Claude Opus keeps making the same spoon and fork errors even when told explicitly
ivan_bezdomny · x · 2026-10-04
Testing image generation, the author found Claude Opus repeatedly makes the same structural mistakes when drawing spoons and forks. Even though the prompt explicitly warned about the error, and follow-up hints like "you have a problem" or "compare your side view with a photo" failed to help, the model either can't see the issue or argues there is none. The author bets a third from-scratch attempt would fail identically.
More from Models
- LocalMaxxing: community-run speed tests for local LLM rigs, 8.6K runs across 592 hardware setups — lxfater · 2026-10-04
- KOL pans OpenAI's 'dots' keynote release, says it's an easy win for Anthropic — kimmonismus · 2026-10-04
- Claude is too pessimistic: user says it shoots down ideas instead of exploring them — Jumpy-Cobbler1020 · 2026-10-04
- Claude is now asking regular users for interviews — even non-coders — imfamilyfriendlysd · 2026-10-04
- GPT image generation wows blogger in page-34 test — aiamblichus · 2026-10-04
- Same open-weights model doom-loops in one engine, runs fine on llama.cpp — pand5461 · 2026-10-04