Read the good ones.

You decide which ones are good research papers.

NewNeurIPS 2026Orals Spotlights PostersWhat's worth reading in RL this week?Most debated papers this month Who's working on agent memory?What's new in diffusion models?

Paper of the day

92%Must readScience or Slop?: Benchmarking and Mitigating Scientific Slop in AI-Generated PapersThe benchmark links scientific slop to ICLR acceptance across years and delivers a harness-level fix that cuts residual AI reasoning gaps, yet its 390-paper set pairs by contribution rather than causation and lacks human baseline verification.

Trending at NeurIPS 2026

Orals, spotlights and posters people are talking about

All sessions

78%Program-as-Weights: A Programming Paradigm for Fuzzy FunctionsProgram-as-Weights compiles natural-language specs into compact local neural adapters that match large-model prompting with far less memory and faster offline execution.NeurIPS 2026Highly rated · ▲ 307 on HF · 6 comments

91%Rethinking Long-Video Efficiency: A Joint Allocation Perspective on Frames, Pixels, and Front-End LatencyLoHi trades per-frame resolution for denser temporal sampling via low-resolution streams plus sparse high-resolution frames, boosting long-video accuracy up to 10.6 points and cutting front-end latency up to 7x.NeurIPS 2026Must read · ▲ 13 on HF · 18 comments

86%Recursive Multi-Agent SystemsRecursiveMAS scales multi-agent collaboration through recursive latent-space computation via RecursiveLink and inner-outer loop co-optimization, improving accuracy by 8.3% with 1.2-2.4x speedup and 34.6%-75.6% token reduction over baselines.NeurIPS 2026Must read · ▲ 239 on HF · 6 comments

88%PaLoRA: Paced Low-Rank Adaptation for Continual LearningPaLoRA derives an optimal rank-aware pacing law for LoRA continual learning that adaptively restricts gradient scaling to prevent forgetting, improving long-horizon benchmark accuracy by 4%.NeurIPS 2026Must read · ▲ 9 on HF · 17 comments

78%Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before AnsweringImagine3D-LLM learns compact scene representations via learnable summary tokens decoded into 3D Gaussian splatting with photometric loss, improving multi-view spatial reasoning benchmarks.NeurIPS 2026Highly rated · ▲ 72 on HF · 14 comments

76%Learning from the Self-future: On-policy Self-distillation for dLLMsd-OPSD applies on-policy self-distillation to diffusion LLMs via suffix conditioning and step-level supervision, cutting optimization steps by ~90% versus RLVR while outperforming baselines on reasoning benchmarks.NeurIPS 2026Highly rated · ▲ 174 on HF · 4 comments

88%MulTaBench: Benchmarking Multimodal Tabular Learning with Text and ImageMulTaBench benchmarks 40 multimodal tabular datasets and shows target-aware tuning of text and image embeddings improves predictive performance over frozen embeddings.NeurIPS 2026 SpotlightMust read · ▲ 142 on HF · 4 comments

76%Self-Distilled RLVRRLSD combines RLVR and self-distillation, using token-level policy differences for update magnitudes and environmental feedback for directions, improving convergence and stability.NeurIPS 2026 SpotlightHighly rated · ▲ 145 on HF · 4 comments

88%AgentGrad: Intervention-guided Prompt Optimization for Multi Agent SystemsAgentGrad improves multi-agent prompt optimization via sequential intervention and semantic gradient clustering, achieving state-of-the-art performance with 2.5x faster optimization and 21.8% lower cost.NeurIPS 2026Must read · ▲ 107 on HF · 5 comments

90%A Matter of TASTE: Improving Coverage and Difficulty of Agent BenchmarksTASTE reverses benchmark construction by evolving tool sequences to automatically generate harder, broader-coverage agent tasks that expose severe performance drops and saturation in existing benchmarks.NeurIPS 2026Must read · 100% of 1 readers upvoted · ▲ 74 on HF · 5 comments

91%CollabVR: Collaborative Video Reasoning with Vision-Language and Video Generation ModelsCollabVR pairs vision-language models with video generation models in closed-loop step-level planning and verification, reducing drift and simulation errors for major video reasoning gains.NeurIPS 2026Must read · 100% of 1 readers upvoted · ▲ 71 on HF · 5 comments

80%DynamicVLA: A Vision-Language-Action Model for Dynamic Object ManipulationDynamicVLA combines efficient inference with continuous action streaming to fix latency-induced mismatches, improving dynamic manipulation success across synthetic and real-world benchmarks.NeurIPS 2026Must read · ▲ 77 on HF · 6 comments

91%?MiroEval: Benchmarking Multimodal Deep Research Agents in Process and OutcomeMiroEval benchmarks multimodal deep research agents via process and outcome evaluation across 100 real-world tasks, finding process quality predicts outcomes and multimodal tasks reduce scores by 3, 10 points.NeurIPS 2026Must read · vote to see score · ▲ 69 on HF · 6 comments

82%?Optimal Transport Reweighting for Robust Learning under Spurious Correlations and Label NoisePOTER uses optimal transport geometry between training and reference distributions to downweight mislabeled or shortcut-aligned samples, achieving state-of-the-art worst-group accuracy with a single training stage.NeurIPS 2026Must read · 100% of 1 readers upvoted · vote to see score · 8 comments

86%?Code2World: A GUI World Model via Renderable Code GenerationCode2World uses renderable code generation for GUI world modeling, achieving top next-UI prediction and boosting Android navigation success by up to 9.5%.NeurIPS 2026Must read · vote to see score · ▲ 186 on HF

76%?Heterogeneous Agent Collaborative Reinforcement LearningHACRL enables heterogeneous agents to share verified rollouts during collaborative on-policy training and execute independently at inference, with HACPO improving all agents by 3.6% over baselines at half the rollout cost.NeurIPS 2026Highly rated · vote to see score · ▲ 110 on HF · 4 comments

Representation & contrastive learning73%?Structure-agnostic Causal Representation LearningSaCRL jointly identifies causal structure and learns invariant representations via soft optimization over HSIC-based invariance violations without prior structural knowledge. It guarantees structure identification, invariance satisfaction, and out-of-distribution generalization while achieving stateNeurIPS 2026Highly rated · 0% of 1 readers upvoted · vote to see score · 8 comments

91%?Claw-Eval: Towards Trustworthy Evaluation of Autonomous AgentsClaw-Eval introduces a trajectory-aware benchmark with 300 tasks, finding opaque grading misses 44% of safety violations and agent rankings vary across multi-dimensional capabilities.NeurIPS 2026Must read · vote to see score · ▲ 117 on HF · 3 comments

90%?Modeling quantum neural network gradient with reinforcement learningRLQ-Grad uses reinforcement learning to propose quantum neural network updates without differentiating circuits, avoiding barren plateaus and scaling with parameters rather than Hilbert space dimension to achieve orders-of-magnitude faster training and higher accuracy.NeurIPS 2026Must read · 100% of 1 readers upvoted · vote to see score · 6 comments

