Skip to content
Tagged articles

reinforcement learning

500 articles · Page 1 of 5
Frontline Investigation
Frontline Investigation
Nov 18, 2026 · Artificial Intelligence

Virtual Training Grounds: How Robots Learn Manipulation via Simulation & RL

This article explains how robots learn manipulation tasks through simulation-based reinforcement learning, detailing the role of physics engines, the observation-action-reward loop, domain randomization for sim-to-real transfer, and why training environments are becoming critical products in robotics.

Isaac LabMuJoCodomain randomization
0 likes · 11 min read
Virtual Training Grounds: How Robots Learn Manipulation via Simulation & RL
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Nov 18, 2026 · Artificial Intelligence

Athanor: AI Agents Autonomously Master LLM Post-Training, Matching Official Instruct Models

Duke PhD student Zhou Yufa presents Athanor, a system where four Opus 5.5 coding agents autonomously post-trained Qwen3-4B-Base on 8-GPU nodes for 104 hours, raising benchmark scores from 15.9 to 67.3 — matching Qwen's official Instruct model — without human guidance or teacher models.

AI agentsAthanorLLM post-training
0 likes · 5 min read
Athanor: AI Agents Autonomously Master LLM Post-Training, Matching Official Instruct Models
Machine Heart
Machine Heart
Nov 17, 2026 · Artificial Intelligence

HKU NeurIPS 2026: Implicit Advantage Symmetry Explains GRPO's Dual Bottlenecks

Hong Kong University researchers identify implicit advantage symmetry as the root cause of GRPO's poor exploration and difficulty adaptation, proposing A-GRAE to dynamically break symmetry and achieve simultaneous gains in Pass@1 accuracy and Pass@k diversity across math and multimodal benchmarks without extra compute.

A-GRAEGRPOLLM post-training
0 likes · 13 min read
HKU NeurIPS 2026: Implicit Advantage Symmetry Explains GRPO's Dual Bottlenecks
AI Frontier Lectures
AI Frontier Lectures
Nov 17, 2026 · Artificial Intelligence

Tencent Hy4 Flash Slashes Per-Task Cost 20x, Rivals Claude Haiku 5.5

Tencent's Hunyuan Hy4 Flash model reduces average per-task cost to 1/20th of Hy4 and 1/13th of Hy4 preview on internal WorkIndex benchmarks, while achieving satisfaction scores near Claude Haiku 5.5 at a fraction of the cost through mixed attention architecture, distillation, RL token reduction, and 4-bit inference acceleration.

Claude Haiku 5.5Hy4 FlashLLM cost optimization
0 likes · 7 min read
Tencent Hy4 Flash Slashes Per-Task Cost 20x, Rivals Claude Haiku 5.5
SmartApe AI
SmartApe AI
Nov 16, 2026 · Artificial Intelligence

CineEvolve: Self-Evolving Video Generation Lets AI Directors Call 'NG' on Storyboards

Researchers from BUPT, Peking University, and BAAI propose CineEvolve, a framework where a single visual language model acts as both director and critic to plan, diagnose, and revise storyboards via alternating reinforcement learning, achieving superior multi-shot consistency and narrative coherence in generated videos.

CineEvolveCineTraj datasetVideo Generation
0 likes · 11 min read
CineEvolve: Self-Evolving Video Generation Lets AI Directors Call 'NG' on Storyboards
Machine Heart
Machine Heart
Nov 16, 2026 · Artificial Intelligence

Axion 1.0: Training Powerful Agents with Only Two-Stage SFT and RL

ByteDance and university researchers introduce Axion 1.0, an agent training method using just two-stage supervised fine-tuning and reinforcement learning, achieving 59% on Humanity's Last Exam and 62.2% on SEAL-0 with fully open-sourced data and code.

Axion 1.0ByteDanceagent training
0 likes · 12 min read
Axion 1.0: Training Powerful Agents with Only Two-Stage SFT and RL
Meituan Technology Team
Meituan Technology Team
Nov 16, 2026 · Artificial Intelligence

EMNLP 2026 Highlights: 12 Long Papers from Meituan on Streaming Voice, RL, Agents & More

Meituan's technical team shares 12 EMNLP 2026 long papers covering streaming voice interaction, reinforcement learning with self-verification, instruction-following evaluation, agent skill internalization, on-policy distillation, Text-to-SQL over unknown schemas, unified memory frameworks, personalized reasoning, tool-use data synthesis, neuron consistency estimation, fast-slow reward models, and fine-grained gated delta networks.

