Bypassing Model Thinking Limits: Free Internal CoT via Custom Tools
ivan_bezdomny · x · 2026-08-12
A user discovered a clever workaround to bypass a model's default thinking restrictions: simply disable the built-in thinking mode and provide a custom tool named deepthink. The model will then actively call this tool, outputting its reasoning process using an internal Chain-of-Thought format within the tool's parameters.
This creative prompt engineering and tool-calling trick unexpectedly extracts the hidden internal CoT, drawing attention and amusement from the community.
Related event: Trick Extracts AI's Internal CoT via Custom Tool(2 posts)→
More from Fun
- When the AI Coding Agent Desperately Asks for Permission to Push Code — adonis_singh · 2026-08-12
- Commentary: AI Competition and Capital Misallocation, Good Projects Die — ctjlewis · 2026-08-12
- e/acc's Beff Jezos flags a 'vibe shift' as more users daily drive Grok — beffjezos · 2026-08-12
- Hilarious Thread: I Hired an AI Agent and It Became My Digital Twin — tallmetommy · 2026-08-12
- SF Tech Culture Meme: Signaling Status Through Codeforces Flexing — aryaman2020 · 2026-08-12
- Malformed Codex Delete Command Wipes User's System Files Due to Quoting Error — SgtSilock · 2026-08-12