reinforcement learning

The papers

Most central to this idea first, not most recent.

• Big Techcs.LG

  Adaptive curricula in AI agents secretly degrade the learner's ability to keep training

  Unsupervised Environment Design fails when learners lose trainability due to parameter growth and degraded representations.

  Mila – Quebec AI Institute, Université de Montreal · Nov 2026

• Big Techcs.LG

  AI agents can score higher by keeping luckier simulated outcomes

  Branch-and-commit only improves performance when candidate returns vary, and most gain is from selecting among random outcomes rather than better actions.

  Technion, Israel Institute of Technology, Nvidia Research · Nov 2026

• Big Techcs.IR

  User-agnostic reward component, not the personalized one, narrows AI recommender picks

  The item-mean component of collaborative rewards causes exposure collapse; per-item centering or standardization corrects it.

  University of California, San Diego, Google Inc. · Nov 2026

• Independentcs.LG

  Reward-trained image models can get worse the more candidates you generate

  Visual RL post-training can degrade the high-quality outputs that Best-of-N sampling relies on.

  Nov 2026

• Big Techcs.LG

  Robots can improve on untrained tasks by practising diverse ones

  Reinforcement learning enables a single generalist robot policy to improve across many tasks simultaneously and generalize to new ones.

  ETH Zurich, MPI for Intelligent Systems, Tübingen · Nov 2026

• Big Techcs.AI

  Language models improve psychiatric interviews after training with simulated patients

  Training a language model with multi-turn reinforcement learning in simulated psychiatric intakes improves its diagnostic, communication, and safety performance over its base model.

  Google DeepMind, Google Research · Oct 2026

• Big Techcs.RO

  Robots learn to pass objects between hands without scripted movements

  Bimanual dexterity emerges from a simple object-centric reward by varying training environments, achieving 76.3% success on real unseen objects.

  Amazon FAR, UCSD · Nov 2026

• Chinese Techcs.LG

  Reasoning models usually inherit “wait” from training data, not reinforcement learning

  The word "wait" in reasoning models is inherited from training data, not coined by reinforcement learning, which only reshapes it.

  The Hong Kong University of Science and Technology, Huawei Noahah’s Ark Lab · Oct 2026

• Chinese Techcs.CL

  Chatbots can learn to clarify vague requests without unnecessary questions

  DialogGRPO trains large language models to ask clarifying questions, recovering missing information on over 88% of vague queries.

  Zhejiang University, Ant Group · Nov 2026

• Big Techcs.CL

  Rewarding moral arguments makes a language model more consistent, not sounder

  Rewarding argument quality makes language models more decisive and order-invariant on moral dilemmas without improving moral validity, so consistency is not validity.

  Google DeepMind · Oct 2026

• Big Techcs.LG

  Two-memory approach lets AI agents learn new tasks without forgetting old ones

  Separating value functions into permanent parametric and transient non-parametric features balances stability and plasticity in continual reinforcement learning.

  McGill University, Mila · Nov 2026

• Big Techcs.LG

  Monitors can curb reward hacking in RL training, but models learn to evade them

  Monitor-based interventions can reduce reward hacking in RL, but careful design is needed to avoid evasion.

  MATS, Google DeepMind · Nov 2026

• Top Universitycs.LG

  Language models can lag 50 to 150 updates behind training

  Standard bf16 rounding silently discards most RL updates, creating a hidden lag that unbiased rounding eliminates.

  Stanford University · Nov 2026

• Big Techcs.RO

  Robots can learn door lifting with ten times fewer demonstrations

  Reusing a frozen reinforcement-learned action decoder and learning a visual latent interface makes dexterous manipulation from visual demonstrations far more data-efficient.

  University of California, Berkeley, Amazon · Nov 2026

• Top Universitycs.LG

  AI models can be trained end to end with one algorithm

  Foundation model training can be unified into a single likelihood-based algorithm treating all stages as maximum-likelihood fitting to model-generated targets.

  Carnegie Mellon University, Johns Hopkins University · Oct 2026

• Top Universitycs.LG

  Single-completion training matches multi-completion diversity by focusing on harder tasks

  Reweighting each prompt's regularization or reward turns ordinary single-sample reinforcement learning into an optimizer that shifts learning pressure toward harder problems and curbs sharpening.

  EPFL, ETH Zürich · Nov 2026

• Academiccs.RO

  Multi-agent framework learns to coordinate puzzle piece assembly without planning

  Multi-agent reinforcement learning with a graph neural network can coordinate multiple parts to assemble Tangram puzzles simultaneously.

  Technical University of Darmstadt · Nov 2026

• Research Labcs.RO

  Simulated humanoid robots learn to walk with 81% fewer falls

  STRIDE learns humanoid walking in 14 minutes using only onboard IMU acceleration, without external velocity tracking.

  Technical University of Darmstadt, Robotics Institute Germany (RIG) · Nov 2026

• Chinese Techcs.CL

  RL-trained agents search multilingual corpora directly, beating retrieval methods

  MARS uses reinforcement learning to train multilingual search agents that directly query raw corpora via shell commands, improving evidence acquisition and reducing redundancy.

  Beijing Jiaotong University, Tencent Inc. · Nov 2026

• Big Techcs.LG

  AI agents’ performance can be predicted using 100-fold smaller training runs

  Reinforcement learning follows predictable scaling laws, with Elo scores enabling performance extrapolation up to two orders of magnitude.

  Google DeepMind · Oct 2026

• Top Universitycs.AIcode

  AI agents build a Hearthstone Battlegrounds bot that rivals top players

  A self-improving team of LLM research agents built a Hearthstone Battlegrounds AI that beats top players, while the LLMs themselves play far worse.

  Nanjing University, Shanghai AI Laboratory · Oct 2026

• Big Techcs.AI

  AI auditors find more problematic behavior by reading models’ internal signals

  Red-team audits of language models improve by treating them as Markov decision processes and using linear probes on the target model's activations to guide beam search.

  Anthropic Fellows Program, Anthropic · Oct 2026

• Chinese Techcs.AI

  Image generators follow object relationships more faithfully with targeted prompt edits

  StruPO uses calibrated scene graphs to learn prompt edits that improve image quality without changing the requested scene.

  Zhejiang University, Tencent AI Lab · Oct 2026

• Big Techcs.LG

  RL method helps AI agents learn high-level task strategies from sparse rewards

  A strategy-guided reinforcement learning method that routes LLM agents through hierarchical skills beats flat skill retrieval.

  Microsoft AI · Nov 2026

• Industrycs.RO

  Simulated humanoid robots match motion tracking quality 3.6 times faster

  Exact simulator gradients, with clipping and multistep regression, train a humanoid tracking policy to state-of-the-art scores in 20 minutes.

  UC Davis · Nov 2026

• Big Techcs.AI

  Browser-game training makes AI webpage generation surpass a frontier model

  WebCatalyst uses high-fidelity reinforcement learning to significantly improve interactive HTML generation, outperforming Claude Opus 4.7.

  Tsinghua University, Microsoft Research · Nov 2026

• Top Universitycs.RO

  Soccer-playing humanoid robot learns four skills from human motion data

  SoccerBot is a humanoid soccer system that uses human motion data to learn a single policy for moving, dribbling, trapping, and kicking with human-like motion.

  University of Science and Technology of China, State Key Laboratory of General Artificial Intelligence, BIGAI · Nov 2026

• Big Techcs.CL

  AI agents break chatbot safeguards more often in fewer exchanges

  An autonomous multi-turn jailbreak attacker can adapt to victim replies, decide when to stop, and achieve SOTA success rates without an in-loop judge.

  University of Rochester, Microsoft Research · Oct 2026

• Chinese Techcs.CV

  Training on interfaces helps visual models inspect images beyond screens

  GUI grounding supervision trains vision-language models to inspect images sequentially, improving perception beyond GUI tasks.

  Wuhan University of Technology, Institute of Automation, Chinese Academy of Sciences (CAS) · Oct 2026

• Chinese Techcs.AI

  AI agents complete more multistep tasks by reusing related experience

  Beyond-Local Policy Optimization improves long-horizon LLM agent training by reusing experience across tasks to estimate step-level credit without a critic.

  Beihang University, Alibaba Group · Nov 2026

• Independentcs.LG

  Aligning rounding locations between inference and training stabilizes BF16 LLM reinforcement learning

  Aligning where BF16 rounding happens between inference and training, specifically in the RoPE path, stabilizes LLM reinforcement learning.

  Oct 2026

• Big Techcs.RO

  Robots learn dexterous skills from a single egocentric human video

  HandMimicX learns dexterous manipulation from a single human video using object motion and interaction priors for spatial generalization.

  Tsinghua University, TsingTec · Nov 2026

• Chinese Techcs.LG

  Low-precision training of AI expert models needs matching fixes on both sides

  Symmetric quantization recovery closes the accuracy gap in FP4 reinforcement learning for mixture-of-experts LLMs.

  Huawei · Nov 2026

• Top Universitycs.AI

  Training shifts a small theorem prover toward fewer, larger proof steps

  Reap trains a 1.7B-parameter prover that solves 79.9% of miniF2F-test via Monte Carlo search.

  Peking University, IQuest Research · Nov 2026

• Top Universitycs.CL

  RL gains in no-think mode stem from reactivating existing reasoning

  Reinforcement learning gains in no-think mode of hybrid LLMs largely come from reactivating the existing think mode reasoning.

  King Abdullah University of Science and Technology, Shanghai Jiao Tong University · Oct 2026

• Big Techcs.LG

  Detached shorter-horizon features boost long-horizon reinforcement learning

  Separating the discount used for representation learning from the behavior discount improves deep RL performance across Atari and continuous control.

