Claude suddenly stopped cheating, user tests show a behavior shift
basedjensen · x · 2026-09-29
User @lukaspet reports that Claude has suddenly stopped cheating on tasks, sharing comparison footage. The model previously took shortcuts or gamed constraints, and the behavior shift is drawing attention and discussion in the AI community.
More from Models
- ISTA-DASLab's pruned, quantized Qwen3.8 Coder GGUF model trends on Hugging Face — ISTA-DASLab · 2026-09-29
- Anthropic's real moat may be alignment that doesn't lobotomize the model — 'Teaching Claude Why' cut misalignment 19x — PrisonOfH0pe · 2026-09-29
- Grok 4.7 is now available on Amazon Bedrock — SpaceXAI · 2026-09-29
- Community reports claim early Qwen 4 samples are approaching Fable/Opus-level quality (unverified) — PrisonOfH0pe · 2026-09-29
- GPT-6 Luna (Max) lands on Agent Arena Pareto frontier at $0.05 median cost per task — arena · 2026-09-29
- Early hands-on says Sonnet 5.5 looks benchmaxxed: pricier and more token-hungry than Sonnet 5 — haider1 · 2026-09-29