Good Papers: read the good ones

When2Tool finds LLMs linearly encode tool necessity in hidden states, and Probe&Prefill uses this to cut unnecessary tool calls by 48% with minimal accuracy loss.

Chung-En Sun, Linbo Liu, Ge Yan, Zimo Wang and 1 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 2 on Hugging Face · Code ★ 16

– ReadersNo votes yet

18/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

88%Must read

?Must readVote to see the score

NeurIPS 2026U California, San Diego U California San Diego UC San Diego UCSD Diffusion models

Distance Marching for Generative Modeling

Distance Marching improves time-unconditional generative models via distance-focused losses and inference, surpassing flow matching FID with fewer steps and aiding OOD detection.

Zimo Wang, Ishit Mehta, Haolin Lu, Xunpeng Huang and 4 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

15/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 15 of 20 reviewers recommend it

lenient 3/5

medium 9/10

strict 3/5

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026UC San Diego U California, San Diego UC Berkeley Google Deepmind / Harvard UC San Diego, Amazon Video generation

Equilibrium Forcing: Adaptive Video Generation Without Noise Conditioning

Equilibrium Forcing removes noise conditioning from video diffusion to enable adaptive closed-loop inference that improves generation quality and consistency.

Hansen Lillemark, Alex Rojas, Zachary Novack, Runqian (Ray) Wang and 4 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 8 of 20 reviewers recommend it

lenient 2/5

medium 5/10

strict 1/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U California, San Diego UC San Diego U North Carolina at Chapel Hill Bytedance Seed; The University o UCSD Efficient inference & serving

SimSD: Simple Speculative Decoding in Diffusion Language Models

SimSD proposes a plug-and-play masking strategy that enables token-level speculative decoding in diffusion language models, achieving up to 7.46x faster throughput without training.

Junxia Cui, Haotian Ye, Runchu Tian, Hongcan Guo and 8 more

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

– ReadersNo votes yet

11/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 11 of 20 reviewers recommend it

lenient 4/5

medium 6/10

strict 1/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U California, San Diego Samsung Research America Duke UCSD Vision-language models

AMPS: Adaptive Modality Preference Steering via Functional Entropy

AMPS uses instance-aware functional entropy to adaptively steer multimodal model modality preferences, improving control while minimizing inference errors.

Zihan Huang, Xintong Li, Rohan Surana, Tong Yu and 4 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 10 of 20 reviewers recommend it

lenient 4/5

medium 6/10

strict 0/5

88%Must read

?Must readVote to see the score

NeurIPS 2026Meta / Ecole Normale Supérieure U California, San Diego INRIA - Ecole Normale Superieure Facebook AI Research Meta / FAIR RL for LLMs

Don’t Let Gains FADE: Breaking Down Policy Gradient Weights in RL

A framework decomposes RL advantage functions into gradient mass axes, showing trade-offs shift during training and motivating FADE, which adapts weights dynamically to accelerate convergence and improve accuracy-diversity trade-offs.

Juliette Decugis, Sean O&#x27;Brien, Francis Bach, Gabriel Synnaeve and 1 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

15/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 15 of 20 reviewers recommend it

lenient 4/5

medium 10/10

strict 1/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U California, San Diego Harvard University, Harvard Computer Science and Engineering UCSD Explainable AI

Measuring What Matters: Synthetic Benchmarks for Concept Bottleneck Models

Synthetic benchmarks for concept bottleneck models generate controlled labeled datasets to evaluate decision support and automation use cases, diagnose failure modes, and guide testing.

Julian Skirzynski, Harry Cheon, Shreyas Kadekodi, Meredith Stewart and 1 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

91%Must read

?Must readVote to see the score

NeurIPS 2026Yale Stanford U Chicago U Chicago Booth School of Busine Toyota Technological Institute a LLM reasoning & chain of thought

INFUSER: Influence-Guided Self-Evolution Improves Reasoning

INFUSER co-evolves a question generator and solver via influence-guided rewards, improving reasoning by over 20% on math benchmarks without curated data.