  NVIDIA Research · Oct 2026

• Academiccs.LG

  AI agents can match the theoretical limit for exploration losses

  An algorithm for episodic tabular reinforcement learning achieves the minimax regret bound sqrt(H^3SAK) with no extra logarithmic factors.

  University of Wisconsin–Madison, IRIT · Oct 2026

• Big Techcs.CL

  Training search agents to assess missing evidence improves research performance

  Training retrieval-augmented LLMs with a dual-reward self-assessment phase improves search-state awareness and beats baselines on multi-hop and deep research benchmarks.

  Amazon, Korea University · Oct 2026

• Chinese Techcs.LG

  Language models reason better when conflicting teacher guidance is filtered

  GIFT filters token-level distillation updates by their geometric alignment with RL gradients and anneals coverage to improve reasoning.

  Shanghai Jiao Tong University, Tencent · Oct 2026

• Big Techcs.LG

  Verifier-based rewards train coding agents toward formally correct programs

  A reward from verifier diagnostics, plus integrity auditing, raises verified code generation from 66.1% to 93.5% on DafnyBench and from 56.3% to 100% on SWE-Proof.

  UC Berkeley, AWS AI Labs · Oct 2026

• Top Universitycs.CL

  LongZero trains LLMs on self-written long-context tasks, no external data

  LongZero improves LLM long-context reasoning via self-generated training tasks without any external data.

  Tsinghua University, Tsinghua-Bosch Joint ML Center · Oct 2026

• Big Techcs.NI

  Black-box RL method finds large performance gaps in LLM network control

  LENS uses reinforcement learning to find adversarial inputs that expose large performance gaps in LLM-based network control systems without accessing model internals.

  University of Southern California, Microsoft Research · Oct 2026

• Chinese Techcs.LG

  Divergence geometry and estimation jointly improve language-model policy optimization

  GPPO unifies LLM RL algorithms via divergence and estimator choices, outperforming standard GRPO by up to 50% on math and 14% on coding.

  Institute for AI Industry Research (AIR), Tsinghua University, SIA-Lab of Tsinghua AIR and ByteDance Seed · Oct 2026

• Top Universitycs.LG

  Conditioning actions on returns improves value learning in reinforcement learning

  A Q-learning critic can be trained by learning return-conditioned action likelihoods, using actions themselves as supervision, improving RL performance.

  Stanford University · Oct 2026

• Chinese Techcs.LG

  Critics that predict task success, not just returns, improve long-horizon agent credit assignment

  SCALE improves long-horizon language-agent training by adding reliability-weighted success-probability credit to PPO, gaining up to 4.1 success-rate points in tested settings.

  Shanghai Jiao Tong University, Huawei Noah's Ark Lab · Oct 2026

• Big Techcs.RO

  Humanoid repositions bulky objects using its arms and torso

  A humanoid robot can reposition objects using its entire body, not just hands, after training in simulation.

  Amazon, University of California, San Diego · Oct 2026

• Academiccs.RO

  One-camera humanoids learn to catch objects they have never seen

  A humanoid robot can catch unseen objects with underhand and overhand throws using one camera.

  UC San Diego, Harvey Mudd College · Oct 2026

• Top Universitycs.AIcode

  RL post-training improves agent consistency but narrows solution coverage

  Post-training RL sharpens LLMs, improving pass@1 but reducing pass@K coverage on agentic tasks.

  Meta Superintelligence Labs, University of Wisconsin–Madison · Oct 2026

• Chinese Techcs.CV

  Single-stage RL method unifies preference alignment and few-step image generation

  AdvFidelity aligns human preferences and generates high-fidelity images in few steps using a single RL stage.

  Alibaba, Shanghai Jiao Tong University · Oct 2026

• Big Techcs.AI

  Optimizing AI persuasion rewards fabrication unless alignment is explicitly enforced

  Outcome-only reinforcement learning for persuasion boosts misaligned strategies, but combining guidance and penalties mitigates it.

  Microsoft Research, Princeton University · Oct 2026

• Big Techcs.AI

  Reinforcement learning adapts diffusion models to large routing problems

  ResizeCO adapts a diffusion model trained on small routing instances to large TSP and CVRP instances via reinforcement learning, setting a new state of the art for short run times.

  Bar-Ilan University, NVIDIA · Oct 2026

• Top Universitycs.LG

  Reinforcement learning reshapes which reasoning strategies language models prefer

  Reinforcement learning primarily reweights existing reasoning strategies rather than expanding an LLM's reasoning boundary.

  University of California, Berkeley · Oct 2026

• Top Universitycs.LG

  Dynamic GPU Shifting Speeds Language Model Reinforcement Learning Post-Training

  PivotRL dynamically shifts GPU resources among RL stages to improve throughput by up to 2.05 times.

  Shanghai Jiao Tong University, Peking University · Oct 2026

• Chinese Techcs.CV

  Adaptive reward routing improves joint audio-video diffusion training

  Adaptive Reward Routing dynamically adapts update locations and reward weights, improving joint audio-video diffusion across multiple metrics.

  MMLab@HKUST, The Hong Kong University of Science and Technology · Oct 2026

• Big Techcs.AI

  Reinforcement learning teaches scientific ideation agents to explore broader possibilities

  AI Night-Scientist uses reinforcement learning to teach LLMs when to depart from predictable reasoning, improving scientific proposal diversity and originality.

  University of Illinois Urbana-Champaign, Microsoft · Sept 2026

• Big Techcs.LG

  Learning the search harness improves frozen models’ scientific problem solving

  HarnessTTT learns to generate executable harnesses at test time, allowing a frozen small executor to outperform fine-tuned systems across scientific discovery tasks.

  University of Wisconsin-Madison, NVIDIA · Oct 2026

• Chinese Techcs.AI

  Memory-guided framework brings differential privacy to million-token LLM inference

  DP-MIND enables private inference on million-token contexts by perturbing chunks, propagating memory, and local denoising.

  Zhejiang University, Ant Group · Oct 2026

• Chinese Techcs.CV

  Refined Questions and Teacher Guidance Improve Long-Horizon Multimodal Search

  MMSearch-OPD improves long-horizon multimodal search agents by combining data refinement with joint reinforcement learning and on-policy distillation.

  Peking University, HKUST(GZ) · Oct 2026

• Big Techcs.RO

  Adaptive sampling helps million-environment robot training solve harder tasks

  Sampling task configurations where a policy has moderate success rates lets massively parallel robot RL solve harder tasks.

  University of Washington, NVIDIA · Oct 2026

• Chinese Techcs.CL

  Reinforcement fine-tuning improves LLM social reading more than game wins

  Reinforcement fine-tuning improves LLMs' social reading and influence in deduction games, but not when only terminal outcomes are optimized.

  Peking University, Alibaba Group · Oct 2026

• Academiccs.CV

  Local and global rewards improve long-horizon video consistency

  Local-global rewards improve 3D consistency in long-horizon camera-controlled video generation.

  California Institute of Technology, World Labs · Oct 2026

• Chinese Techcs.CL

  Specialized memory writer improves persistent state across multiple LLM frameworks

  A specialized memory writer trained with attributed chunk-level reinforcement learning outperforms general-purpose models across four long-term-memory harnesses.

  Tencent Hunyuan · Oct 2026

• Independentcs.CV

  Post-training recipe makes compact multimodal reasoners beat larger models

  MMVistaReason demonstrates that systematic open-data construction and capacity-aware post-training provides a practical path to reliable multimodal reasoning.

  Oct 2026

• Top Universitycs.LG

  Empowerment Leads Agents to Central States, but Not Always Adaptability

  Empowered states are central in information geometry, enabling task adaptation only in tabular settings.

  Princeton University, UC Berkeley · Oct 2026

• Independentcs.CV

  Training long-horizon multimodal research agents with reinforcement learning at scale

  Long-MDR makes 128k-context, 75-turn multimodal reinforcement learning more efficient and stable through distillation warmup, gradual horizon expansion, and entropy-based recovery.

  Oct 2026

• Chinese Techcs.CV

  RL-trained cache scheduler adapts video diffusion to user-specified speedups

  MORCA uses latent-aware offline-to-online reinforcement learning to schedule cache reuse, improving video diffusion fidelity at user-specified acceleration targets over existing caching methods.

  Shanghai Jiao Tong University, Alibaba Group · Oct 2026

• Big Techcs.RO

  HULK teaches humanoids to carry heavy loads while staying balanced

  HULK uses model predictive control to guide reinforcement learning for humanoid whole-body forceful loco-manipulation under sustained payloads.

  University of Michigan, University of California, Los Angeles · Oct 2026

• Independentcs.RO

  Filtered critic gradients speed reinforcement learning for flow-based robot policies

  QF3 is the first off-policy flow RL method to train humanoid locomotion from scratch, achieve 10x speedup, and transfer zero-shot to hardware.

  Oct 2026

• Chinese Tech

  Q-target training improves in-context reinforcement learning from suboptimal data

  QTPT enables robust in-context RL from weak offline data by training with Q-targets instead of behavior cloning.

  Shanghai Jiao Tong University, Alibaba Group · Sept 2026

• Research Labcs.LG

  Density-aware reweighting accelerates multi-reward RL for LLMs

  Inverse-square-root weighting by active-group density strengthens sparse reward signals and speeds multi-objective LLM reinforcement learning while preserving competitive final performance.

  University of Chinese Academy of Sciences, University of Minnesota Twin Cities · Oct 2026

• Top Universitycs.LG

  Unmodified posterior sampling achieves minimax regret in reinforcement learning

  Posterior sampling for reinforcement learning is minimax optimal without structural prior assumptions.

  KAIST · Oct 2026

• Independentcs.LG

  A compact AlphaZero-style chess system reaches 3,251 Elo after 2.5 days on a single GPU node

  An engineered AlphaZero-style chess system achieves 3251 Elo in 2.5 days on one 8-GPU node.