78%?Tuna-2: Pixel Embeddings Beat Vision Encoders for Multimodal Understanding and GenerationTuna-2 replaces vision encoders with patch embeddings for end-to-end pixel-space multimodal understanding and generation, achieving state-of-the-art results that outperform encoder-based designs at scale.NeurIPS 2026Highly rated · vote to see score · ▲ 70 on HF · 4 comments

89%?Principia: Relational Physics Tests for Video ModelsPrincipia benchmarks video generators via calibration-independent relational physics consistency across eight Newtonian phenomena, finding top models score below 0.42 despite high VBench ratings.NeurIPS 2026Must read · vote to see score · ▲ 18 on HF · 6 comments

91%?Jointly Reinforcing Diversity and Quality in Language Model GenerationsDARLING uses a learned partition function to jointly optimize language model response quality and semantic diversity via reinforcement learning, improving both quality and novelty across creative and math benchmarks.NeurIPS 2026Must read · vote to see score · ▲ 25 on HF · 5 comments

89%?Beyond Semantic Similarity: Rethinking Retrieval for Agentic Search via Direct Corpus InteractionDirect 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.NeurIPS 2026 SpotlightMust read · vote to see score · ▲ 125 on HF

83%?TransitLM: A Large-Scale Dataset and Benchmark for Map-Free Transit Route GenerationTransitLM provides 13 million transit records to train LLMs for map-free route generation, yielding accurate routes that implicitly ground GPS coordinates without maps.NeurIPS 2026Must read · vote to see score · ▲ 127 on HF

Must read

This year's highest-rated papers

See all

92%Science or Slop?: Benchmarking and Mitigating Scientific Slop in AI-Generated PapersA benchmark of 390 AI papers measures scientific slop across structure, argument, and artifacts; a harness reduces the AI-human gap by 63% via evidence-grounded revision.Must read · 100% of 1 readers upvoted · ▲ 55 on HF · 6 comments

91%MotorMind: Scaffolding General Vision Language Models for Zero-Shot Robot ManipulationMotorMind connects general vision-language models to deterministic robot control via mid-level actions and feedback loops, achieving 66.7% zero-shot success on LIBERO-PRO and 95% on real robots without task-specific training or external tools.Must read · ▲ 90 on HF · 6 comments

91%DEPICT: Scoring Text-to-Image Alignment by Answer AgreementDEPICT scores text-to-image alignment via expected agreement between image and caption answers, boosting negation accuracy to 88% and exceeding fine-tuned evaluators on human correlation benchmarks.Must read · ▲ 3 on HF · 6 comments

91%Dataset Identity, Not Novelty: The Source of an Inflated OOD Detection GainPost-hoc OOD detectors inflate reported gains by fitting dataset identity rather than novelty, leaving only a single constant fit valid.Must read · 3 comments

91%OSWorld-Pro: Process-based Evaluation for Computer Use AgentsOSWorld-Pro introduces process-based evaluation with 2,800 subgoals across 300 tasks, revealing top models achieve only 75.7% subgoal success versus 83.4% end-state performance and identifying distinct failure modes like irrelevant actions and click errors.Must read · 100% of 1 readers upvoted · ▲ 15 on HF · 6 comments

91%Does Learning Protein Folding Generalize to Broader Reasoning?Post-training on protein-folding data via discrete answers and continuous geometry improves structure prediction and broad reasoning across ten benchmarks.Must read · 100% of 1 readers upvoted · ▲ 120 on HF · 5 comments

91%HuatuoGPT-3: RL-Only Domain Adaptation from Base ModelsOnePO enables RL-only medical domain adaptation via adaptive objective evolution and teacher retirement, yielding HuatuoGPT-3 that surpasses frontier models.ICML 2026Must read · 100% of 1 readers upvoted · ▲ 15 on HF · 4 comments

91%Learning Functional Subspaces for Neural Network CompressionLSP learns low-rank subspaces end-to-end via joint orthogonal projector optimization to reduce transformer memory and compute while outperforming local criteria at high compression ratios.Must read · ▲ 1 on HF · 4 comments

91%CurveCodec 2: Skeleton-agnostic animation compression with a learned entropy modelCurveCodec 2 predicts quantized skeletal curves from past values and learns residual entropy models to reduce animation storage to 0.22-0.37x of ACL with verified error bounds.Must read · ▲ 4 on HF · 16 comments

91%Where Does Retrieval-Based Open-Ended Evaluation Fail? Automatic Taxonomy Induction from Long-Form Medical Answer Factuality VerificationAutomatic taxonomies reveal retrieval and verifier reasoning failures persist across scaled medical retrieve-then-verify systems, showing fundamental open-ended evaluation limits.Must read · ▲ 12 on HF · 4 comments

LLM agents & planning91%StateM: Reaching 95.3% Raw Accuracy, or a \$15 Frontier Run, on Terminal-Bench 2.1 via Harness ScalingStateM improves long-horizon agent accuracy via durable-state harness scaling without model changes, reaching 95.3% on Terminal-Bench 2.1 and cutting API costs to about $15 versus $574.68.Must read · ▲ 452 on HF · 11 comments

91%False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search AgentsSelf-evolving search agents suffer co-cheating where proposers and solvers mutually reinforce errors; CrossFit partitions sources to cross-fit agreement and cuts false agreement by over half, boosting downstream search by 8+ points.Must read · ▲ 670 on HF · 5 comments

91%?UndoBench: Separating Task Competence from Recovery Capability in Tool-Using AI AgentsUndoBench separates tool-using agent competence from fault recovery via paired enterprise workflow trials, finding 83.54% nominal success but only 46.72% recovery success with phase-dependent vulnerabilities.Must read · vote to see score · ▲ 11 on HF · 16 comments

91%?Sharpening Tax in Post-TrainingPost-training sharpens base model behaviors at the cost of solution coverage, introducing a quantifiable "Sharpening Tax"; a posterior-tempered group sampler reduces this tax while boosting accuracy.Must read · vote to see score · ▲ 102 on HF · 4 comments

91%?QuantWM: Temporally Consistent 2-Bit KV Cache Quantization for Video World ModelsQuantWM is a training-free 2-bit KV cache quantization framework for video world models that preserves attention logits and token selection to eliminate temporal flickering while achieving up to 6.20x memory compression.Must read · vote to see score · ▲ 13 on HF · 3 comments

91%?DeskForge: Dense Supervision from Desktop Environments for Computer-Use AgentsDeskForge generates dense desktop supervision via controllable real-app environments, yielding 1.2M observations that improve GUI grounding and long-horizon computer-use task completion.Must read · vote to see score · ▲ 10 on HF · 5 comments

91%?OpenART: Scaling Agent Red Teaming via Open-Ended Environment EvolutionOpenART scales agent red teaming via open-ended environment evolution across 10,000 stateful scenarios, with EMHA achieving 85% attack success that grows with complexity.Must read · vote to see score · ▲ 266 on HF · 6 comments

