Glitch Exposes Gemini's Internal Thoughts
Regular_Preference64 · reddit · 2026-08-28
Users reported an anomaly with Gemini today where the model's internal processing was accidentally exposed in responses.
Symptoms:
- Model starts responses with "sff" during normal conversations
- Outputs text like "The user wants to know" which appears to be hidden system prompts or chain-of-thought
- Proceeds to give the actual response afterwards
Speculation:
It appears that the "sff" marker, likely intended as a hidden delimiter, is being rendered in the chat interface due to a system error.
More from Models
- Grok gives wrong answers on movie times, park status, and car specs — JoeJustice · 2026-08-28
- ThursdAI Recap: GLM 5.3 and Qwen 27B Drop, OpenAI Pauses RL for Safety — altryne · 2026-08-28
- GLM-5.2 Introduces Monitors to Combat Reward Hacking in RL — burny_tech · 2026-08-28
- Anthropic Luna Max test shows generous limits, high speed — timpera · 2026-08-28
- User calls Grokbot 'terrible', cites missing tasks — krishnan · 2026-08-28
- Prime Intellect Evaluates Autonomous AI Research Capabilities Across 18 Frontier Models — mariofilhoml · 2026-08-28