Siyu Chen, Miao Lu, Beining Wu, Heejune Sheen and 6 more

Atlanta Poster Session 4, Thu, Dec 10, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

17/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 17 of 20 reviewers recommend it

lenient 4/5

medium 10/10

strict 3/5

89%Must read

?Must readVote to see the score

NeurIPS 2026U California, Santa Barbara U California, San Diego UC Berkeley In-context learning & prompting

Learning to Follow In-Context Watermark Instructions via Self-Distillation

ICWBench reveals current LLMs fail at in-context watermarking, and self-distillation with reinforcement learning raises watermark detectability near perfect while preserving quality.

Yepeng Liu, Tianyi Chen, Xuandong Zhao, Dawn Song and 1 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

16/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 16 of 20 reviewers recommend it

lenient 5/5

medium 6/10

strict 5/5

83%Must read

?Must readVote to see the score

NeurIPS 2026U California, San Diego Massachusetts Institute of Techn UC San Diego Northwestern Mila; University of Montreal; MI Music

Live Music Diffusion Models: Efficient Fine-Tuning and Post-Training of Interactive Diffusion Music Generators

Live Music Diffusion Models modify diffusion inference with block-wise KV caching to surpass discrete autoregressive efficiency, enabling stable alignment via ARC-Forcing and real-time interactive generation on consumer hardware.

Zachary Novack, Stephen Brade, Haven Kim, Hugo Flores and 5 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 3 on Hugging Face · Code ★ 50

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 13 of 20 reviewers recommend it

lenient 4/5

medium 7/10

strict 2/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U California, San Diego / Meta U California, San Diego Image restoration

Single Shot HDR Recovery via a Video Diffusion Prior

By recasting single-shot HDR reconstruction as video diffusion of exposure brackets fused by a lightweight UNet, it achieves high input fidelity and outperforms generative baselines on benchmarks and in 72% of human comparisons.

Chinmay Talegaonkar, Jinshi He, Nicholas Antipa

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 10 of 20 reviewers recommend it

lenient 5/5

medium 5/10

strict 0/5

83%Must read

?Must readVote to see the score

NeurIPS 2026UC San Diego U California, San Diego Time series foundation models

Handwriting decoding as a challenging motor task for EEG Foundation Models

Handwriting decoding evaluates EEG foundation models, revealing confounded datasets and inferior performance versus smaller specialist models despite scaling.

Srinivas Ravishankar, Ishayu Ghosh, Nora Zajzon, Teng (Simon) Fei and 1 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 13 of 20 reviewers recommend it

lenient 5/5

medium 7/10

strict 1/5

70%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U California, San Diego U British Columbia The University of British Columb UCSD Learning theory

A second order regret bound for NormalHedge

A NormalHedge variant achieves second-order ε-quantile regret O(sqrt(V_T log(V_T/ε))) for easy sequences via self-concordance analysis.

Yoav S Freund, Nicholas Harvey, Victor S. Portella, Yabing Qi and 1 more

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

5/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 5 of 20 reviewers recommend it

lenient 1/5

medium 2/10

strict 2/5

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U California, San Diego Amazon Amazon Store Foundational AI / S Meta Superintelligence Labs Texas A&M University - Colle RL for LLMs

Stepwise Penalization for Length-Efficient Chain-of-Thought Reasoning

SWAP assigns step-level length penalties based on reasoning contribution, cutting chain-of-thought length by 64.3% and boosting accuracy 5.7% over base models.

Xintong Li, Sha Li, Rongmei Lin, Hongye Jin and 9 more

Atlanta Poster Session 3, Thu, Dec 10, 10:00 AM–1:00 PM, Hall C1 · Published 2026

– ReadersNo votes yet

9/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 9 of 20 reviewers recommend it

lenient 5/5

medium 4/10

strict 0/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Northwestern Illinois Institute of Technology U Michigan - Ann Arbor U California, Los Angeles U California, San Diego Long-context modeling

Attention Sinks and Outliers in Attention Residuals

OASIS stabilizes dual-normalized attention-residual architectures via null routing and token-to-depth null coupling, reducing activation outliers by 81.75% and improving low-bit quantized reasoning by 42.11%.