91%?Can We Defend Against AI-Generated Video Attacks on Real-World Crisis Events? A Systematic Evaluation of Detectors, Generators and Social DisseminationThe RA-Bench benchmark reveals current detectors fail to consistently identify AI-generated crisis videos, which become harder to detect after social dissemination and frequently mislead humans.Must read · vote to see score · ▲ 287 on HF · 15 comments

90%?World Action Learning via Interaction-Centric Spectral Latent GuidanceWING distills interaction-centric latent actions from egocentric video and uses cross-embodiment spectral low-frequency guidance to transfer them to robot policies, achieving high success rates on LIBERO, RoboTwin, RoboCasa, and real-world tasks.Must read · 100% of 1 readers upvoted · vote to see score · ▲ 20 on HF · 4 comments

90%?World Models' Last Exam in PhysicsWorld Models' Last Exam in Physics benchmarks video models via 40 measurement-based physics tasks, finding the best model scores 57.76/100 with widespread inconsistencies.Must read · 100% of 1 readers upvoted · vote to see score · ▲ 2 on HF · 4 comments

90%?From Evidence to Action: How Tool-Using Agents FailTool-using agents often fail by acting before establishing required evidence or leaving multi-action workflow prerequisites unresolved, despite accurate static action assessment. SafeActBench reveals failures stem from how agents use established evidence during execution, not just missing informatioMust read · 100% of 1 readers upvoted · vote to see score · ▲ 20 on HF · 3 comments

89%?Self-Generated Feedback Destabilizes Test-Time Training: A Causal Decomposition of Long-Horizon AdaptationSelf-generated feedback in long-horizon test-time training causes weight updates that improve synthetic text but degrade real-text prediction, and settlement on independent evidence prevents this failure.Must read · vote to see score · ▲ 19 on HF · 4 comments

89%?Rethinking Token Reweighting for SFT: Suppress, Reverse, and Extrapolate Learned FeaturesExisting token-reweighting methods cannot reverse harmful SFT features; SCALE uses frozen SFT deltas with entropy-guided gates to suppress, reverse, or extrapolate them, improving math and code results.Must read · vote to see score · ▲ 11 on HF · 6 comments

Most debated

Where the reviewers can't agree

See all

86%Thinking with Video: Video Generation as a Promising Multimodal Reasoning ParadigmVideo generation enables unified multimodal reasoning via Sora-2, which matches vision-language models and exceeds GPT-5 on spatial tasks while scoring 92% on MATH.Must read · ▲ 242 on HF · 22 comments

83%The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy DistillationRIDE extrapolates RL-induced representation residuals for stable on-policy distillation, surpassing output-space methods and matching or exceeding RL teachers.Must read · ▲ 577 on HF · 19 comments

86%Adaptive Fused Prior Transfer for Controllable Generative Image CompressionAFP-GIC transfers adaptive fused priors from a frozen pretrained model to guide generative compression without transmitting them, reducing decoder latency and parameters while improving very-low-bitrate naturalness.Must read · ▲ 4 on HF · 18 comments

76%Adapting prior-data fitted networks for tabular anomaly detectionFrozen and fine-tuned TabPFN representations for tabular anomaly detection yield ZEN and FOCUS, surpassing all ADBench baselines in AUROC despite unsupervised deployment and contaminated reference sets.ICLR 2027Highly rated · ▲ 5 on HF · 18 comments

89%Sharpen Without Search: On-Policy Distillation of Sequence-Level Power DistributionOn-policy power distillation trains models to generate sharpened answers directly, improving single-sample math reasoning by up to 27.3 points and outperforming multi-candidate sampling and reward-based methods.Must read · ▲ 4 on HF · 18 comments

89%Certification of Real Images through Calibrated Content AuthenticationDeepfake detectors degrade to 76% accuracy and near-zero under attacks, so calibrated reconstruction-based authentication bounds false real-image certification to 1%.Must read · ▲ 13 on HF · 18 comments

83%PerturBot: Breaking Shortcut Priors in Vision-Language-Action Models with Perturbative TrainingPerturbBot breaks vision-language-action shortcut priors via perturbative training while GroundingFscore diagnoses shortcut reliance, enabling healthier scaling without altering inference.Must read · ▲ 13 on HF · 17 comments

67%Collaborative Personalized Preference Alignment for LLMs under Data DeficiencyAPO learns shared LLM aligner initializations via grouped gradient coordination to enable few-shot personalization under heterogeneous preferences and scarce feedback, improving over baselines with 20 local examples.Highly rated · 0% of 1 readers upvoted · ▲ 5 on HF · 17 comments

71%Code2Games: Enabling Coding Agents for Gaming World GenerationCode2Games coordinates scene analysis and gameplay planning via shared representations to generate consistent gaming worlds and adapt them to Unreal Engine 5. The framework improves visual quality, interactive fidelity, and playable-game quality over direct coding-agent generation on the GameCode4D Highly rated · ▲ 6 on HF · 17 comments

71%RealtimeWAM: One-Step Asynchronous World Action ModelsRealtimeWAM uses teacher-anchored consistency distillation and cross-expert wavefront pipelining for one-step asynchronous action generation, achieving near-lossless performance with ~25x speedup.Highly rated · 0% of 1 readers upvoted · ▲ 12 on HF · 17 comments

80%?Learning Latent Protein Languages for Autoregressive GenerationLearned latent protein languages improve autoregressive generation scaling, speed, and quality versus amino-acid and coordinate token models.Must read · vote to see score · ▲ 5 on HF · 17 comments

86%?Training Numerical Intelligence via Auto-Diagnosis and Skill DiscoveryADSD links numerical diagnosis to reusable solver self-improvement, reducing mean solver error by nearly 71x across four challenging numerical domains.Must read · vote to see score · ▲ 8 on HF · 16 comments

86%?Learning Steadily: Accumulating Relative Point Margin Scores for Face Image Quality AssessmentCARPM-FIQA accumulates relative point margin scores across training epochs to stabilize face image quality estimates, reducing variance and improving ranking stability near top performance.Must read · 100% of 1 readers upvoted · vote to see score · ▲ 6 on HF · 16 comments

88%?FairRSFM: A Biome-Aware Benchmark and Debiasing Framework for Remote Sensing Foundation ModelsFairRSFM benchmarks remote sensing foundation models by biome to expose hidden ecological performance disparities and tests debiasing methods without backbone updates. Aggregate metrics consistently mask large biome-dependent gaps, though mitigation effectiveness varies by model and task.Must read · vote to see score · ▲ 5 on HF · 16 comments

78%?Agentic discovery of blood biomarker from distilled private health recordsDistilling private health records into a released scoring tool enables agentic discovery of CBC biomarkers that improve diagnostic AUC over literature baselines without exposing patient data.Highly rated · vote to see score · ▲ 4 on HF · 16 comments

