EMNLP paper introduces Concept-Targeted Attribution, tracing the circuits behind linear probes
ZhijingJin · x · 2026-10-25
Zhijing Jin's team published a new paper at #EMNLP2026, How Do Linear Probes Emerge?, introducing Concept-Targeted Attribution (CTA).
- Existing probes only tell whether a model has linearly encoded a concept, not which features or circuits produce that encoding; CTA trains attribution graphs around a probe's concept direction instead.
- Paper and code links are in the thread; authors include Vedant Palit, Bernhard Schölkopf, and Zhijing Jin.
Related event: New EMNLP Paper Introduces CTA to Explain How Linear Probes Emerge(2 posts)→
More from Research
- AI for systems research: basic agent swarms plus good metrics get ideas shipped — sh_reya · 2026-10-25
- 1,000+ mathematicians form Association for Human Mathematics, pledge not to publish AI-generated papers — Hesamation · 2026-10-25
- CritPt, SciCode and Terminal-Bench Science: three science benchmarks as RL environments — geoffwolfe · 2026-10-25
- Robotics researcher: 2 days is enough to ship a follow-up paper on supermimic — ChongZzZhang · 2026-10-25
- bbtree claims fastest concurrent btree: 5x faster than std BTreeMap single-threaded — tianyin_xu · 2026-10-25
- Covert channel math: 128-bit secret leaks in 13s, 1TB weights take centuries — vishalmisra · 2026-10-25