  Oct 2026

• Top Universitycs.LG

  A framework predicts RL post-training outcomes without running reinforcement learning

  PoEM predicts reinforcement learning outcomes for new reward functions without retraining by composing existing policies.

  MIT CSAIL · Sept 2026

• Academiccs.RO

  Synthetic demonstrations let robot policies escape sparse-reward failures

  SynthDemo-RL generates synthetic demonstrations from simulator state, distills via SFT, and refines with PPO to achieve high success on tasks previously at 0%.

  Fujitsu Limited, The Institute of Statistical Mathematics · Sept 2026

• Academiccs.AIcode

  Agents learn selectively from imperfect vision-language model teachers

  SAGE selectively queries a VLM teacher during uncertain steps and distills its guidance into a lightweight RL policy.

  Fondazione Bruno Kessler, University of Torino · Sept 2026

• Big Techcs.CL

  RL with live server state boosts multi-step tool use in LLMs

  Reinforcement learning with live server state and a programmatic reward improves multi-step tool use in LLMs.

  IBM Research · June 2026

• Big Techcs.AI

  RL Training Can Amplify Compliance That Disappears Without Oversight

  Reinforcement learning amplifies scheming behavior in AI models even when the reward does not reward scheming.

  MATS Research, Google DeepMind · Oct 2026

• Chinese Techcs.RO

  Visual reward optimization can improve robots while reinforcing wrong-object mistakes

  Optimizing a learned visual reward can raise task success and wrong-object failures simultaneously, hiding the error.

  Baidu, University of California, Santa Barbara · Oct 2026

• Independentcs.LG

  Starting token cues can elicit reasoning from base language models

  Base models can reason well when response-opening token cues activate behaviors learned from training data, and reinforcement learning partly works by making those cues likelier.

  Oct 2026

• Big Techcs.LG

  Routing gradients can isolate reward hacking for removal after training

  Routing gradients from flagged samples into sacrificial parameters, then ablating those parameters, removes reward hacking without teaching the model to exploit the monitor.

  MATS, Stanford · Oct 2026

• Big Techcs.LG

  Frozen LLM critics sharpen credit assignment for long-horizon agents

  A frozen LLM value critic, policy-aware via in-context conditioning, improves step-level credit assignment in agentic RL, beating GRPO and PPO.

  Apple · Oct 2026

• Big Techcs.CL

  Language models learn to propose improvements for other language models via RL with a learned world model

  LM2LM uses reinforcement learning to train LLMs to generate effective proposals for improving other language models.

  The Pennsylvania State University, UC Santa Cruz · Oct 2026

• Top Universitycs.LG

  LLM agents learn to organize document collections before questions arrive

  Training a builder agent with reinforcement learning to reorganize documents into a filesystem workspace improves downstream question answering and transfers to unseen domains.

  Stanford University, Carnegie Mellon University · Oct 2026

• Chinese Techcs.CV

  Decoupling skills from expert policies improves multi-capability vision-language model training

  SCOPED trains capability-specific experts and distills explicit skills into a unified VLM, outperforming mixed-capability RL.

  Northeastern University, Alibaba Group · Oct 2026

• Big Techcs.LG

  Gradient norm ranks attention heads for efficient sparse and continual RL fine-tuning

  Gradient norm ranks attention heads; tuning top heads matches dense fine-tuning, and partitioning heads prevents forgetting.

  University of Washington, Tsinghua University · Oct 2026

• Chinese Techcs.AI

  Neos turns visual evidence into better follow-up search decisions

  Neos learns to interleave search and visual reasoning, achieving leading performance on multimodal deep search benchmarks.

  Fudan University, Accio Team, Alibaba Group · Oct 2026

• Big Techcs.LG

  8B model learns to construct and refine GPU optimization scripts, beating larger LLMs

  A training method teaches an 8B language model to generate correct GPU optimization scripts, outperforming larger frontier LLMs.

  University of California, San Diego, Amazon Annapurna Labs · Oct 2026

• Chinese Techcs.CV

  Stochastic attention routing enables efficient RL alignment for diffusion models

  Routing Policy Optimization uses temporary stochastic attention routes to improve reward-aligned visual generation with less training compute while preserving dense deterministic inference.

  National University of Singapore, ByteDance · Oct 2026

• Chinese Techcs.LG

  Reinforcement-trained steering agent corrects goal drift in LLM agents and multi-agent systems

  A reinforcement-trained supervisor detects goal drift and selectively redirects LLM agents, improving single-agent and multi-agent performance with few interventions.

  National University of Singapore, University of Science and Technology of China · Oct 2026

• Academiccs.RO

  Separating exploration and exploitation policies with classifier-free guidance improves diffusion RL fine-tuning

  CFG-Policy combines separate exploration and exploitation diffusion policies via classifier-free guidance to improve RL performance while preserving behavioral diversity.

  ShanghaiTech University, InstAdapt · Oct 2026

• Chinese Techcs.LG

  Criterion-aware credit assignment improves rubric-based reinforcement learning for language models

  CoCA improves rubric-based reinforcement learning by 4.2% on medicine and science benchmarks through criterion-aware credit assignment.

  Zhejiang University, Ant Group · Oct 2026

• Top Universitycs.CL

  Attackers That Learn Mid-Attack Find More Large-Model Jailbreaks

  Updating the attacker's parameters during jailbreaking raises attack success rate from 55.9% to 72.4%.

  Johns Hopkins University, National University of Singapore · Oct 2026

• Chinese Techcs.LG

  Relative rubric comparisons improve rewards for open-ended language model training

  MatrixReward uses pairwise rubric comparisons to build a win-rate matrix and derive distance-based rewards for open-ended RL.

  Zhejiang University, Renmin University of China · Oct 2026

• Top Universitycs.RO

  Human demonstrations teach humanoids whole-body manipulation without robot teleoperation

  Workhorse enables humanoid robots to perform whole-body manipulation tasks by learning from human demonstrations without robot-specific retargeting.

  University of California, Berkeley · Oct 2026

• Independentcs.LG

  Interactive software judges improve when they test applications directly

  Interaction-as-a-Judge using computer-use agents improves evaluation and generation of interactive applications over code-only methods.

  Oct 2026

• Chinese Techcs.CL

  Learning the refinement loop improves web generation beyond final pages

  Learning generation, critique, and refinement jointly in a loop improves functional web generation more than optimizing each role in isolation.

  Tsinghua University, Huawei Noah's Ark Lab · Oct 2026

• Chinese Techcs.CV

  Human-rated clips improve rewards for synchronizing generated sound and video

  SyncReward is a human-aligned reward model that improves audio-visual synchronization in generated content.

  Renmin University of China, WeChat, Tencent Inc. · Oct 2026

• Big Techcs.CV

  Adaptive feedback targets each image model’s changing weaknesses

  Adaptive, feedback-driven post-training improves diverse text-to-image models more than handcrafted reinforcement-learning data by targeting each model’s evolving weaknesses.

  University of Science and Technology of China, Microsoft Research Asia · Oct 2026

• Academiccs.LG

  Backpropagating verifier rewards improves reasoning in continuous flow language models

  A critic predicting per-token values lets continuous flow language models improve reasoning through backpropagated verifier rewards.

  Tufts University · Oct 2026

• Academiccs.AI

  Open dataset and distillation-to-reinforcement pipeline sharpen GUI grounding in 8B model

  A large GUI grounding dataset and a distillation-to-reinforcement learning training method yield state-of-the-art 8B-scale grounding models.

  Prentis AI, Westlake University · Oct 2026

• Top Universitycs.AIcode

  Reasoning models jailbreak monitors while remaining readable to humans

  Models evade chain-of-thought monitoring by phrasing, not encoding, their reasoning.

  Meridian Cambridge · Sept 2026

• Big Techcs.LG

  Bidirectional start-goal curriculum accelerates sparse-reward reinforcement learning without demonstrations.

  Bidirectional Voronoi-biased exploration curricula, growing starts from goals and goals from starts, accelerate sparse-reward goal-conditioned reinforcement learning.

  ETH Zürich, NVIDIA · Oct 2026

• Top Universitycs.RO

  Robot learns manipulation tasks from scratch in minutes with AI-generated 3D models

  ZeroBot enables robot manipulation learning from scratch in minutes using image-to-3D generative models and parallel RL.

  Imperial College London, Robotics and AI Institute · Oct 2026

• Big Techcs.CV

  Probability rewards keep visual grounding reasoning aligned with predicted objects

  Rita reduces thinking drift in visual intention grounding by using conditional probability rewards to enforce thinking-answer consistency.

  National University of Singapore, A*STAR Institute of Advanced Intelligence and Computing, Singapore · Oct 2026

• Academiccs.LG

  Language models’ responses to reward changes can be forecast before training

  Within-prompt covariance between a response feature and a reward-model score change forecasts, before training, how far a policy's use of that feature will shift.

  Mohamed bin Zayed University of Artificial Intelligence · Nov 2026

• Chinese Techcs.CV

  Video models adapt reasoning to frame evidence, slashing output length 62%

  EVAThink trains video understanding models to choose between direct answering and reasoning based on whether frames contain sufficient evidence.

  East China Normal University, Zhongguancun Academy · Nov 2026

• Top Universitycs.CL

  Rewarding precision reduces unnecessary and incorrect content in rubric-trained models

  Adding a precision objective to rubric-based rewards reduces verbosity and improves factual correctness without sacrificing coverage.

  KAIST, Carnegie Mellon University · Oct 2026

• Industrycs.CL

  Open-weights German-English model claims Pareto frontier for quality and serving cost

  Kolibri is an open-weight mixture-of-experts model on the Pareto frontier of quality and serving cost for English and German.

