interwhen: test-time verification framework for steering reasoning models lands at NeurIPS
rao2z · x · 2026-09-25
The paper interwhen, done during the authors' time at Microsoft Research, is accepted to NeurIPS. It is a single-trajectory verification framework: it periodically polls the reasoning trace and forks inference to recover intermediate states, running verifiers asynchronously alongside generation with negligible overhead on correct executions. To address verifier scarcity beyond math/code, it synthesizes verifiers automatically from natural-language policy documents.
Related event: Interwhen: single-trajectory verification framework accepted at NeurIPS(2 posts)→
More from Research
- Google launches AlphaGenome 2 to predict how DNA variants affect specific cell types — sundarpichai · 2026-11-12
- New learning rule boosts Hopfield network memory capacity over 10×, per Nature Communications study — bravo_abad · 2026-11-12
- Google team pushes neural cellular automata to multi-step visual reasoning — bratton · 2026-11-12
- NeoMME: 255x smaller multimodal embeddings while keeping >95% nDCG@10 — antoine_chaffin · 2026-11-12
- Study: Reflective Data Center Roofs Actually Cool Daytime Ground by 1.8-2.2°C — SumitGup · 2026-11-12
- Tsinghua's BaGuaLu-2 tames MoE expert-load spikes at runtime, accepted to SC26 — jiqizhixin · 2026-11-12