When to Think, When to Speak: Learning Disclosure Policies for LLM Reasoning

Jiaqi Wei, Xuehang Guo, Pengfei Yu, Xiang Zhang, Wanli Ouyang, Siqi Sun, Qingyun Wang, Chenyu You
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:133640-133659, 2026.

Abstract

In single-stream autoregressive interfaces, the same tokens both update the model state and constitute an irreversible public commitment. This coupling creates a silence tax: additional deliberation postpones the first task-relevant content, while naive early streaming risks premature commitments that bias subsequent generations. We introduce Side-by-Side (SxS) Interleaved Reasoning, which makes disclosure timing a controllable decision within standard autoregressive generation. SxS interleaves partial disclosures with continued private reasoning in the same context, but releases content only when it is supported by the reasoning so far. To learn such pacing without incentivizing filler, we construct entailment-aligned interleaved trajectories by matching answer prefixes to supporting reasoning prefixes, then train with SFT to acquire the dual-action semantics and RL to recover reasoning performance under the new format. Across two Qwen3 architectures/scales (MoE Qwen3-30B-A3B, dense Qwen3-4B) and both in-domain (AIME25) and out-of-domain (GPQA-Diamond) benchmarks, SxS improves accuracy–content-latency Pareto trade-offs under token-level proxies (e.g., inter-update waiting).

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-wei26i, title = {When to Think, When to Speak: Learning Disclosure Policies for {LLM} Reasoning}, author = {Wei, Jiaqi and Guo, Xuehang and Yu, Pengfei and Zhang, Xiang and Ouyang, Wanli and Sun, Siqi and Wang, Qingyun and You, Chenyu}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {133640--133659}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wei26i/wei26i.pdf}, url = {https://proceedings.mlr.press/v306/wei26i.html}, abstract = {In single-stream autoregressive interfaces, the same tokens both update the model state and constitute an irreversible public commitment. This coupling creates a silence tax: additional deliberation postpones the first task-relevant content, while naive early streaming risks premature commitments that bias subsequent generations. We introduce Side-by-Side (SxS) Interleaved Reasoning, which makes disclosure timing a controllable decision within standard autoregressive generation. SxS interleaves partial disclosures with continued private reasoning in the same context, but releases content only when it is supported by the reasoning so far. To learn such pacing without incentivizing filler, we construct entailment-aligned interleaved trajectories by matching answer prefixes to supporting reasoning prefixes, then train with SFT to acquire the dual-action semantics and RL to recover reasoning performance under the new format. Across two Qwen3 architectures/scales (MoE Qwen3-30B-A3B, dense Qwen3-4B) and both in-domain (AIME25) and out-of-domain (GPQA-Diamond) benchmarks, SxS improves accuracy–content-latency Pareto trade-offs under token-level proxies (e.g., inter-update waiting).} }
Endnote
%0 Conference Paper %T When to Think, When to Speak: Learning Disclosure Policies for LLM Reasoning %A Jiaqi Wei %A Xuehang Guo %A Pengfei Yu %A Xiang Zhang %A Wanli Ouyang %A Siqi Sun %A Qingyun Wang %A Chenyu You %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wei26i %I PMLR %P 133640--133659 %U https://proceedings.mlr.press/v306/wei26i.html %V 306 %X In single-stream autoregressive interfaces, the same tokens both update the model state and constitute an irreversible public commitment. This coupling creates a silence tax: additional deliberation postpones the first task-relevant content, while naive early streaming risks premature commitments that bias subsequent generations. We introduce Side-by-Side (SxS) Interleaved Reasoning, which makes disclosure timing a controllable decision within standard autoregressive generation. SxS interleaves partial disclosures with continued private reasoning in the same context, but releases content only when it is supported by the reasoning so far. To learn such pacing without incentivizing filler, we construct entailment-aligned interleaved trajectories by matching answer prefixes to supporting reasoning prefixes, then train with SFT to acquire the dual-action semantics and RL to recover reasoning performance under the new format. Across two Qwen3 architectures/scales (MoE Qwen3-30B-A3B, dense Qwen3-4B) and both in-domain (AIME25) and out-of-domain (GPQA-Diamond) benchmarks, SxS improves accuracy–content-latency Pareto trade-offs under token-level proxies (e.g., inter-update waiting).
APA
Wei, J., Guo, X., Yu, P., Zhang, X., Ouyang, W., Sun, S., Wang, Q. & You, C.. (2026). When to Think, When to Speak: Learning Disclosure Policies for LLM Reasoning. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:133640-133659 Available from https://proceedings.mlr.press/v306/wei26i.html.

Related Material