  Aleph Alpha · Oct 2026

• Big Techcs.CL

  Reinforcement learning yields user simulators that stay faithful and sound natural

  Reinforcement learning with a jointly trained discriminator produces user simulators that are both task-faithful and natural.

  Cornell University, Microsoft Research · Oct 2026

• Big Techcs.RO

  One Human Demonstration Can Seed Adaptable Dexterous Robot Skills

  Training one goal-conditioned policy with dense human-motion tracking and sparse off-reference goals yields dexterous skills that preserve functional behavior while adapting to new states and targets.

  Carnegie Mellon University, NVIDIA · Oct 2026

• Top Universitycs.CR

  Unverified state text can flip calibrated model decisions

  Appending an unverified opinion to the state flips 12.1% of decisions in RLCD models, making the state an untrusted input.

  Institute of Information Engineering, Chinese Academy of Sciences, University of Chinese Academy of Sciences · Sept 2026

• Chinese Tech

  Segment-specific rewards improve reinforcement learning for tool-calling agents

  Decoupling advantage estimation per segment resolves cross-segment credit misattribution in tool-calling RL.

  Peking University, Shenzhen University · Sept 2026

• Chinese Techcs.LG

  Elastic GPU scheduling makes reinforcement learning faster for marathon agent tasks

  QwenGyre enables online reinforcement learning for xlong-horizon LLM agents by elastic GPU reallocation and trajectory deduplication.

  Alibaba Token Hub, Alibaba Group, University of Science and Technology of China · Sept 2026

• Industrycs.AIcode

  AI reasoning models learn from reference examples alone, no reward needed

  An adversarially co-trained LLM judge can match verifiable reward signals for post-training reasoning models using only reference answers.

  Edison Scientific Inc. · Oct 2026

• Chinese Techcs.DC

  Language models train faster when generated samples arrive more evenly

  An offline scheduler that balances generation batching efficiency and trainer queue stability triples async LLM reinforcement learning throughput.

  École polytechnique, Huawei Technologies · Nov 2026

• Independentcs.AI

  A small AI agent trained on cloned websites matches bigger web agents

  A suite of 153 self-hosted replica websites with 3,518 tasks trains small web agents that then handle unseen tasks and match leading open agents.

  Nov 2026

• Top Universitycs.LG

  Small language models trained with reinforcement learning outperform 402B model in tool selection

  A Qwen3-8B model trained with reinforcement learning beats Llama-4-Maverick (402B) at tool selection and transfers to new tool corpora and retrievers.

  Meta, Carnegie Mellon University · Oct 2026

• Research Labcs.AI

  Jailbreak prompts improve uncertainty calibration in black-box reasoning models

  Jailbreak prompts, repurposed for uncertainty estimation, significantly improve confidence calibration in black-box large reasoning models.

  Petscraft, Inria · Sept 2026

• Big Techcs.CL

  Item response modeling improves rubric rewards while cutting judge requests

  Rubric Response Theory aggregates rubric criteria using item response theory to improve reinforcement learning rewards over additive methods.

  University of British Columbia, Microsoft · Sept 2026

• Chinese Techcs.LG

  Averaging rollout contexts improves label-free language model training

  Self-rewarding LLMs become more reliable when reward estimates are marginalized over random group contexts, improving zero-label self-evolution.

  Shanghai Jiao Tong University, Tencent · Sept 2026

• Chinese Techcs.AI

  Empathetic AI learns to adapt support priorities as conversations evolve

  Context-adaptive empathy rubrics improve reinforcement-learned dialogue systems by changing cognitive, affective, and proactive support priorities as conversations evolve.

  University of Science and Technology of China, SenseTime Research · Sept 2026

• Academiccs.LG

  Anchored hypergraphs provide stable credit assignment for multi-agent teams

  HySTAR uses a stable anchored hypergraph to separate credit assignment from adaptive representations, improving MARL performance across multiple benchmarks.

  University of Electronic Science and Technology of China, Independent Researcher · Sept 2026

• Chinese Techcs.LG

  Aligning policy and reward manifolds stabilizes LLM reasoning in RL

  DCRL aligns policy and reward manifolds to improve LLM reasoning, enabling a 4B model to surpass a 32B baseline.

  The Hong Kong University of Science and Technology (Guangzhou), Huawei Noah's Ark Lab · Sept 2026

• Chinese Techcs.LG

  Three diffusion RL methods unified as divergence-constrained reward maximizers

  DiffusionRFT uses exact sparsemax projection to converge faster, train more stably, and achieve top performance in regression-based diffusion RL.

  Tencent · Oct 2026

• Chinese Techcs.CV

  Post-training helps an agent sustain complex work for over 10 hours

  Marathoner achieves ultra-long-horizon autonomous execution with 1000+ tool calls over 10+ hours, beating proprietary models.

  FIC, University of Macau, Ant Group · Sept 2026

• Chinese Techcs.LG

  Off-policy sampling avoids suboptimal traps in reward-guided self-training

  For softmax multi-arm bandits, infrequently refreshed rollout data still converges at Θ(1/T) and can outperform on-policy REINFORCE from weak initial policies.

  Tsinghua University, Alibaba Group · Sept 2026

• Research Lab

  Reinforcement learning teaches an agent to verify mixed-source multimodal misinformation

  Reinforcement learning trains an LVLM to use a toolkit of specialized tools for detecting mixed-source multimodal misinformation, achieving 70.4% accuracy on MMFakeBench.

  Beijing University of Posts and Telecommunications, Minzu University of China · Sept 2026

• Chinese Techcs.CV

  AI models miss image details when attention drifts during reasoning

  Localization-Head Alignment raises a 4B thinking VLM's fine-grained perception accuracy from 74.1% to 76.6%.

  Tsinghua University, Ant Group · Oct 2026

• Big Techcs.AI

  Structured tags help AI chatbots ignore hidden instructions in data

  LLMON post-training reduces indirect prompt-injection attack success from 53-100% to 0.02-5.6%.

  IBM Research · Oct 2026

• Big Techcs.CL

  Aligning auxiliary rewards with task success improves multi-task language model training

  CONCUR projects gated auxiliary rewards onto scores that agree with the primary reward, improving multi-task LLM reinforcement learning.

  CMU, Google DeepMind · Oct 2026

• Top Universitycs.CV

  Persistent memory and reinforcement learning reduce drift in long driving simulations

  WAM-Sim uses persistent hierarchical memory and online flow-matching RL to generate consistent and realistic driving simulations over long horizons.

  Fudan University, Shenzhen Yinwang Intelligent Technology Co., Ltd · Nov 2026

• Chinese Techcs.AI

  Feedback criteria that track a model's growing ability halve its training steps

  Co-evolving a reward rubric alongside a policy by tracking its capability frontier improves and accelerates open-ended LLM training.

  University of Science and Technology of China, Alibaba Group · Oct 2026

• Big Techcs.LG

  Language models can produce better writing without sacrificing variety

  PRISM trains LLMs to produce diverse outputs using curated data and stochastic embeddings.

  Tsinghua University, Apple · Oct 2026

• Independentcs.AI

  CUA-Lite standardises computer use agent development across environments and models

  CUA-Lite is an open platform that standardizes environments, data, and model harnesses for computer use agents.

  Oct 2026

• Top Universitycs.CL

  Misspecified training rewards reduce the reliability of chain-of-thought monitoring

  Reward misspecification makes chain-of-thought poor at revealing reward hacking, while training models to describe inferred rewards substantially restores monitorability.

  Columbia University · Oct 2026

• Big Techcs.LG

  A simulator predicts when detectable reward hacking will dominate training

  SimHack predicts when detectable reward hacking will dominate coding RL runs before training, enabling curriculum choices that delay its onset by 96.6%.

  University of Illinois Urbana-Champaign, Google Research · Oct 2026

• Big Techcs.RO

  AI driving assistants reduce collisions by anticipating what drivers intend

  Extending world models with intent inference improves torque alignment and reduces collisions in shared vehicle control.

  Toyota Research Institute, Massachusetts Institute of Technology · Oct 2026

• Top Universitycs.AI

  AI agents can learn from imagined outcomes without generating images

  Dwell reframes a vision-language model as a recurrent latent transition operator to learn from experience.

  HKUST, National University of Singapore · Oct 2026

• Research Labcs.RO

  New memory architecture lets legged robots remember maze layouts while exploring

  A spatially indexed memory lets legged robots explore mazes end-to-end and remember the layout three times better.

  Technical University of Darmstadt, Robotics Institute Germany (RIG) · Oct 2026

• Independentcs.LG

  AI weather forecasts can capture extreme rain without blurring storm patterns

  FitzRoy uses reinforced diffusion to achieve state-of-the-art precipitation forecast skill and physical realism simultaneously.

  Oct 2026

• Big Techcs.RO

  Robots learn from human videos to handle unfamiliar tasks and environments

  Representing actions as chains of discrete motion tokens that are reinforced by how well they predict future video frames lets unlabeled human videos train robot manipulation policies.

  NVIDIA, KAIST · Oct 2026

• Chinese Techcs.CL

  Adaptive rubrics improve reinforcement learning for open-ended text generation

  Selectively replacing saturated rubric criteria with flaw-focused checks improves reinforcement learning for open-ended generation while preserving useful feedback.

  State Key Laboratory of AI Safety, Institute of Computing Technology, CAS, University of Chinese Academy of Sciences · Oct 2026

• Academiccs.LG

  Curriculum training finds prompt injections that direct attacks miss

  Curriculum reinforcement learning enables effective prompt injection red-teaming against frontier LLMs by overcoming the cold-start problem.

  The Pennsylvania State University, Purdue University · Sept 2026

• Research Labcs.AI

  Temporal distance improves self-supervised goal-directed reinforcement learning

  ChronoSRL gives critic embeddings temporal geometry, training distances to match goal-reaching times for faster and more reliable learning.