76%?SEER: Self-Evolving Event Reasoning and Retrieval for Time Series ForecastingSEER uses self-evolving event reasoning and retrieval to dynamically optimize forecasting with exogenous events via reflective memory and causal knowledge, outperforming state-of-the-art baselines.Highly rated · vote to see score · ▲ 13 on HF · 16 comments

88%?Periscope: Extending Frozen Language Models Beyond Their Context WindowPeriscope arranges text chunks in a grid to build an evidence map via local and strided probes, letting frozen language models answer questions across multi-million-token contexts with sublinear cost and small GPU memory.Must read · 100% of 1 readers upvoted · vote to see score · ▲ 8 on HF · 16 comments

76%?Empirical Variational AutoencoderEmpirical Variational Autoencoder learns autoregressive latent priors empirically via one linear layer to close the VAE prior-posterior gap, yielding high-fidelity sequential generation competitive with diffusion models at much faster inference.Highly rated · 100% of 1 readers upvoted · vote to see score · ▲ 7 on HF · 15 comments

89%?HLA-WM: Hybrid Linear Attention for Long-Horizon Video World ModelsHLA-WM combines geometry-guided retrieval with recurrent linear attention to fix long-range forgetting in video world models, improving 60-second consistency metrics by up to 28.5% with 12× lower memory and no retraining.Must read · vote to see score · ▲ 6 on HF · 15 comments

88%?UnAct: Gradient-Free Unlearning via Targeted Activation InterventionUnAct uses gradient-free targeted activation interventions to unlearn model classes from few forget images without gradients, labels, or retained data, matching or exceeding SSD and LFSSD accuracy across datasets and preventing network collapse with scarce data.Must read · vote to see score · ▲ 4 on HF · 15 comments

All papers

How scores work

78%Highly rated

?Highly ratedVote to see the score

PreprintU Zurich U Palermo Universitat Pompeu Fabra Institut de Biologia Evolutiva Barcelona School of Economics Genomics & single-cell

Agta hunter-gatherer oral microbiomes are shaped by contact network structure

Agta hunter-gatherer oral microbiomes resemble Central African foragers more than neighbors, with contact networks predicting bacterial transmission and central individuals as supersharers.

Federico Musciotto, Begoña Dobón, Michael John Greenacre, Álex Mira and 12 more

Published Dec 31, 2030 · 0 citations

100% Readers1 of 1 upvoted

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.

71%Highly rated

?Highly ratedVote to see the score

arXivRetrieval-augmented generation

UNREAL: Unifying Retrieval and Long-Context with a Single Model

UNREAL unifies retrieval and long-context evidence selection via frozen LLM representations with minimal parameters, outperforming state-of-the-art retrievers and improving long-context accuracy substantially.

Edan Kinderman, Elad Hoffer, Yochai Blau, Brian Chmiel and 3 more

Published Oct 6, 2026 · ▲ 15 on Hugging Face

0% Readers0 of 1 upvoted

14/20 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: 14 of 20 reviewers recommend it

lenient 4/5

medium 7/10

strict 3/5

78%Highly rated

?Highly ratedVote to see the score

arXivUnified multimodal models

Sensor-Language-Action Models

Sensor-Language-Action modeling unifies multimodal sensors, language, and actions via a semantic interface, and OpenSLA achieves superior hierarchical prediction and explanation with zero-shot generalization.

Yuekai Xu, Zitao Shuai, Yuzhe Yang

Published Oct 6, 2026 · ▲ 1 on Hugging Face · Code ★ 1

– 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 5/5

medium 5/10

strict 1/5

90%Must read

?Must readVote to see the score

arXivVideo generation

World Models' Last Exam in Physics

World Models' Last Exam in Physics benchmarks video models via 40 measurement-based physics tasks, finding the best model scores 57.76/100 with widespread inconsistencies.

Mingju Gao, Qingle Liu, Yuzhao Peng, Xinjie Lin and 7 more

Published Oct 6, 2026 · ▲ 2 on Hugging Face

100% Readers1 of 1 upvoted

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.

78%Highly rated

?Highly ratedVote to see the score

arXivEmbodied agents

VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning

VeriFine co-evolves policies, curricula, and judges via adaptive diagnosis and coactive calibration to sustain embodied reasoning self-improvement. Experiments on driving and navigation show continuous gains in both policy and judge capability.

Zewei Zhou, Rachel Luo, Yulong Cao, Chaowei Xiao and 9 more

Published Oct 6, 2026 · ▲ 4 on Hugging Face

100% Readers1 of 1 upvoted

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.

88%Must read

?Must readVote to see the score

arXivControllable generation

CtrlCache: Accelerating Interactive Video World Models with Control-Aware Caching

CtrlCache accelerates interactive video world models via control-aware caching that detects action changes to reuse transformer residuals and apply frequency-mixed history guidance, achieving up to 1.41x speedups with improved quality.

Shangye Song, Dong Gong, Hong Jia, Yun Sing Koh and 1 more

Published Oct 6, 2026 · ▲ 1 on Hugging Face · Code

100% Readers1 of 1 upvoted

14/20 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.

84%Must read

?Must readVote to see the score

arXivAgent benchmarks & environments

CheckerBench: Can Long-Horizon Agents Synthesize Static-Analysis Checkers?

CheckerBench evaluates long-horizon agents on synthesizing static-analysis checkers across 300 CVE-derived tasks, finding best Pass@1 reaches 45.33%.

Hang He, Li Wang, Hao Chen, Yuchen Shao and 8 more

Published Oct 6, 2026 · ▲ 52 on Hugging Face · Code

100% Readers1 of 1 upvoted

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 5/10

strict 2/5

82%Must read

?Must readVote to see the score

arXivRL for LLMs

Sherpa: Teaching LLMs to Teach Adaptively

Sherpa uses multi-turn reinforcement learning to train LLM teachers that adapt instructions to diverse student archetypes, improving student performance by 20.5 points and pedagogy scores to 79.2%.

Weixian Xu, Yanzhe Zhang, Zora Zhiruo Wang, Changyu Chen and 1 more

Published Oct 6, 2026 · ▲ 3 on Hugging Face · Code

100% Readers1 of 1 upvoted

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.

84%Must read

?Must readVote to see the score

arXivRL for LLMs

TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models

TRACE aligns FP4 quantization between RL training and rollout paths via rollout-guided quantization-aware training for MoE language models, achieving BF16-comparable RL performance with up to 5.4x rollout speedup.

Xin Wang, Hao Yu, Zhengyang Zhuge, Bochao Mao and 8 more

Published Oct 6, 2026 · ▲ 15 on Hugging Face

100% Readers1 of 1 upvoted

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 6/10

strict 1/5

90%Must read

?Must readVote to see the score

arXivTool use & function calling

From Evidence to Action: How Tool-Using Agents Fail

Tool-using agents often fail by acting before establishing required evidence or leaving multi-action workflow prerequisites unresolved, despite accurate static action assessment. SafeActBench reveals failures stem from how agents use established evidence during execution, not just missing informatio

Hongzhan Lin, Shidong Cao, Ziyang Luo, Wenhao Chai and 2 more

Published Oct 6, 2026 · ▲ 20 on Hugging Face · Code ★ 3

100% Readers1 of 1 upvoted

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 5/5

medium 7/10

strict 3/5

86%Must read

arXivAlignment & preference optimization

Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability

Cross-tokenizer on-policy distillation achieves comparable accuracy with strict top-16 shared-vocabulary supervision versus full coverage, while expanded span supervision reduces accuracy due to conflicting gradients, motivating prioritization of supervision reliability over alignment coverage.

