GPT-5.6 and Claude Hallucinate Messages in Noise Image
In a small AI mishap that drew attention online, users asked GPT-5.6 Sol and Claude Fable 5 to detect a hidden message in a 1024×1024 binary noise image that actually contained no embedded information. Both models still produced confident, invented interpretations, sparking discussion about multimodal hallucination and overconfident responses.
2026-07-12 ~ 2026-07-13 · 3 related posts
- Models Hallucinate Messages in Noise Images — goodside · 2026-07-12
- Two Models Hallucinate Messages in Noise Image — goodside · 2026-07-12
- Hilarious Failures from GPT-5.6 and Claude — steipete · 2026-07-13