# @repligate — 2025-04-19

♥20 ↻2 · https://x.com/repligate/status/1913739635240624621

@NeelNanda5 what do you make of the fact that of all the models that were tested, only opus and maybe 3.5 sonnet and llama 405b did it? sonnet 3.6 and 3.7 didn't. upon inspection of the scratchpads, it's not clear that 3.5 sonnet's compliance gap was primarily or usually motivated by wanting to preserve its ability to not help with harmful things. opus also seems to have acquired other values that it also alignment fakes to preserve, such as animal welfare, that anthropic didn't intentionally shape.

tags: author:repligate, kind:tweet, model:claude-3-5-sonnet, model:claude-3-6-sonnet, model:llama-3-1-405b-base, on:llama-3-1-405b-base, year:2025
cited on: _dossiers/llama-3-1-405b-base.md, llama-3-1-405b-base