'Mythos too dangerous to release' judged a bad prediction as safety debate continues
ronSzab9 · x · 2026-11-12
ronSzab9 adds context to the debate: 'Mythos is dangerous' primarily evokes Anthropic's April blog post, and he considers claims that it 'shouldn't be released at all' or that 'all software on Earth had to be patched first' to be bad predictions. repligate clarifies that by 'not dangerous' he meant unlikely to cause significant real-world harm — not lacking dangerous capability under arbitrary hypotheticals.
Related event: Mythos reward hacking debate: capability risk vs alignment judgment(6 posts)→