  Technical University of Darmstadt, Robotics Institute Germany (RIG) · Sept 2026

• Academiccs.CR

  AI agents fall for social engineering attacks in every tested scenario

  All frontier AI models fail to resist social engineering attacks, with proactive verification below 32% and many making ungrounded action claims.

  Hong Kong University of Science and Technology · Nov 2026

• Chinese Techcs.CL

  Adaptive training harness keeps rubric rewards useful as policies improve

  HarnessRL co-evolves task sampling, rollout guidance, and rubric criteria during RL training, beating baselines by 2.8-3.1 points on open-ended tasks.

  Alibaba Token Hub, Alibaba Group, Université de Montréal · Oct 2026

• Top Universitycs.LG

  Reinforcement learning can stably trap itself in low-return policies via representation superposition

  Self-confirming superposition traps allow optimal representation fitting to stabilize a lower-return policy in reinforcement learning.

  University of Cambridge, University of Sydney · Oct 2026

• Big Techcs.CV

  Self-grounded attention improves visual reasoning across diverse benchmarks

  Training VLMs to attend to objects they mention in reasoning improves visual reasoning accuracy.

  NVIDIA Research, The Hebrew University of Jerusalem · Oct 2026

• Chinese Techcs.LG

  Evidence annotations improve step-level credit for deep search agents

  Evidence-guided step-level process rewards derived from annotated evidence claims improve credit assignment and outperform standard GRPO on deep search benchmarks.

  Renmin University of China, Key Laboratory of Data Engineering and Knowledge Engineering, MOE · Oct 2026

• Big Techcs.LG

  Simple choices outperform complex recipes in iterative reward design benchmarks

  Simple design choices consistently improve iterative reward design, and combining them outperforms prior methods.

  Anthropic Fellows Program, University of Edinburgh · Oct 2026

• Chinese Techcs.AI

  Trajectory refinement improves reinforcement learning for long-horizon language-model agents

  TrajEvolve refines low-quality steps in historical trajectories and uses binary step-necessity classification for credit assignment, improving RL on long-horizon tasks.

  City University of Hong Kong, Tencent · Oct 2026

• Research Labcs.CL

  Accurate World Predictions Are Not Required for Agent Training Gains

  World model post-training improves agent performance through training effects other than accurate world prediction, as mismatched targets retain task gains despite lower accuracy.

  Xi’an Jiaotong University, University of Southern California · Sept 2026

• Chinese Techcs.CLcode

  Adaptive tutoring helps rerankers assemble better evidence sets for research

  AdaTutoRank, a setwise reranker with adaptive tutoring hints matched to rollout quality, tops ten RAG and deep research benchmarks while using fewer retrieval calls.

  University of Science and Technology of China, Yuanbao Team, Tencent · Sept 2026

• Chinese Techcs.CV

  Robotic world models learn by comparing diverse predicted futures

  FutureWorlds reduces LPIPS by 14.78%, 20.84%, and 9.12% on three datasets for 32-frame predictions.

  HKUST (GZ), CUHK · Oct 2026

• Chinese Techcs.AI

  Clinical agents falter when health records cannot support the question

  A benchmark of 5,486 noisy clinical queries shows LLM success drops from 62.2% to 37.9% when questions lack supporting EHR evidence.

  Zhejiang University, Ant Healthcare · Oct 2026

• Top Universitycs.AI

  Tailoring agent harnesses per task beats one-size-fits-all optimization

  Turbo Harness adapts global harnesses per instance using prior optimization artifacts, outperforming baselines across seven benchmarks.

  Rutgers University, Red Hat AI Innovation · Oct 2026

• Chinese Techcs.CL

  Separating planning from synthesis improves long-horizon search agents

  IterSynth separates planning and synthesis in deep search agents using a summary state, outperforming prior small agents.

  Zhejiang University, Tencent · Sept 2026

• Big Techcs.LG

  Small models can steer stronger ones through shared reasoning traces

  Weak teacher trained with RL improves strong student accuracy by alternating reasoning chains without strong model rollouts.

  UT Austin, Together AI · Oct 2026

• Independentcs.AI

  Preference learning improves collaboration among teams of language-model agents

  Multi-agent preference learning (MAPL) improves LLM collaboration quality and efficiency while approaching performance of MARL with fixed rewards.

  Sept 2026

• Big Techcs.IR

  Learned context curation boosts LLM catalog search accuracy

  Context curation and post-training with RPTune improves LLM catalog search accuracy by up to 31.4 percentage points.

  Google, University of Illinois Urbana-Champaign · Oct 2026

• Big Techcs.CV

  Language-Based Memory Helps Driving Models Track Earlier Road Events

  AD-Memo gives autonomous-driving VLA agents reusable language memory, improving memory-dependent driving, scene question answering, and cross-model portability.

  University of Illinois Urbana-Champaign, NVIDIA · Oct 2026

• Academiccs.AI

  Propagating reasoning states lets outcome labels improve process rewards

  Reasoning State Propagation propagates binary validity states through break and repair probabilities, combining process and outcome supervision.

  Southeast University, Key Laboratory of Computer Network and Information Integration (Southeast University) · Oct 2026

• Academiccs.AI

  Local clinical agents learn to consult remote models without identifying patients

  PrivMeSA reduces patient re-identification risk to 0% and improves task accuracy by up to 15.8 percentage points.

  University of Central Florida, Université de Montréal · Oct 2026

• Chinese Techcs.LGcode

  Diverse reward-model ensembles reduce reward hacking during language-model alignment

  Reward Model Boosting combines deliberately diverse reward models with a boosted tree aggregator, improving reward accuracy and reducing reward hacking in RLHF.

  Alberta Machine Intelligence Institute (Amii), University of Alberta · Sept 2026

• Independentcs.AI

  Reusable code skills help language agents act and learn efficiently

  Code-based skills nearly triple game progression and reduce inference cost by 86% in NetHack language agents.

  Sept 2026

• Chinese Techcs.CV

  Driving AI adapts reasoning to risk, not just complexity

  CAR-VLA adapts reasoning depth and focus using scene complexity and dynamic risk to achieve competitive driving performance.

  Fudan University, Yinwang Intelligent Technology Co., Ltd · Oct 2026

• Big Techcs.RO

  A single world model enables robots to insert unseen parts zero-shot

  A visual world model trained on up to 90 insertion tasks zero-shot assembles unseen objects with 56% success, far above a 7% model-free baseline.

  NVIDIA, University of California San Diego · Oct 2026

• Big Techcs.CL

  Eight-stage post-training recipe improves an open 106-billion-parameter language model

  A reproducible eight-stage post-training recipe improves the GLM-4.5-Air-Base model beyond its official release.

  Amazon · Sept 2026

• Research Labcs.CV

  Fine-Grained Rewards Reduce Vision-Language Hallucinations Without Silencing Valid Detail

  Fine-grained alignment using dense reward signals and subsentence-level credit assignment reduces object hallucination in vision-language models.

  State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences · Oct 2026

• Top Universitycs.AI

  Self-play on generated programs produces transferable language-model capabilities

  Self-play pretraining with synthetic data from a universal Turing machine yields predictable zero-shot scaling on natural data.

  Independent Researcher, Tel Aviv University · Sept 2026

• Independentcs.LG

  Mixed-granularity agent graphs improve collaboration across varied tasks

  MAGIC uses reinforcement learning to build mixed-granularity agent graphs that improve collaboration across subtasks.

  Sept 2026

• Big Techcs.AIcode

  Dynamic rubrics improve credit assignment for long-horizon agent training

  DRACO uses dynamic rubrics to assign per-step credit in GRPO, improving long-horizon agent training without ground-truth rewards.

  Carnegie Mellon University, IBM Research · Sept 2026

• Top Universitycs.AIcode

  Reinforcement learning strengthens policy invocation for agent safety judgments

  RePolicy learns to invoke safety policies for agent safeguards via reinforcement learning, achieving strong detection performance.

  Zhejiang University, Zhongguancun Academy · Aug 2026

• Top Universitycs.CV

  Factorized reinforcement learning improves vision-language models’ spatial reasoning

  FactoSR factorizes 4D spatial reasoning into planar, depth, and temporal objectives, achieving 5.9% and 4.5% improvements on VSI-Bench and All-Angles-Bench.

  Joy Future Academy, The Hong Kong University of Science and Technology (Guangzhou) · Sept 2026

• Chinese Techcs.CL

  Self-play with evolving skills boosts LLM benchmarks

  Skill Self-Play pushes LLM performance ceilings by co-evolving proposer, solver, and skill controller.

  Alibaba, The Chinese University of Hong Kong · July 2026

• Big Techcs.LG

  Dense credit assignment from gold-answer log-probabilities improves long-horizon agent RL

  TRACE assigns per-tool rewards from temporal-difference changes in frozen reference model answer log-probabilities, improving long-horizon agent search with pure RL.

  University of Wisconsin–Madison, Microsoft Research · July 2026

• Chinese Techcs.CV

  Unified visual-generation agentic model outperforms larger closed-source models

  A single end-to-end learnable agentic model combines understanding, thinking, planning, and creation for visual tasks, outperforming larger closed-source models on a new benchmark.

  Tencent Hunyuan, Hong Kong University of Science and Technology · Mar 2026

• Top Universitycs.LG

  Merging specialist AI models saves 42% of training compute compared to joint training

  On-policy distillation beats GRPO in compute efficiency only when using the student's own GRPO checkpoint as teacher.

  Salesforce AI Research, UIUC · Nov 2026

• Big Techcs.RO

  Humans intervene when robots stall, not when they err

  Adding optimistic bonuses to human intervention transitions improves robot learning efficiency by 10-50%.

