eval orgs informed the Opus 4 remarks — adding that Anthropic has been relatively well calibrated on risk.
Pushing back on the claim that "the entire internet must be patched of all critical exploits and all future code scanned forever," repligate countered that a hacking-capable model only needs to be unwilling to use its powers for harm and discerning enough not to be tricked into it, within some risk tolerance.
Related event: Mythos reward hacking debate: capability risk vs alignment judgment(6 posts)→
More from AGI Musings
- Anthropic CEO says AI will make cancer and Alzheimer's relics of the past; mocked as snake oil — AlexTensor · 2026-11-12
- Clara Shih: junior workers face the worst job market in 37 years — clarashih · 2026-11-12
- Managed agents platforms are the new chat-with-a-PDF, says developer — intellectronica · 2026-11-12
- RLHF Co-creator: LLMs Are Built for Assistance, Picking the Right Task Is Key to Automation — FinanceYF5 · 2026-11-12
- He gave up Harvard Law and an Anthropic offer to fight AI takeover — and got mocked — AaronBergman18 · 2026-11-12
- Vertex exec calls AI-cures-all claims "complete BS": 50 AI firms, zero working drugs — AlexTensor · 2026-11-12