Clarification as Supervision lands NeurIPS Oral: denser training signal via model interaction
iatitov · x · 2026-10-02
The paper "Clarification as Supervision" is accepted as an Oral at NeurIPS 2026 (top 0.3%).
- Core insight: when a text-only reasoner answers from a VLM's caption and gets it wrong, the caption often omitted details the reasoner needed — not a reasoning failure.
- Approach: let models interact richly during training while pressuring them to reduce that interaction, yielding a denser learning signal and more effective one-shot collaboration.
More from Research
- Agility's Digit runs end-to-end autonomous whole-body manipulation for ~12 hours at IROS — chris_j_paxton · 2026-10-02
- Gumbel Straight Flow: distilling autoregressive models into one-step flow maps — sedielem · 2026-10-02
- Hypothesis: human learning is hill climbing — hard-to-verify tasks aren't relatively harder for AI — Afinetheorem · 2026-10-02
- Google unveils next-gen federated learning with TEE-based verifiable differential privacy — gaganghotra_ · 2026-10-02
- Beyond ChatGPT: Anima Anandkumar on making AI understand physics — nordicinst · 2026-10-02
- Quantum solver cracks drug discovery problem in 25 min; classical solver stalls 40% short after 3 hrs — MJBiercuk · 2026-10-02