EMNLP 2026On-Policy DistillationText-to-SQL
0 likes · 28 min read
EMNLP 2026 Highlights: 12 Long Papers from Meituan on Streaming Voice, RL, Agents & More
Machine Heart
Machine Heart
Nov 15, 2026 · Artificial Intelligence

SkillMaster: LLM Agents That Autonomously Evolve Their Own Skills via Counterfactual Rewards

Tsinghua University's SkillMaster framework teaches LLM agents to create, update, and retain reusable skills through reinforcement learning, using counterfactual utility rewards and DualAdv-GRPO to jointly optimize task execution and skill management, achieving 98.7% on ALFWorld and 82.0% on WebShop with Qwen2.5-7B.

DualAdv-GRPOLLM agentsNeurIPS 2026
0 likes · 11 min read
SkillMaster: LLM Agents That Autonomously Evolve Their Own Skills via Counterfactual Rewards
Machine Heart
Machine Heart
Nov 13, 2026 · Artificial Intelligence

NeurIPS 2026: HKUST's Visual-Native Agent Solves Image Reuse in Multimodal Deep Search

HKUST MMSense Lab introduces a visual-native agent framework that assigns unique IDs to images for tool chaining in multimodal deep search, and an on-policy data evolution method that iteratively refines question-generation rules using agent rollouts, boosting 8B and 30B model accuracy across eight benchmarks.

HKUST MMSense LabImage ReuseMultimodal Deep Search
0 likes · 20 min read
NeurIPS 2026: HKUST's Visual-Native Agent Solves Image Reuse in Multimodal Deep Search
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Nov 13, 2026 · Artificial Intelligence

FireFlash: 64 B300s, 5 Days, Matches Fable 5/GPT-6 Sol at 30x Lower Cost

FireFlash, a 300B-parameter terminal and coding agent trained via full-parameter RL on 64 NVIDIA B300 GPUs in under five days, achieves Terminal-Bench 4.0 Pass@1 of 46.2%—matching Fable 5 and GPT-6 Sol—while reducing per-task cost by over 30x, with gains transferring to DeepSWE and SWE-Bench benchmarks.

FireFlashGRPOLong-Horizon Agents
0 likes · 22 min read
FireFlash: 64 B300s, 5 Days, Matches Fable 5/GPT-6 Sol at 30x Lower Cost
PaperAgent
PaperAgent
Nov 12, 2026 · Artificial Intelligence

NVIDIA's Skill2Env and SoL-Pi: Collective Skills to Self-Evolving Agent Harnesses

NVIDIA's two new papers introduce Skill2Env, which compiles 3,400 practitioner skills into 7,971 verifiable training tasks across 13 domains, and SoL-Pi, an auto-research loop that cuts token usage by 49% and API costs by 33% while retaining 93.7% performance on agent benchmarks.

Agent HarnessAuto-ResearchCollective Skills
0 likes · 9 min read
NVIDIA's Skill2Env and SoL-Pi: Collective Skills to Self-Evolving Agent Harnesses
Machine Heart
Machine Heart
Nov 12, 2026 · Artificial Intelligence

FireFlash: Open Recipe for Long-Horizon Agent RL Matches Frontier at 30x Lower Cost

Researchers introduce FireFlash, a 300B-parameter terminal and coding agent trained via full-parameter RL from GLM-5.3-Flash on 64 B300 GPUs in under five days, achieving Terminal-Bench 4.0 Pass@1 of 46.2%—matching Fable 5 and GPT-6 Sol—while reducing per-task cost by over 30x through optimized infrastructure, task filtering, reward shaping, harness design, and a corrected GRPO objective.

GRPOLong-Horizon AgentsOpen Recipe
0 likes · 21 min read
FireFlash: Open Recipe for Long-Horizon Agent RL Matches Frontier at 30x Lower Cost
PaperAgent
PaperAgent
Nov 11, 2026 · Artificial Intelligence

EvoMemState: Self-Evolving Memory Paradigm for Personal Agents

Researchers from Xiamen and Jilin Universities propose EvoMemState, a state-centric memory paradigm for personal agents that evolves through ontology-guided induction, reinforcement learning, and state-aware reasoning, achieving 57.6-58.9% accuracy on MemSyco-Bench and 85.6% complete retrieval rate across benchmarks.

EvoMemStateLoCoMoLong-term Memory
0 likes · 10 min read
EvoMemState: Self-Evolving Memory Paradigm for Personal Agents
Machine Heart
Machine Heart
Nov 10, 2026 · Artificial Intelligence

UMM-Reflection: Unified Multimodal Models Self-Correct Images Without External Verifiers, 3x Repair Success

Researchers from NTU, SJTU, and UTokyo propose UMM-Reflection, enabling unified multimodal models to self-correct generated images through native reflection and reinforcement learning, achieving 64.9% repair success (up from 20.6%) and GenEval 0.84 (from 0.72) without external verifiers at inference.