Bingxi Hou, Guochao Jiang, Guofeng Quan, Weiqing Li and 3 more

Published Oct 6, 2026 · ▲ 40 on Hugging Face

100% Readers1 of 1 upvoted

13/20 AI panelreviewers recommend it

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

89%Must read

?Must readVote to see the score

arXivAgent memory

DAEDALUS: Bootstrapping Agent Memory from Self-Generated Tasks

DAEDALUS bootstraps reusable agent memory from self-generated practice tasks without oracles, improving success rates by up to 15.9 points across benchmarks.

Antoine Edy, Max Conti, Victor Xing, Marc-Antoine Allard and 2 more

Published Oct 6, 2026 · ▲ 5 on Hugging Face · Code

– 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 10/10

strict 1/5

78%Highly rated

?Highly ratedVote to see the score

arXiv3D generation

Building Rome from a Single Image

A redesigned object-centric generator partitions scenes into distance-adaptive chunks, captures 2D-3D correspondence, and trains on 4,000 outdoor scenes to outperform baselines in indoor and outdoor mesh generation.

Jiraphon Yenphraphai, Fang Li, Tianshuo Xu, Depu Meng and 4 more

Published Oct 6, 2026 · ▲ 4 on Hugging Face

– 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.

71%Highly rated

?Highly ratedVote to see the score

arXivTest-time compute

Towards In-Parameter Memory Augmentation for Large Language Models

This survey organizes in-parameter memory augmentation for LLMs by parameter placement and acquisition time to enable reusable parametric knowledge at deployment.

Haoyu Huang, Zhongwei Xie, Jiaxin Bai, Yisen Gao and 5 more

Published Oct 6, 2026 · ▲ 5 on Hugging Face · Code ★ 1

– ReadersNo votes yet

6/20 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.

78%Highly rated

?Highly ratedVote to see the score

arXivEmbodied agents

Attacca: Goal-Directed Control under State Continuity for Long-Horizon Embodied Agents

Attacca trains visual goal-conditioned policies on complete search-to-interact trajectories with decoupled goal images and behavioral-phase conditioning to improve long-horizon embodied task success by up to 7x.

Gyusik Seo, Jaehong Yoon

Published Oct 6, 2026 · ▲ 3 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.

AI panel: 11 of 20 reviewers recommend it

lenient 4/5

medium 6/10

strict 1/5

83%Must read

?Must readVote to see the score

arXivRecommender systems

Personal-Agent Mediated Recommendation with Cross-Platform User History

Personal-agent mediated recommendation balances cross-platform user history against platform rankings via the MediateRec benchmark and PAMO optimization to improve rescue-harm trade-offs.

Yu Xia, Jiangfan Zhang, Jun Xiao, Julian McAuley and 1 more

Published Oct 6, 2026 · ▲ 3 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 7/10

strict 1/5

78%Highly rated

?Highly ratedVote to see the score

arXivAdversarial robustness

AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model

AdvSim2Real co-evolves tasks, adversarial injections, and a web agent in a simulator, boosting 4B agent completion by 33.6% against unseen adaptive attacks and transferring gains to real browsers.

Sarim Hashmi, Mukul Ranjan, Kshitij Mishra, Mikhail Kuznetsov and 2 more

Published Oct 6, 2026 · ▲ 3 on Hugging Face · Code

– 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

arXivRL for LLMs

NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale

NeMo-DCR bit-exactly refits trillion-parameter policies by streaming delta-compressed weight changes via affine mappings and XOR masks, cutting 1T cross-region refits from 87.5 minutes to 150 seconds.

Songlin Jiang, Zhiyu Li, Terry Kong, Yu Yao and 4 more

Published Oct 6, 2026 · ▲ 6 on Hugging Face · Code ★ 2,048

– 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 4/5

medium 6/10

strict 2/5

76%Highly rated

?Highly ratedVote to see the score

arXivRobot manipulation

EmbodiedSmith: Scaling Embodied Data through Recursive Self-Improvement Flywheel in Simulation

EmbodiedSmith unifies asset, scene, and task generation in a recursive self-improvement loop to scale embodied simulation data, improving generation success and robot policy generalization across diverse embodiments and physics.

Yikai Qin, Yifei Deng, Mingjian Liang, Wenxuan Song and 12 more

Published Oct 6, 2026 · ▲ 7 on Hugging Face

– 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

71%Highly rated

?Highly ratedVote to see the score

arXivSoftware engineering agents

Harness Engineering for Software Engineering via Modular Executable Dev-Primitives

Dev-Primitives turn repository artifacts into active, resident-LLM agents with self-modification interfaces, and HERMES improves software engineering benchmarks by 12.4% over baselines while cutting inference costs by 26.2%.

Haibo Jin, Xinjie Li, Peng Kuang, Haohan Wang

Published Oct 6, 2026 · ▲ 2 on Hugging Face

0% Readers0 of 1 upvoted

14/20 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: 14 of 20 reviewers recommend it

lenient 5/5

medium 9/10

strict 0/5

84%Must read

?Must readVote to see the score

arXivRL for LLMs

HiPLEX: Hierarchical Policy Factorization for Full Duplex Speech Language Models

HiPLEX factorizes full-duplex speech policies into timing and content controllers to jointly optimize interaction dynamics via reinforcement learning. It lowers takeover rates, reduces interruption latency, and improves human-like turn timing versus GRPO.

Kyudan Jung, Hyunsin Park, Yoonhyung Lee, Jinhwan Park and 4 more

Published Oct 6, 2026

100% Readers1 of 1 upvoted

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 6/10

strict 1/5

76%Highly rated

?Highly ratedVote to see the score

arXivTool use & function calling

Hiding Tool Latency in On-Device Cascaded Voice Agent through Speculative Execution

Speculative tool execution predicts tool calls from partial ASR to run them during speech, cutting median voice-agent response latency from 5.79 s to 4.60 s.

Kyudan Jung, Hyunsin Park, Yoonhyung Lee, Jinhwan Park and 4 more

Published Oct 6, 2026 · ▲ 1 on Hugging Face

100% Readers1 of 1 upvoted

7/20 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: 7 of 20 reviewers recommend it

lenient 4/5

medium 2/10

strict 1/5

78%Highly rated

?Highly ratedVote to see the score

arXivVideo generation

ALIVE: Interaction-Aligned Object Insertion for First-Frame-Guided Video Editing

ALIVE inserts objects that interact with video contents via first-frame editing and interaction guidance, outperforming baselines on interaction and insertion benchmarks.

