Qwen3-32B can name which paragraph it attended to most — above chance, no injection needed
Sauers_ · x · 2026-11-19
In follow-up experiments, Sauers shows Qwen3-32B can identify which paragraph it naturally attended to most — without any injection — either self-reporting in the same chat or judging from just reading the transcript.
Chart data: rank correlation between self-report and true attention is 0.42 (p<10⁻⁵); 0.49 when a second instance reads the transcript, with 96% top-1 accuracy on the correct paragraph. The ability to control attention peaks sharply at early-mid layers, near zero in early layers.
More from Research
- RhoAgent in Nature Biotech brings AI design from proteins to programmable functional RNA — EricTopol · 2026-11-19
- Designing a humanoid robot foot: heel strikes hit 1511N, whole foot weighs 117g — dee_hw · 2026-11-19
- Google's GUARD paper: frozen base model, self-editing agent harness gains +8.89 points — omarsar0 · 2026-11-19
- Kezek splits mono two-person conversations into per-speaker tracks for full-duplex dialogue training data — manifestro · 2026-11-19
- Elie Bakouch: current AI research benchmarks overfit and can't track RSI — eliebakouch · 2026-11-19
- Hutter Prize text compression jumps ~10% by overfitting small transformers for arithmetic coding — ricklamers · 2026-11-19