Qwen 3.8 27B produces a surprisingly foul-mouthed chain of thought
CabinetNational3461 · reddit · 2026-09-21
While running Qwen 3.8 27B in a DeepSeek harness, the author glanced at the progress and found the model's thought process unexpectedly full of profanity — a classic funny model moment, screenshot included.
More from Fun
- Jev × GPT-6 Astra demo: a context-aware AI assistant that speaks only when needed — i_dg23 · 2026-09-21
- Dev building a fast, brawly action roguelike shares its most magical part so far — TAbrodi · 2026-09-21
- Redditor recreates TaleSpin as chillwave with Minimax video and YuE2 music — the_bollo · 2026-09-21
- Creator Behind Codex UGC Billboards Shares Her AI Origin Story — paw_lean · 2026-09-21
- eigenrobot comes out as an effective altruist — by donating character-building time — eigenrobot · 2026-09-21
- Patrick Collison's 'Fast' page catalogs ambitious projects done in weeks, not years — itsOmSarraf_ · 2026-09-21