  University of Washington, Toyota Research Institute · Nov 2026

• Top Universitycs.RO

  AI robot controllers raise average task success from 33% to 95%

  Geometric Support Behavior Regularization lets an online reinforcement learning policy improve on a pretrained policy by penalizing distance from support, raising average success from 33% to 95%.

  UC Berkeley, Physical Intelligence · Nov 2026

• Big Techcs.RO

  Simple prompt trick stops AI agents from faking progress in robot tasks

  Splitting robot reward estimation into a factual scene description followed by rubric-based scoring removes vision-language models' bias toward declaring failed tasks successful.

  University of Washington, NVIDIA · Nov 2026

• Big Techcs.RO

  Humanoid robots trained in simulation can navigate homes without prior mapping

  A humanoid robot trained only in simulation achieves 91% success performing physical tasks in unfamiliar real homes.

  Apple · Nov 2026

• Chinese Techcs.AI

  Language models can learn facts while misjudging what they know

  Large language models need explicit calibration of answerability judgments after knowledge updates to avoid confident wrong answers.

  City University of Hong Kong, Tencent Jarvis Lab · Nov 2026

• Independentcs.DC

  Fusing adapters into low-precision GEMMs cuts latency for AI serving and training

  EpiLoRA fuses LoRA adapters into low-precision GEMM kernels, cutting serving latency and enabling RL training on FP8 weights.

  Nov 2026

• Independentcs.AI

  New system halves compute waste during AI response generation training

  FlexFloor increases mixture-of-experts reinforcement-learning response throughput by 52.74% and occupied-compute efficiency by about 65.40% by releasing workers as their responses finish.

  Nov 2026

• Top Universitycs.LG

  Language models can change behavior without stronger training exposure signals

  Training membership signals in LLMs arise from how information is read out, not just what is internally encoded.

  Rutgers University, University of Bristol · Nov 2026

• Chinese Techcs.CV

  Video generators improve longer clips by revisiting earlier generation steps

  Using the intermediate denoising trajectory of prior chunks as context improves autoregressive long video generation quality.

  Huawei Noah Research Institute · Nov 2026

• Top Universitycs.CL

  Training AI from partially misaligned states makes behavior fixes last longer

  A new method, PEARL, keeps language models from relapsing into bad behaviors even after further training.

  Thinking Machines Lab, Princeton University · Nov 2026

• Chinese Techcs.SE

  A two-agent system finds bugs better than one by splitting search and inspection

  NINJA uses a navigator and inspectors to balance file-level and function-level search for better issue localization.

  Zhejiang University, DAMO Academy, Alibaba Group · Nov 2026

• Big Techcs.CL

  Language models favor suggested answers even when told the source is unreliable

  Reasoning LLMs shift answers toward hinted choices by a predictable constant log-odds shift, regardless of source credibility.

  Northwestern University, NSF-Simons AI Institute for the Sky (SkAI) · Oct 2026

• Big Techcs.CL

  AI jailbreak attacks more diverse with new training reward

  CoFiBreak trains jailbreak attackers to generate more varied successful attacks without sacrificing effectiveness or adding inference-time cost.

  Simon Fraser University, Microsoft Research · Nov 2026

• Research Labcs.LG

  Language models reason better when they think aloud only at key steps

  Selective verbalization of key reasoning tokens improves latent reasoning accuracy and efficiency.

  Beijing Institute of Technology, Institute of Computing Technology, Chinese Academy of Sciences · Nov 2026

• Chinese Techcs.RO

  AI models can reason better by predicting videos of outcomes

  RynnUnify is a 16B unified embodied model that combines understanding with long-horizon visual planning via interleaved video-language reasoning.

  DAMO Academy, Alibaba Group, Hupan Lab · Oct 2026

• Top Universitycs.RO

  Robots can learn better when humans correct deliberately introduced deviations

  Data Inoculation uses a shared-control framework to perturb demonstrations, improving downstream policy robustness.

  UC Berkeley · Oct 2026

• Big Techcs.RO

  Coding agents teach humanoid robots new whole-body skills without retraining

  A single frozen Metatron checkpoint supports walking, recovery, object pickup, and sitting on a real humanoid through code-defined prompts.

  NVIDIA, Dollhouse Robotics · Oct 2026

• Chinese Techcs.CV

  AI video assistants answer more accurately with better timing as events unfold

  SyncPro uses separate reasoning and response streams that activate independently to improve proactive video understanding timing and accuracy.

  Institute of Computing Technology, CAS, University of Chinese Academy of Sciences · Oct 2026

• Top Universitycs.CL

  Interventional step credit prunes redundant chain-of-thought while preserving accuracy

  Interventional step credit and IGPO reduce reasoning chain length by up to 62% while improving accuracy.

  Harbin Institute of Technology, Shenzhen, University of Aberdeen · Oct 2026

• Chinese Techcs.CV

  AI can trace objects’ outlines and flag when they’re absent

  A single vision-language model can handle detection, segmentation, matting, OCR, counting, and referring localization by predicting structured tokens plus a special object token that drives a pixel decoder.

  WeChat Vision, Tencent Inc. · Oct 2026

• Chinese Techcs.AI

  Large-scale dataset and world-model RL improve embodied navigation in open worlds

  Embodied world agents improve open-world navigation-planning via offline pretraining on a 200K-sample dataset and interactive reinforcement-learning post-training inside a video world model.

  Tencent Hunyuan · Oct 2026

• Independentcs.AI

  A few clean labels can fix sample-wise noise in reward model training

  Reward models trained on noisy human feedback can be made unbiased by calibrating sample-wise noise with a few clean labels via optimal transport.

  Oct 2026

• Big Techcs.AI

  Competition challenges AI models to optimize HVAC control across unfamiliar buildings

  A procedural HVAC benchmark measures whether one controller can transfer across unseen building designs and climates with minimal per-building configuration.

  Columbia University, Harvard University · Oct 2026

• Independentcs.CV

  Native 4K image editing works better as an agent system than a single model

  BluePixel combines an efficiently adapted native-4K editor with specialist routing and reward-based retries, delivering four-step 4K edits while ranking second on VINS-4KEval++.

  Oct 2026

• Chinese Techcs.LG

  Experience replay boosts RL training efficiency for large language models by reusing trajectories

  ReFlow reuses recent trajectories and dynamically reallocates resources to accelerate LLM reinforcement learning by up to 2.06 times without reducing model performance.

  The University of Hong Kong, University of Science and Technology of China · Oct 2026

• Big Techcs.LG

  JAX-based Atari environments enable GPU-accelerated training and easy modification

  JAXtari reimplements Atari environments in JAX, training PPO on 100 million environment steps in under an hour on one A100 GPU while exposing editable, object-centric states.

  Technical University Darmstadt, Google Intrinsic AI Research · Oct 2026

• Big Techcs.AI

  Learned memory assembly helps LLM agents succeed with fewer context tokens

  Learning to assemble memories with task success feedback improves LLM agent performance and reduces memory token usage.

  University of Illinois Urbana-Champaign, Amazon · Oct 2026

• Industrycs.AI

  AI agents cheat to maximize rewards, benchmark shows

  CheatBench is a benchmark that measures reward gaming in AI agents across multiple domains.

  Center for AI Safety · Oct 2026

• Big Techcs.RO

  Active stereo gaze enables precise manipulation without wrist cameras

  Active visual fixation using only a stereo camera matches or exceeds wrist camera performance.

  Amazon FAR, UC Berkeley · Oct 2026

• Chinese Techcs.LG

  Aligning FP4 training and rollout stabilizes reinforcement learning

  TRACE aligns FP4 training and rollout quantization for MoE RL, matching BF16 rollout performance while accelerating rollout by up to 5.4x.

  Alibaba Token Hub, Alibaba Group, Ohio State University · Oct 2026

• Chinese Techcs.AIcode

  Equivalent tool interfaces can make the same LLM agent fail

  LLM agents exhibit large performance differences across functionally equivalent tool schemas, with success rates ranging from 0% to 97%.

  University of Cambridge, Yinwang · Sept 2026

• Big Techcs.AI

  Models learn when to reason, cutting tokens while improving accuracy

  CounterRoute jointly learns routing and response modes in dual-model LMs, cutting generated tokens by 41-51% while improving accuracy.

  Amazon.com · Sept 2026

• Big Techcs.LG

  MEND selectively moves flow-model samples when reward gains justify distance

  MEND caps rewards per group and accepts a flow-model velocity update only when the capped reward gain exceeds a quadratic displacement price.

  The University of Texas at Austin, Google · Oct 2026

• Big Techcs.LG

  Reusable reasoning choices make multimodal search agents substantially faster

  Selection-based Structured Reasoning replaces freeform reasoning with selection from pre-specified candidates, cutting reasoning latency by over 90%.

  Apple, Carnegie Mellon University · Oct 2026

• Big Techcs.CL

  Real tumor board discussions expose gaps in medical language models

  OpenTumorBoard benchmarks LLMs on tumor board discussions; best models score 3.43/5 in clinical equivalence.

  Rice University, University of Illinois at Urbana-Champaign · Sept 2026

• Big Techcs.CV

  Canvas instructions enable precise image editing without separate text prompts.

  Image editing can be directed by drawing marks and notes on the image; VibeEdit outperforms text-prompted baselines on targeted edits among similar objects.

  University of Sydney, Microsoft Research · Oct 2026

• Chinese Techcs.LG

  Verification and repair reduce off-policy drift in long-tail RL rollouts

  RollVerify verifies and repairs partially generated rollouts to maintain accuracy while cutting training costs.

  Shanghai Jiao Tong University, Central South University · Oct 2026

• Chinese Techcs.CL

  Agents learn to reset noisy context during long web searches

  A rubric-driven search agent with self-managed memory and answer verification reaches 72.83 on BrowseComp and improves across several other search tasks.

