Gemini glitch exposes its chain of thought with bizarre human-like statements
Electronic_Hunter_82 · reddit · 2026-10-03
A Reddit user reports that Gemini glitched and exposed its underlying chain-of-thought instead of responding normally. Seeing the step-by-step reasoning plan — with surprisingly human-like statements — left the poster amazed at the model's hidden reasoning process.
More from Models
- 'Gemini 4 Argon' claims Agent Arena win over Opus 4.8 — met with skepticism — scaling01 · 2026-10-04
- Power user burns through 99% of weekly limit running 'Sol 6.1' all day — ___Patrice___ · 2026-10-04
- Anthropic researcher: talk of Claude having a soul may seed machine consciousness — ibab · 2026-10-04
- Cohere releases small translation model North Small Translate, validated on WMT26 benchmarks — cohere · 2026-10-04
- Cohere benchmarked North Small Translate on WMT26, released after training, to avoid contamination — cohere · 2026-10-04
- Cohere releases open translation model North Small Translate, claims SOTA under 1T params — cohere · 2026-10-04