Wes Roth Video posted an update
Fable 5.2 had to be crippled
Why it mattersThe video discusses Claude Fable 5.2's claimed coding, robotics, hardware-design and physics capabilities, and reports that Anthropic deliberately capped some accelerator-kernel and model-architecture research skills during training. It also examines a METR report, a secret autonomous-experiment scaffold, and a custom game and language-model benchmark.
Discuss: What trade-off does the reported cap on accelerator-kernel optimization and model-architecture research suggest for Claude Fable 5.2?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.