Haozheng Luo, Haoran Dai, Shaoyang Zhang, Xi Chen and 9 more

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 1 on Hugging Face · Code ★ 3

– ReadersNo votes yet

11/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

80%Must read

?Must readVote to see the score

NeurIPS 2026UC San Diego U California, San Diego Abiomed UC San Diego, Amazon Offline RL

Generative OOD-regularized Model-based Policy Optimization

GORMPO integrates generative density estimation into model-based offline RL to restrict policy updates to high-density dataset regions, improving performance by 17% on medical data while linking OOD detection quality to policy gains under stable dynamics.

Aysin Tumay, Jiahe Huang, Elise Jortberg, Rose Yu

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 12 of 20 reviewers recommend it

lenient 5/5

medium 7/10

strict 0/5

83%Must read

?Must readVote to see the score

NeurIPS 2026UCSD UC San Diego U California, San Diego Behavior & reasoning editing

Steer2Edit: From Activation Steering to Component-Level Editing

Steer2Edit converts inference-time activation steering into training-free, component-level rank-1 weight edits that improve safety, truthfulness, and reasoning efficiency over global interventions.

Chung-En Sun, Ge Yan, Zimo Wang, Lily Weng

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 1 on Hugging Face

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 13 of 20 reviewers recommend it

lenient 5/5

medium 8/10

strict 0/5

91%Must read

?Must readVote to see the score

NeurIPS 2026U California San Diego U California, San Diego UC San Diego West Virginia UCSD Efficient inference & serving

AgentKVShift: Efficient KV Cache Reuse for Agentic Memory Systems

AgentKVShift uses probe-guided KV residual correction to reuse agentic memory caches with near-full accuracy at 10-30% recompute, yielding 2-3.5x prefill speedups.

Nilesh Pandey, Jason Kong, Lanxiang Hu, Quanling Zhao and 4 more

Atlanta Poster Session 1, Wed, Dec 9, 10:00 AM–1:00 PM, Hall C1 · Published 2026

– ReadersNo votes yet

17/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 17 of 20 reviewers recommend it

lenient 5/5

medium 9/10

strict 3/5

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026NYU U California, San Diego U Munich, Ludwig-Maximilians-Uni New York AI policy & governance

On the Meta-Design of Allocation Problems

The paper defines meta-design for resource allocation by optimizing upstream design parameters like data, capacity, and quality, and demonstrates the framework in German employment and Ethiopian cash transfer programs.

Unai Fischer Abaigar, Emily Aiken, Christoph Kern, Juan C Perdomo

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 8 of 20 reviewers recommend it

lenient 4/5

medium 4/10

strict 0/5

89%Must read

?Must readVote to see the score

NeurIPS 2026SpotlightGoogle DeepMind U California, San Diego U Waterloo Stanford; Noeri AI Tencent LLM agents & planning

Beyond Semantic Similarity: Rethinking Retrieval for Agentic Search via Direct Corpus Interaction

Direct corpus interaction uses terminal tools to search raw corpora directly, bypassing fixed retrieval interfaces and substantially outperforming sparse, dense, and reranking baselines on agentic search benchmarks.

Zhuofeng Li, Haoxiang Zhang, Cong Wei, Pan Lu and 14 more

Atlanta Poster Session 3, Thu, Dec 10, 10:00 AM–1:00 PM, Hall C1 · Published 2026 · ▲ 125 on Hugging Face · Code ★ 408

– ReadersNo votes yet

16/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 16 of 20 reviewers recommend it

lenient 5/5

medium 9/10

strict 2/5

Show 20 more papers