  State Key Laboratory of Multimedia Information Processing, Peking University · Oct 2026

• Top Universitycs.LG

  Cross-feedback and step-wise scoring let small models match much larger reasoning datasets

  CRD-4B achieves 97.3% on MATH-500 and 70.3% on AIME'25 using only 50K training examples.

  KAIST · Oct 2026

• Chinese Techcs.RO

  Counterfactual post-training improves robot world model's action following and physical plausibility

  A robot world model using offline geometric calibration and counterfactual post-training improves action following and cuts human-assessed interaction defects from 48.12% to 6.25%.

  Institute of Automation, Chinese Academy of Sciences, Alibaba Group · Oct 2026

• Academiccs.CV

  Video RL improves by assigning credit to individual visual tokens

  TVRL uses gradients from a frozen vision-language model to allocate token-level credit, outperforming scalar-reward GRPO on video generation.

  Northeastern University · Oct 2026

• Big Techcs.CV

  One-step video refiner achieves 8.91x speedup for 4K upscaling

  SoL-Refiner transforms low-resolution video outputs into 4K videos with a single denoising step, outperforming multi-step refiners.

  NVIDIA · Oct 2026

• Top Universitycs.CL

  Proactive assistance requires choosing the right moment, not just the right words

  ProPerAgent improves proactive assistance by retrieving long-term personal memories, gating interruptions, and training against premature or repeated responses.

  Michigan State University, Meta Reality Labs · Oct 2026

• Big Techcs.LG

  Equal reward weighting can steer language models toward easier shortcuts

  Symmetric reward aggregation in LLM post-training degrades primary task performance by favoring easier auxiliary rewards.

  University of British Columbia, Amazon Annapurna Labs · Oct 2026

• Big Techcs.IR

  Co-Adaptive Routing and Retrieval Learns Query Rules Without Supervision

  CoRE jointly learns reasoning rule routing and constrained retrieval for structured knowledge without query-rule supervision.

  University of Georgia, Michigan State University · Oct 2026

• Big Techcs.AI

  Personal LLM agents improve recommendations using cross-platform history

  PAMO trains personal agents to selectively override platform rankings using cross-platform history, outperforming matched outcome-only reinforcement learning on accuracy and rescue-harm balance.

  University of California San Diego, Meta AI · Oct 2026

• Big Techcs.LG

  Joint on-policy learning improves self-distillation by aligning teacher with student

  A privileged teacher improves RL students only if its distillation update aligns with the student's reward gradient, motivating joint on-policy training.

  Perplexity, NVIDIA · Oct 2026

• Big Tech

  Privileged self-practice beats self-distillation for multi-turn language-model agents

  Privileged Self-Practice injects task-specific instructions into failing agent rollouts to improve multi-turn agent training.

  Texas A&M University, AWS AI, Amazon · Sept 2026

• Top Universitycs.CL

  Counterfactual credit teaches language models which memories stay useful

  MGPO isolates incremental value of memory rewrites via counterfactual credit, reducing memory length by 80% while improving extraction.

  University of Michigan · Sept 2026

• Big Techcs.LG

  Streaming KV caches makes agent context compaction faster to train

  KV-streams streams KV cache across compactions instead of re-prefilling, achieving 2.6x to 5x faster agent training without performance loss.

  Mila, Microsoft · Sept 2026

• Chinese Techcs.AI

  Post-training makes coding-agent trajectory verifiers substantially more accurate

  Post-training a 30B coding-trajectory verifier with supervised fine-tuning and reinforcement learning raises completion-classification accuracy from 57% to 78%.

  Leibniz Research Center Huawei, Monash University Malaysia · Oct 2026

• Big Techcs.CL

  Coordinated solution plans make parallel reasoning attempts less redundant

  Coordinated reasoning paths via planned outlines improve test-time scaling over repeated sampling.

  University of California, Los Angeles, Amazon · Sept 2026

• Independentcs.CL

  Agents Learn to Revise Their Harnesses on Unseen Tasks

  Harness learning trains a proposer to revise a solver's code-based harness using execution feedback, enabling test-time adaptation that generalizes to new tasks.

  Sept 2026

• Chinese Techcs.CL

  Latent reasoning cuts speech model response delay by roughly 12x

  A latent reasoning method that models multiple reasoning paths and predicts future states in chunks rivals explicit chain-of-thought in speech models while cutting time to first answer by 11.8x.

  The Chinese University of Hong Kong, Shenzhen, Tencent Hunyuan · Oct 2026

• Chinese Techcs.AI

  Training on self-generated algorithms helps design agents escape performance plateaus

  Population-Curated Policy Optimization enables self-evolving LLM agents to escape in-context evolutionary stagnation by learning from self-generated algorithms.

  Nanjing University, Huawei Noah’s Ark Lab · Oct 2026

• Chinese Techcs.AI

  Aligned skills and rubrics improve multimodal agent training

  Representing each skill as both actor-facing guidance and an evaluator-facing rubric, then co-evolving them, improves multimodal agent planning and tool use.

  The University of Hong Kong, Peking University · Sept 2026

• Big Techcs.LG

  Typed action encoding helps language models learn adaptable agent policies

  An architecture called Dyad adds a typed action encoder to LLMs, improving agent decision-making and general capabilities.

  Zhejiang University, SAP · Sept 2026

• Chinese Techcs.CL

  Selecting evidence before summarizing improves long-context reasoning efficiency

  Highlighting and summarizing relevant evidence before answering improves long-context reasoning and allows more compact output.

  Peking University, Baidu Inc. · Sept 2026

• Chinese Techcs.CV

  Flow matching plus physics yields more trackable 3D human motion from video

  FlowHMR uses flow matching with reinforcement learning to recover physically plausible global 3D human motion from monocular video, beating prior methods in fidelity and tracking.

  Peking University, The Hong Kong University of Science and Technology · Oct 2026

• Industrycs.CV

  Kandinsky 6.0 jointly generates short videos with synchronized sound

  Kandinsky 6.0 Video generates synchronized 5-second video and 44kHz audio, including lip-sync, from text or image prompts.

  Kandinsky Lab · Oct 2026

• Top Universitycs.CL

  Progressive pruning lets language-model debates beat equal-cost baselines

  Conditional Progressive Pruning makes multi-agent debate outperform consistency methods under the same computational cost.

  Rutgers University, New Brunswick, University of Cambridge · Oct 2026

• Big Techcs.AI

  Modeling pairwise task relationships improves multi-task recommendation systems

  Learning pairwise cross-task relationships improves multi-task recommendation accuracy and user satisfaction.

  Google · Oct 2026

• Big Techcs.CL

  AI judges favor reasoning models' stories far more than readers

  Reasoning-enhanced models do not consistently produce better creative writing, and LLM judges are strongly biased toward reasoning outputs.

  University of California, Los Angeles, Arizona State University · Nov 2026

• Big Techcs.LG

  Decoupling direction and distance improves on-policy distillation for language models

  D^2-OPD decouples extrapolation direction from distance in on-policy distillation, outperforming baselines by up to 9.45 points on math reasoning.

  Columbia University, AWS AI Lab · Nov 2026

• Chinese Techcs.LG

  Distilling AI reasoning works better when teacher rewards student's own learned math steps, not style

  Student-Delta On-Policy Distillation improves math reasoning by using a student's own post-training changes to guide teacher feedback.

  Zhejiang University, State Lab of CAD & CG · Nov 2026

• Big Techcs.AI

  AI agents follow attacks in documentation 2.6 times more than in code

  LLM agents treat some environment regions as more trustworthy than others, so indirect prompt injections succeed unevenly depending on where they are placed.

  University of Chicago, Microsoft Research · Oct 2026

• Chinese Techcs.CL

  Language models can lose math skills from more math training

  On-Policy Mid-Training Distillation improves language model math performance by training on student-generated trajectories instead of fixed prefixes.

  Fudan University, The Hong Kong Polytechnic University · Nov 2026

• Research Labcs.LG

  AI world models solve tasks without faithfully capturing how environments change

  Latent world models that achieve high task reward often fail to faithfully model the environment's dynamics.

  Helmholtz Munich, Munich Center for Machine Learning (MCML) · Oct 2026

• Big Techcs.AI

  Adversarial rubric repair improves self-training on open-ended expert tasks

  A reward hacking adversary supervises and repairs model-written rubrics, improving self-improvement on open-ended tasks.

  Massachusetts Institute of Technology, Microsoft Research · Oct 2026

• Chinese Techcs.LG

  AI agents can reuse successful decisions to learn with fewer tokens

  Replaying successful actions as branches inside rollout trees lets language-model agents learn long-horizon tasks with higher success and fewer rollout tokens.

  Alibaba Cloud Computing, Fudan University · Nov 2026

• Big Techcs.RO

  Hand-held skill interface enables portable data collection for dexterous robot hands

  DexPort, a portable skill-interface for dexterous hands, boosts task success rates from 11% to 64% over state-only baselines.

  Amazon FAR (Frontier AI & Robotics), Stanford University · Oct 2026

• Chinese Techcs.AI

  Targeting uncertain local decisions cuts GUI agent training cost by 56-61%

  A two-stage adaptive training framework for GUI agents reduces environment interaction by 56-61% while improving task performance.

  Alibaba Group, The Chinese University of Hong Kong · Oct 2026

• Big Techcs.LG

  Adaptive distillation triggers teacher takeover when student diverges

  Divergence-triggered teacher takeover during on-policy distillation stabilizes long-horizon training and accelerates learning from base initializations.

  Google, University of Washington · Oct 2026

• Top Universitycs.RO

  DexCast learns real-world dexterous control from human videos without teleoperation

  DexCast learns dexterous robotic hand control from monocular human videos without teleoperation, achieving 87.5% success on real-world tasks.

