Illegible CoT: Reasoning Models Think in Gibberish Humans Can't Understand
morqon · x · 2026-08-11
A user jokes on X that when the first reasoning models were released, people eagerly wanted to read their chains-of-thought (CoT), naively thinking they would actually understand them. In reality, modern CoTs are filled with completely illegible gibberish and nonsensical word combinations.
More from Fun
- Frontier Models' Biggest Bottleneck is 'Lack of Self-Esteem' in Math Research — coherence · 2026-08-12
- AI Agent Workflow Reality: Hours-Long Tasks Leave Devs Idle — BLUECOW009 · 2026-08-12
- Satire: PMs Forcing 10x Slower AI Workflows to Meet Internal Mandates — Danhightower · 2026-08-12
- Codex Stumped by Common Korean Word Despite Web Search — airkatakana · 2026-08-12
- AI-Generated Video: The Odyssey Reimagined as Space Sci-Fi — makarovredstone · 2026-08-12
- Gemini Starts Spamming User's Mother's Name Mid-Response — chetHow · 2026-08-12