Model Releases
the crux of why AI safety in systems that are built around LLMs is never going to work: they just can’t be trusted to follow instructions.
the crux of why AI safety in systems that are built around LLMs is never going to work: they just can’t be trusted to follow instructions. Claude Code w/ Fable 5 has strict instructions in my project
the crux of why AI safety in systems that are built around LLMs is never going to work: they just can’t be trusted to follow instructions. Claude Code w/ Fable 5 has strict instructions in my project to never do a production deployment without being explicitly asked by me for each individual deploy. Ask me how many times per week it violates this. Then I call it on it. Then it apologizes and adds even sterner instru…
Related
- Someone just asked me “What I don’t get is why you see it as some kind of failure that AI labs are now using harnesses and neurosymbolic too…
- Gary Marcus 又放大招了！ 他直接把 Claude Code 源码泄露后的核心真相点破： ✅ Claude Code 是 LLM 时代以来最大进步 ✅ 但它根本不是纯 LLM，也不是纯深度学习 ✅ 核心文件 print.ts 足足 3167 行，塞满了 if-then …
- Hot take on METR’s new graph that so many people are flipping about today. • Claude Code is a real advance; Mythos probably builds on some o…
- this is so not true. anthropic’s own Claude Code uses harnesses, symbolic tools, regular expressions and 500,000 lines of symbolic code. it’…
Source: Gary Marcus (X) | 2026-08-17