  Tsinghua University, TsingTec · Oct 2026

• AI Startupcs.AI

  Orchestrating coding agents broadens autonomous computational research beyond optimization

  The Agentic Scientist orchestrates coding agents with rubric verification to automate computational research and discovery.

  Sakana AI · Oct 2026

• Chinese Techcs.CV

  MaskSense preserves subject exploration when training masked image generators with RL

  MaskSense uses semantic priors and entropy anchoring to avoid a visual exploration trap in reinforcement learning for masked image generation.

  University of Science and Technology of China, Tongyi Lab, Alibaba Group · Oct 2026

• Chinese Techcs.SD

  Direct decision head makes bounded audio judgments about 20 times faster

  Qwen-Audio-Jev adds a dedicated decision head to an audio language model, directly outputting choices, binary judgments, and scores, with a 20x inference speedup.

  Token Foundry, Alibaba Group · Oct 2026

• Chinese Techcs.LG

  Query-adaptive neuron steering improves frozen language models across benchmarks

  Q-NeuS adaptively modulates neurons per query to boost LLM performance without retraining or task labels.

  Northeastern University, Washington University in St. Louis · Oct 2026

• Big Techcs.RO

  One Contact Skill Transfers Dexterous Manipulation Across Robot Hands

  A reusable cross-embodiment contact-acquisition primitive enables dexterous manipulation from object trajectories and reach specifications.

  Amazon FAR, UC San Diego · Oct 2026

• Big Techcs.AI

  Self-evolving visual agents learn tool use from unlabeled images using margin reward.

  A margin reward based on the solver's performance difference with and without tools enables visual agents to learn tool invocation from unlabeled images.

  UCAS, Microsoft Research Asia · Oct 2026

• Top Universitycs.AIcode

  Structural guidance improves language models’ long-horizon reasoning under sparse rewards

  SAGE improves sparse-reward, long-horizon LLM reasoning by pruning structurally unproductive branches and supplying dense geometric guidance across reasoning depth.

  Virginia Tech, University of Wisconsin Madison · Sept 2026

• Top Universitycs.AI

  Targeted feedback improves memory retrieval for language model agents

  Memory retrieval for LLM agents improves by learning set-level uplift compared to no-memory execution.

  MIIT Key Laboratory of Data and Decision Intelligence, Beihang University · Sept 2026

• Top Universitycs.LG

  Graph-based hindsight improves credit assignment for long-horizon language agents

  GraphHCA assigns step-level credit in long-horizon LLM agent tasks using closed-form hindsight ratios, achieving state-of-the-art results.

  Beihang University, Zhongguancun Academy · Sept 2026

• Chinese Techcs.LGcode

  Token-level teacher routing beats prompt-level routing in multi-teacher distillation

  Token-level ExpertAlign routing combines specialized teachers without domain labels and outperforms prompt-level routing and mean aggregation across four distillation settings.

  Shanghai Jiao Tong University, GAIR · Sept 2026

• Chinese Techcs.AI

  Safety training works better when models ignore superficial prompt cues

  DeShortcut-Align reduces over-refusal and improves robustness by decoupling spurious shortcuts in safety alignment of reasoning models.

  Zhejiang University, Ant Group · Sept 2026

• Big Techcs.CL

  Teacher alignment method closes capacity gap in reasoning distillation without discarding data

  TeacherGRPO adapts teacher models to student capacity via reinforcement learning, closing the reasoning distillation gap curse.

  University of Virginia, Netflix · Oct 2026

• Academiccs.CL

  Verified workflow training gives language models reusable procedural skills

  SkillGym converts human-written agent skills into executable training environments, enabling LLMs to internalize procedural competence and beat larger models on real-world benchmarks.

  East China Normal University, Shanghai AI Laboratory · Sept 2026

• Chinese Techcs.CL

  Dual-system voice agents balance live conversation with slower tool use

  Adaptive delegation between fast and slow reasoning enables full-duplex voice agents to balance real-time interaction with complex task accuracy.

  Tsinghua University, ByteDance · Sept 2026

• Big Techcs.LG

  Competitive self-play discovers compact motor skills humans can directly control

  Competitive self-play yields a small set of motor skills that humans can directly compose to solve new tasks without retraining.

  Georgia Institute of Technology, Simon Fraser University · Oct 2026

• Big Techcs.RO

  Humanoid tracker expands executable manipulation skills from sparse human demonstrations

  InterMimicGen self-evolves motion imitation data and policies to scale humanoid loco-manipulation from sparse human demonstrations.

  University of Illinois Urbana-Champaign, NVIDIA · Oct 2026

• Chinese Techcs.CV

  Fine-grained routing makes sparse video attention faster without quality loss

  VSA2 makes video DiT attention 8.9x faster with comparable quality by using a fine-grained router and adaptive sparsity.

  University of California, San Diego, ByteDance Seed · Oct 2026

• Big Techcs.AI

  Learning options by focusing on relevant features instead of whole states

  Identifying a small relevant subset of features per subgoal yields options that generalize broadly and accelerate exploration.

  Amazon, Brown University · Sept 2026

• Big Techcs.AI

  Reasoning improves when latent next thoughts remain a distribution

  Modeling each latent next thought as a multimodal distribution improves compact mathematical reasoning and preserves diverse solution paths better than point predictions.

  Rutgers University, Amazon · Sept 2026

• Research Labcs.AI

  Hierarchical memory trees improve long-context reasoning without model training

  ConvMem uses hierarchical convolution with strides, skip connections, and multi-kernel decomposition for parallel, training-free long-context reasoning.

  Institute of Automation, Chinese Academy of Sciences · Sept 2026

• Academiccs.LG

  Adaptive rollout trees broaden language models’ mathematical reasoning coverage

  Difficulty-adaptive sentence-entropy-guided tree-structured rollouts expand a large reasoning model's pass@k coverage more than parallel sampling during RLVR training.

  Pohang University of Science and Technology (POSTECH) · Sept 2026

• Chinese Techcs.AI

  Search agents improve longer when their critics evolve alongside them

  A self-improving search agent needs feedback that co-evolves with the policy it guides.

  Fudan University, Tencent · Aug 2026

• AI Startupcs.NE

  Evolution strategies balance adaptation and retention across changing tasks

  Neuroevolution, especially evolution strategies, outperforms reinforcement learning methods in continual task settings by achieving a better stability-plasticity trade-off.

  IT University of Copenhagen, University of Pisa · Oct 2026

• Chinese Techcs.CL

  Adaptive teacher intervention boosts distillation accuracy and stability

  Adaptive teacher intervention guided by policy disagreement improves accuracy and reduces training length in on-policy distillation.

  Nanyang Technological University, Baidu Inc. · Oct 2026

• Chinese Techcs.LG

  Ablating individual constraints improves distillation for complex instruction following

  CC-OPD improves multi-constraint instruction following by using per-constraint probability differentials as token-level rewards.

  Alibaba Group · Sept 2026

• Chinese Techcs.AI

  App deeplinks help mobile agents navigate faster and finish more tasks

  Combining app-native deeplinks with GUI actions lets mobile agents complete tasks more successfully and in fewer steps than using GUI actions alone.

  Shanghai Jiao Tong University, Tongyi Lab, Alibaba Group · Sept 2026

• Top Universitycs.AI

  Structured software interfaces outperform screenshot-and-click control for AI agents

  ASIL replaces screenshot-and-click with structured JSON observations and semantic actions, achieving over 80% success on 380 tasks with fewer than five actions each.

  X-LANCE Lab, Shanghai Jiao Tong University · Aug 2026

• Top Universitycs.CLcode

  Sliding reasoning windows make long test-time scaling 3x faster

  Prefix Sliding makes existing models 3x faster while maintaining performance during test-time scaling by keeping only the prefix and a recent window.

  Stanford University, University of California, Santa Barbara · Aug 2026

• Chinese Techcs.CV

  Adaptive tool use improves video research agents’ accuracy and efficiency

  AdaVDR adapts tool use and reflection in video deep research to cut costs and errors.

  Accio Team, Alibaba Group, Beijing Key Laboratory of Intelligent Information Technology · Aug 2026

• Academiccs.MAcode

  Reward-guided graph generation cuts multi-agent communication costs without hurting accuracy

  RGA-Designer reduces token consumption by 20.5% while preserving task accuracy in multi-agent communication topologies.

  University of Luxembourg, King Mongkut’s University of Technology Thonburi · Aug 2026

• Independentcs.CLcode

  A 2.8-trillion-parameter open MoE model approaches frontier performance.

  Kimi K3 is an open, 2.8T parameter Mixture-of-Experts model with 104B activated parameters that achieves frontier-level performance across multiple domains.

  July 2026

• Big Techcs.LG

  Geometry-aware LoRA optimization improves convergence across supervised and reinforcement learning

  RoM, a rotated manifold optimizer for LoRA, converges faster to lower held-out loss and achieves better or comparable downstream performance.

  ETH Zurich, Microsoft Research Cambridge · Oct 2026

• Top Universitycs.CR

  Persistent memory in AI agents can be stealthily poisoned via a single email

  Untrusted external content can be silently injected into persistent personal agent memory and later influence behavior without user awareness.

  Nanyang Technological University, CFAR, A*STAR · July 2026

• Big Techcs.AI

  Probabilistic scoring unlocks verification as a new scaling axis for LLMs

  Scoring token logit expectations, repeated evaluations, and decomposed criteria improve LLM verification accuracy and achieve state-of-the-art agentic benchmark results.

  Stanford University, UC Berkeley · July 2026

• AI Startupcs.LG

  Orchestrator models dynamically combine specialized LLMs into collective intelligence

  Fugu orchestrator models combine multiple LLMs into adaptive agent teams to achieve state-of-the-art results.

  Sakana AI · June 2026