Kaomoji emotion probe validated: 0.68 correlation with valence direction, 0.79 when trained
Sauers_ · x · 2026-11-21
The full writeup of the Qwen3-8B kaomoji experiment with three validation numbers: the model's self-rating of a conversation correlates 0.52 (rank) with its next kaomoji; directly projecting layer-18 activations onto a valence direction reaches 0.68; and a probe trained on kaomoji ratings predicts them at 0.79. The setup induces "moods" in real chats (good or bad news first), then reads them out via a prefilled "I'm feeling" + kaomoji. Conclusion: the model's kaomoji choices reliably reflect an internal emotional state that's linearly readable from activations — a zero-cost, reproducible introspection probe.
Related event: Emoji Probes Quantify Qwen3-8B's "Happiness"(3 posts)→
More from Models
- METR Q4: $90k of expert humans still can't beat frontier models, horizons hit 27h — iruletheworldmo · 2026-11-21
- Kaomoji as an emotion probe: Qwen3-8B's internal "happiness" is measurable — Sauers_ · 2026-11-21
- 27,262 thinking tokens, zero answer: Google auto-billed 23 empty Vertex AI crashes via HTTP 200 — Plenty-Jackfruit-218 · 2026-11-21
- Thinking Machines' new model claimed 5x faster than Qwen3.8, 15% more accurate — LiTianleli · 2026-11-21
- Gemini silently saves every Google search that gets an AI response, user finds — The-Neofighter · 2026-11-21
- Frontier Design Benchmark: Claude Fable 5.2 tops ranking, just 26 Elo below human designs — yuwen_lu_ · 2026-11-21