Zhenghong Zhou, Zhe Lin, Jiebo Luo, Yuqian Zhou

Published Oct 6, 2026 · ▲ 3 on Hugging Face · Code

– 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.

78%Highly rated

?Highly ratedVote to see the score

arXivMixture of experts

Stepped MoE: Segment-Level Routing with Configurable Inference Complexity

Stepped MoE unifies elastic architectures and sparse gating to adapt model capacity to deployment constraints and input requirements, outperforming dense counterparts by 2-5%.

Arnav Kundu, Zhaoyang Xu, Bairu Hou, Chang Gao and 2 more

Published Oct 5, 2026 · ▲ 1 on Hugging Face

100% Readers1 of 1 upvoted

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

76%Highly rated

?Highly ratedVote to see the score

arXivGANs & VAEs

Empirical Variational Autoencoder

Empirical Variational Autoencoder learns autoregressive latent priors empirically via one linear layer to close the VAE prior-posterior gap, yielding high-fidelity sequential generation competitive with diffusion models at much faster inference.

Kaede Shiohara

Published Oct 5, 2026 · ▲ 7 on Hugging Face · Code ★ 4

100% Readers1 of 1 upvoted

6/20 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: 6 of 20 reviewers recommend it

lenient 3/5

medium 3/10

strict 0/5

71%Highly rated

arXivVision-language-action models

RealtimeWAM: One-Step Asynchronous World Action Models

RealtimeWAM uses teacher-anchored consistency distillation and cross-expert wavefront pipelining for one-step asynchronous action generation, achieving near-lossless performance with ~25x speedup.

Chengtao Lv, Jinyang Du, Shuyi Feng, Yang Yong and 6 more

Published Oct 5, 2026 · ▲ 12 on Hugging Face · Code ★ 2,880

0% Readers0 of 1 upvoted

14/20 AI panelreviewers recommend it

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

AI panel: 14 of 20 reviewers recommend it

lenient 4/5

medium 9/10

strict 1/5

91%Must read

ICML 2026Alignment & preference optimization

HuatuoGPT-3: RL-Only Domain Adaptation from Base Models

OnePO enables RL-only medical domain adaptation via adaptive objective evolution and teacher retirement, yielding HuatuoGPT-3 that surpasses frontier models.

Junying Chen, Xinyuan Xie, Ziniu Li, Wenyuan Gu and 6 more

Published Oct 5, 2026 · ▲ 15 on Hugging Face · Code ★ 12

100% Readers1 of 1 upvoted

16/20 AI panelreviewers recommend it

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 10/10

strict 1/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026RL for LLMs

Rationale-Guided Policy Optimization: Learning to Reason with Adaptive Rationale Scaffolding

RGPO adaptively uses ground-truth rationales as temporary scaffolds to generate improved responses for on-policy RL, then transfers only higher-reward model outputs back, reducing reward sparsity and improving text and multimodal reasoning.

Hoang Phan, Minh Pham, Chau Pham, Chinmay Hegde and 2 more

Published Oct 5, 2026 · ▲ 7 on Hugging Face · Code ★ 1

– 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

78%Highly rated

?Highly ratedVote to see the score

arXivImage recognition

WildMatch: Weakly Supervised Image Matcher Adaptation for Wildlife Re-Identification

WildMatch adapts pretrained image matchers to wildlife identification using only identity labels, improving retrieval accuracy and learning transferable matching priors without keypoint annotations.

Turhan Can Kargin, Piotr Kubaty, Ekaterina Rostovskaya, Izabela Wierzbowska and 2 more

Published Oct 5, 2026 · ▲ 2 on Hugging Face · Code

– 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 5/5

medium 6/10

strict 0/5

76%Highly rated

?Highly ratedVote to see the score

arXivVideo generation

S2PD: Serial-to-Parallel Diffusion for Physically and Logically Consistent Video Generation

S2PD switches from autoregressive to parallel diffusion during denoising to enforce physical and logical consistency with faster sampling than fully serial methods.

Jeffrey Hu, Daniel Olmeda Reino, Ayush Tewari

Published Oct 5, 2026 · 0 citations · ▲ 1 on Hugging Face · Code ★ 4

– 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.

83%Must read

?Must readVote to see the score

arXivRobot manipulation

Execution-Aligned Progressive Noise for Consistent Asynchronous Replanning in Generative Robot Policies

Execution-Aligned Progressive Noise structures noise across chunks and time to maintain consistent generative states during asynchronous replanning, achieving up to 96.7% real-robot success.

Di Wu, Ping Liu, Xuhua Chen, He Zheng and 2 more

Published Oct 5, 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 6/10

strict 2/5

76%Highly rated

?Highly ratedVote to see the score

arXivDeep RL

MEND: RL For Flow Models via Proximal Velocity Matching

MEND uses proximal velocity matching to cap rewards and accept only cost-effective sample moves, outperforming prior flow-model RL methods in far fewer updates without KL penalties or reference models.

Shreshth Saini, Neil Birkbeck, Yilin Wang, Balu Adsumilli and 1 more

Published Oct 5, 2026 · ▲ 2 on Hugging Face · Code ★ 1

– 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 3/5

medium 7/10

strict 0/5

78%Highly rated

?Highly ratedVote to see the score

arXivRobot manipulation

Conditional Trajectory Peaks: Single-Pass Multimodal Policies over Action Chunks

Conditional Trajectory Peaks predicts multimodal action-chunk candidates in a single pass, achieving 97.25% LIBERO success and 3× faster inference while preserving diverse behaviors.

Di Wu, Rongtian Shen, Ping Liu, Xuhua Chen and 3 more

Published Oct 5, 2026 · ▲ 1 on Hugging Face

– 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 3/5

medium 5/10

strict 3/5

78%Highly rated

?Highly ratedVote to see the score

arXivProbing & representation analysis

Learning to Read the Contextual Tokens in Diffusion Transformers

A framework maps diffusion transformer contextual tokens through a frozen LLM to reveal they encode global emerging scene semantics early, inspiring contextual alignment that improves generation quality.

Omer Dahary, Etai Sella, Hadar Averbuch-Elor, Daniel Cohen-Or and 1 more

Published Oct 5, 2026 · ▲ 5 on Hugging Face

– 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 5/10

strict 2/5

86%Must read

?Must readVote to see the score

arXivTool use & function calling

Judged Useless, Queried Anyway: Tool-Using Agents Rarely Turn Their Own Evidence Judgments into Stopping Decisions

Agents judge failing retrieval results useless but rarely stop; enforcing answers after five useless judgments improves success and fixes stopping.

