User says Anthropic deliberately nerfed models' AI R&D capabilities during training
burny_tech · x · 2026-11-10
- X user burnytech reports that while experimenting with combining evolutionary strategies and novelty search in LLM training, he worried Claude might refuse or sandbag, and switched to Astra instead.
- The post quotes theobearman's observation that Anthropic took steps to nerf the models' AI R&D capabilities during training.
- A rare signal that a frontier lab may be intentionally limiting its models' automated AI-research capabilities.
More from Models
- Fable 5.2 tops Remote Labor Index at 41.3%, nearly doubling next-best Opus 5.5 — haider1 · 2026-11-10
- Re-running an old castle prompt: Fable 5.2's procedural Blender output is a visible jump — aitrendz_xyz · 2026-11-10
- Fable 5.2 plus a few Opus subagents produces a complex 3D scene in ~30 minutes — aitrendz_xyz · 2026-11-10
- Same prompt head-to-head: Fable 5.2 spends 4x more than Opus 5.5 for better lighting — aitrendz_xyz · 2026-11-10
- mark_k pours cold water on Fable 5.2 AGI hype: "Let's not get carried away" — mark_k · 2026-11-10
- Fable 5.2 Reviews Itself From 160 User Reports: 5 Rules for Using It Cheaply — henrik392 · 2026-11-10