Artificially boosted attention: Qwen3-32B notices and names the injected paragraph
Sauers_ · x · 2026-11-19
Sauers' attention-injection experiment: researchers artificially amplified Qwen3-32B's attention on one paragraph (×1.8 to ×9.9) and asked whether it noticed the boost.
The stronger the injection, the more often the model answered "Yes, paragraph [P2]" — agreement rose from 13% to 22% (+63% at ×9.9, p=2×10⁻⁴), and correctly naming the boosted paragraph rose from 25% to 59% (+135%, p<10⁻⁵). This suggests genuine introspective awareness of its own attention state. The setup is conceptually modeled on Jack W. Lindsey's activation introspection methods.
More from Research
- FoodTruck Bench, an LLM agent business-sim benchmark, lands on Papers with Code — NielsRogge · 2026-11-19
- MIT reads and steers LLMs' internal physics intuition via hidden-state shifts in Gemma — ProfBuehlerMIT · 2026-11-19
- harness-hub open-sources RL-trained LoRA adapters tuned per agent harness, recipe coming Friday — yb2698 · 2026-11-19
- RL-Trained Hermes LoRA on Qwen3.8-27B Adds +13.7 Pts on SWE-Bench Pro While Using Fewer Tokens — yb2698 · 2026-11-19
- Freenet releases new video on network routing with old-school ML — Sanity · 2026-11-19
- Empire AI creates 120+ early-career research positions in New York, Cornell deadline Dec 15 — dianarycai · 2026-11-19