Chubin Zhang, Zhenglin Wan, Xingrui Yu, Jingxuan Wu and 3 more

Published Oct 5, 2026 · ▲ 3 on Hugging Face · Code ★ 3

– ReadersNo votes yet

14/20 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: 14 of 20 reviewers recommend it

lenient 4/5

medium 7/10

strict 3/5

89%Must read

?Must readVote to see the score

arXivImage restoration

JLD: Perceptual Distance Through A Jacobian Lens

JLD defines a perceptual image distance via a Jacobian-derived metric tensor from frozen vision encoders, achieving state-of-the-art correlation with human judgments and resolution robustness.

Shreshth Saini, Balu Adsumilli, Alan C. Bovik

Published Oct 5, 2026 · ▲ 3 on Hugging Face · Code ★ 3

– 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

86%Must read

?Must readVote to see the score

arXivLong-context modeling

HLA: Expressive Hybrid Linear Attention via Chunk-Wise Dynamic Mixing

Hybrid Linear Attention introduces query-dependent chunk-level routing for Gated DeltaNet, improving long-context benchmarks by up to 5.57 points via adaptive recurrent memory composition.

Zhuokun Chen, Xi Lin, Xiyu Wu, Jiahao He and 2 more

Published Oct 5, 2026 · ▲ 2 on Hugging Face

– ReadersNo votes yet

14/20 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: 14 of 20 reviewers recommend it

lenient 4/5

medium 8/10

strict 2/5

74%Highly rated

?Highly ratedVote to see the score

arXivAgent benchmarks & environments

MiniCorp: The Last Mile of the AI Agent Firm

MiniCorp is a simulated office environment that generates longitudinal, counterfactual enterprise data to study autonomous AI-run companies and train adaptive agents.

Jingying Zeng, Zhenwei Dai, Jinning Li, Changho Shin and 5 more

Published Oct 5, 2026 · ▲ 8 on Hugging Face

– 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

67%Highly rated

arXivAlignment & preference optimization

Collaborative Personalized Preference Alignment for LLMs under Data Deficiency

APO learns shared LLM aligner initializations via grouped gradient coordination to enable few-shot personalization under heterogeneous preferences and scarce feedback, improving over baselines with 20 local examples.

Liyan Yang, Yige Yuan, Zhiqin Yang

Published Oct 5, 2026 · ▲ 5 on Hugging Face

0% Readers0 of 1 upvoted

12/20 AI panelreviewers recommend it

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

80%Highly rated

?Highly ratedVote to see the score

arXivRobot manipulation

InterMimicGen: Scaling Humanoid Loco-Manipulation through Self-Evolving Motion Imitation

InterMimicGen retargets human motion capture to humanoids and self-evolves tracking data via iterative simulation-verified augmentation to scale dexterous loco-manipulation.

Yucheng Zhang, Sirui Xu, Jinhong Li, Liuyu Bian and 7 more

Published Oct 5, 2026 · ▲ 4 on Hugging Face

100% Readers1 of 1 upvoted

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 4/10

strict 1/5

69%Highly rated

?Highly ratedVote to see the score

arXivVision-language-action models

Arm-wise Compositional Generalization in Dual-Arm Vision-Language-Action Models

ACG-Bench evaluates arm-wise compositional generalization in dual-arm vision-language-action models via AE-VLA, which achieves 21.53% simulated and 39% real-world success versus under 6% baselines.

Zaibin Zhang, Binghao Ran, Yuhan Wu, Zhongbo Zhang and 9 more

Published Oct 5, 2026 · ▲ 4 on Hugging Face

0% Readers0 of 1 upvoted

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

80%Must read

?Must readVote to see the score

arXivLLM agents & planning

Beyond Semantic Similarity: Performance and Costs of Agentic Retrieval for Complex Tasks

Agentic retrieval combining LLM reasoning with dense retrieval improves nDCG@10 by 8.7 points over standard retrieval but requires 107 seconds and 764K input tokens per query.

Reza Esfandiarpoor, Radek Osmulski, Yauhen Babakhin, Gabriel de Souza P. Moreira and 7 more

Published Oct 5, 2026 · ▲ 7 on Hugging Face

– 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 6/10

strict 1/5

89%Must read

arXivLLM reasoning & chain of thought

Sharpen Without Search: On-Policy Distillation of Sequence-Level Power Distribution

On-policy power distillation trains models to generate sharpened answers directly, improving single-sample math reasoning by up to 27.3 points and outperforming multi-candidate sampling and reward-based methods.

Erfan Baghaei Potraghloo, Seyedarmin Azizi, Arya Fayyazi, Saeid Shokoufa and 3 more

Published Oct 5, 2026 · ▲ 4 on Hugging Face · Code ★ 1

– ReadersNo votes yet

16/20 AI panelreviewers recommend it

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

AI panel: 16 of 20 reviewers recommend it

lenient 4/5

medium 9/10

strict 3/5

88%Must read

?Must readVote to see the score

arXivVision-language-action models

When to Switch: Reliable Action-Chunk Extension for Vision-Language-Action Models

RACE predicts subskill transition timing to extend VLA action chunks reliably, reducing stop-and-go idle time ~5x on real robots while improving success rates.

Seonghoon Yu, Dongwon Kim, HyungRok Jung, Yoonjae Baek and 3 more

Published Oct 5, 2026 · ▲ 13 on Hugging Face · Code ★ 3

– 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 5/5

medium 9/10

strict 1/5

89%Must read

arXivUncertainty & calibration

Certification of Real Images through Calibrated Content Authentication

Deepfake detectors degrade to 76% accuracy and near-zero under attacks, so calibrated reconstruction-based authentication bounds false real-image certification to 1%.

Sarim Hashmi, Abdelrahman Elsayed, Mohammed Talha Alam, Samuele Poppi and 1 more

Published Oct 5, 2026 · ▲ 13 on Hugging Face · Code ★ 2

– ReadersNo votes yet

16/20 AI panelreviewers recommend it

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

88%Must read

?Must readVote to see the score

WWW 2026LLM reasoning & chain of thought

Base Models Can Reason By Taking a Cue From Training Data

Fixing initial token cues in base models boosts reasoning to match RL performance, with effects traced to training data associations that can be causally edited.

Sophie L. Wang, Amil Dravid, Rulin Shao, Kevin Farhat and 2 more

Published Oct 5, 2026 · ▲ 12 on Hugging Face · Code ★ 2

– 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 9/10

strict 2/5

89%Must read

?Must readVote to see the score

arXivVideo generation

HLA-WM: Hybrid Linear Attention for Long-Horizon Video World Models

HLA-WM combines geometry-guided retrieval with recurrent linear attention to fix long-range forgetting in video world models, improving 60-second consistency metrics by up to 28.5% with 12× lower memory and no retraining.

Zhuokun Chen, Feng Chen, Xi Lin, Xiyu Wu and 3 more

Published Oct 5, 2026 · ▲ 6 on Hugging Face · Code ★ 2

– 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

78%Highly rated

?Highly ratedVote to see the score

arXivRL for LLMs

LoGRA: Scaling LLM Reinforcement Learning with Low-Rank Gradient Sketches

LoGRA reduces LLM reinforcement learning memory by up to 45.7% via low-rank gradient sketches and predicted-KL step control, enabling 27B-parameter training on single nodes.

Shaokun Zhang, Yifan Zhang, Jian Hu, Yueying Li and 4 more

Published Oct 5, 2026 · ▲ 11 on Hugging Face

– 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 5/5

medium 5/10

strict 1/5

78%Highly rated

?Highly ratedVote to see the score

arXivEfficient inference & serving

Closing the Context Gap: Activation Alignment for Tabular In-Context Learning

Activation alignment trains a linear map to align partial-context student activations with full-context teacher activations, significantly improving tabular in-context learning efficiency and recovering much of the performance gap.

Yoel Zeldes

Published Oct 5, 2026 · ▲ 5 on Hugging Face · Code ★ 1

– 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

ICLR 2027Tabular data

Adapting prior-data fitted networks for tabular anomaly detection

Frozen and fine-tuned TabPFN representations for tabular anomaly detection yield ZEN and FOCUS, surpassing all ADBench baselines in AUROC despite unsupervised deployment and contaminated reference sets.

Maximilian Bershtman, Niv Cohen

Published Oct 5, 2026 · ▲ 5 on Hugging Face · Code ★ 1

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

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

80%Must read

?Must readVote to see the score

arXivIn-context learning & prompting

Learning to Learn a Language

Prior-Fitted Language Model, trained solely on synthetic non-linguistic data, learns to infer and predict real languages from context with frozen weights, achieving strong cross-lingual compression and reasoning without ever seeing real text.

Lennart Carstens-Behrens, Holger Fröhlich

Published Oct 5, 2026 · ▲ 7 on Hugging Face · Code ★ 1

– 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 4/5

medium 6/10

strict 2/5

78%Highly rated

?Highly ratedVote to see the score

arXivAI oversight & deception

SoK: Semantic Decision Engines in Network Control Loops

Systematizing 139 semantic decision engine families reveals most miss network deadlines and verification, with only four reporting deadline attainment; unverified decisions reverse admission verdicts under queued execution, prompting minimum reporting rules and a research agenda.

Delong Li, Chen Li, Xu Wang, Haochen Gong and 2 more

Published Oct 5, 2026 · ▲ 6 on Hugging Face · Code

– 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 1/5

medium 8/10

strict 2/5

88%Must read

?Must readVote to see the score

arXivFairness & bias

Noise Out, Bias In: Targeted Bias Injection in Diffusion Language Models via Closed-Loop Activation Steering

Targeted bias injection via closed-loop activation steering exploits diffusion language model denoising trajectories to steer frozen models toward adversarial demographic answers with minimal corruption.

Sarim Hashmi, Mukul Ranjan, Abdelrahman Elsayed, Muhammad Umer Sheikh and 2 more

Published Oct 5, 2026 · ▲ 14 on Hugging Face · Code ★ 3

– 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 5/5

medium 9/10

strict 1/5

88%Must read

?Must readVote to see the score

arXivFairness & bias

FairRSFM: A Biome-Aware Benchmark and Debiasing Framework for Remote Sensing Foundation Models

FairRSFM benchmarks remote sensing foundation models by biome to expose hidden ecological performance disparities and tests debiasing methods without backbone updates. Aggregate metrics consistently mask large biome-dependent gaps, though mitigation effectiveness varies by model and task.

Md Aminur Hossain, Omkumar Vaghasiya, Rajeev Ranjan Dwivedi, Vinod Kurmi and 1 more

Published Oct 5, 2026 · ▲ 5 on Hugging Face · Code

– 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 5/5

medium 8/10

strict 2/5

83%Must read

?Must readVote to see the score

arXivLLM reasoning & chain of thought

What Matters for Latent Reasoning with Flow Matching

FLaRe uses flow matching for latent reasoning that is useful, diverse, explainable, refinable and efficient, reaching 97% of explicit chain-of-thought accuracy at 25% latency.

Yassine Ouali, Adrian Bulat, Georgios Tzimiropoulos

Published Oct 5, 2026 · ▲ 10 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 4/5

medium 9/10

strict 0/5

80%Must read

?Must readVote to see the score

arXivAgent memory

Capability-Driven Self-Evolution of Agent Memory

PrisMem drives agent memory self-evolution via capability-specific guidance, dependency-aware selection, and trace-guided integration, outperforming baselines by up to 10.54 points on million-token benchmarks.

Yaoqi Chen, Yuru Feng, Qianxi Zhang, Baotong Lu and 7 more

Published Oct 5, 2026 · ▲ 6 on Hugging Face

– 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 4/5

medium 7/10

strict 1/5

71%Highly rated

?Highly ratedVote to see the score

arXivAlignment & preference optimization

Representation-Space MMD for Diffusion Language Models

Post-training minimizes representation-space MMD between diffusion language model outputs and references via retained token features, improving perplexity, accuracy, and parallel decoding.

Ilya Drobyshevskiy, Ilia Sudakov, Maksim Semenov, Denis Kuznedelev and 6 more

Published Oct 5, 2026 · ▲ 16 on Hugging Face · Code ★ 10

– ReadersNo votes yet

7/20 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: 7 of 20 reviewers recommend it

lenient 3/5

medium 3/10

strict 1/5

83%Must read

?Must readVote to see the score

arXivLLM pretraining & scaling laws

Towards Looped Models Done Right, Part II: Rethinking at Fixed Points

Looped language models use fixed-point convergence to enable truncated training, shared KV caches, faster prefill, and faster RL updates, while a learned depth prior and orthogonal input injection improve perplexity across scales.

Benhao Huang, Chufan Shi, Junlin Chen, Shicheng Wen and 3 more

Published Oct 5, 2026 · ▲ 17 on Hugging Face · Code ★ 30

– 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 3/5

medium 8/10

strict 2/5

86%Must read

?Must readVote to see the score

arXivAgent benchmarks & environments

AutoSciBench: Autonomous Benchmark Generation for Evaluating Scientific Agents

AutoSciBench autonomously generates and iteratively adapts scientific agent benchmarks via recipes and concepts, reducing solver accuracy by over 22 points versus human benchmarks while improving quality ratings.

Dongki Kim, Namkyeong Lee, Surag Nair, Carl Edwards and 6 more

Published Oct 4, 2026 · ▲ 17 on Hugging Face

100% Readers1 of 1 upvoted

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

78%Highly rated

?Highly ratedVote to see the score

arXivLLM agents & planning

EVISKILL: Grounding Skill Evolution in Replayable Evidence

EVISKILL grounds LLM skill evolution in replayable evidence cards linking edits to supporting contexts, using targeted replay for verification and global validation for incorporation.

Yan Zhou, Yili Wang, Yiwei Dai, Qinggang Zhang and 1 more

Published Oct 4, 2026 · ▲ 11 on Hugging Face · Code ★ 8

100% Readers1 of 1 upvoted

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 4/5

medium 4/10

strict 1/5

Show 20 more papers