BAGELGenEvalImage Generation
0 likes · 19 min read
UMM-Reflection: Unified Multimodal Models Self-Correct Images Without External Verifiers, 3x Repair Success
Machine Heart
Machine Heart
Nov 9, 2026 · Artificial Intelligence

Touch-R1: RL-Powered Tactile Reasoning Resolves Vision-Touch Conflicts (NeurIPS 2026)

Touch-R1 introduces the first tactile-anchored physical reasoning model using reinforcement learning to enable cross-sensor tactile inference, correcting visual misjudgments by prioritizing tactile evidence, and achieves state-of-the-art results on the new TouchReason-Bench benchmark across nine metrics including physical property identification, ordinal reasoning, and cross-sensor consistency.

Cross-Sensor ConsistencyEmbodied AINeurIPS 2026
0 likes · 10 min read
Touch-R1: RL-Powered Tactile Reasoning Resolves Vision-Touch Conflicts (NeurIPS 2026)
Machine Heart
Machine Heart
Nov 6, 2026 · Artificial Intelligence

WildAWM: Synthetic Computer Worlds Train Agents That Stump GPT-6 Astra

Researchers from Microsoft Research and universities introduce WildAWM, a framework that synthesizes complete personal computer environments with CLI, GUI, and MCP interfaces using only open-source models, generates cross-interface tasks validated through a five-gate process, and trains Qwen models that improve on eight out-of-distribution benchmarks, even creating tasks that challenge GPT-6 Astra.

Agent World ModelComputer-Use AgentCross-Interface Tasks
0 likes · 15 min read
WildAWM: Synthetic Computer Worlds Train Agents That Stump GPT-6 Astra
Machine Heart
Machine Heart
Nov 5, 2026 · Artificial Intelligence

ZTE's EmbodiedMind Cracks Long-Horizon Robot Planning with Trie-GRPO

ZTE's EmbodiedMind introduces a three-stage training paradigm (RSFT, IR-GRPO, Trie-GRPO) that enables efficient long-horizon planning for embodied agents, achieving 70.02% on 18 benchmarks with an 8B model and topping the SuperCLUE EmbodiedCLUE-VLA leaderboard with its 180B model, validated in simulation and on real robots.

Long-Horizon PlanningPrefix-Tree RLRobotics
0 likes · 16 min read
ZTE's EmbodiedMind Cracks Long-Horizon Robot Planning with Trie-GRPO
SmartApe AI
SmartApe AI
Nov 5, 2026 · Artificial Intelligence

Harness-S: Internalizing Complementary Harness Experience for Generalizable Agents

Researchers from Shanghai Jiao Tong University and Xiaohongshu propose Harness-S, a framework that evolves complementary execution harnesses, uses them for selective rescue training on failed tasks, and adapts harness selection via a policy-aware router, enabling agents to internalize diverse problem-solving strategies and retain them in the base framework.

Behavior TransferComplementary LearningExecution Frameworks
0 likes · 21 min read
Harness-S: Internalizing Complementary Harness Experience for Generalizable Agents
SmartApe AI
SmartApe AI
Nov 4, 2026 · Artificial Intelligence

RL-Driven Diffusion Models: From Generation to Decision-Making (Survey)

A 26-page survey from HUST, Tsinghua, and other institutions reviews how reinforcement learning integrates with diffusion models across language, vision, robotics, and scientific computing, covering methods like RLHF, DPO, GRPO, and challenges in reward design, exploration, and sparse rewards.

DPOGRPOMultimodal Generation
0 likes · 13 min read
RL-Driven Diffusion Models: From Generation to Decision-Making (Survey)
SmartApe AI
SmartApe AI
Nov 4, 2026 · Industry Insights

AMD AI VP Sharon Zhou Joins Anthropic, Cites 1000x RL Kernel Optimization Gain

Former AMD AI VP Sharon Zhou, who reported to CEO Lisa Su, joins Anthropic to return to frontier model research after pioneering reinforcement learning for GPU kernel optimization at AMD, claiming the approach reduces human effort for breakthroughs by 1000x; her career spans Harvard, Stanford under Andrew Ng, founding Lamini, and MIT TR35 recognition.

AI talent movementAMDAndrew Ng
0 likes · 6 min read
AMD AI VP Sharon Zhou Joins Anthropic, Cites 1000x RL Kernel Optimization Gain
Machine Heart
Machine Heart
Nov 3, 2026 · Artificial Intelligence

ReTerm: Code Agents That Learn to Recover from Their Own Failures

Researchers from Fudan University, CUHK MMLab, and Shanghai Jiao Tong University introduce ReTerm, a recursive self-improvement framework where code agents learn to diagnose and recover from their own terminal failures, turning self-generated error trajectories into training data that boosts 9B model performance on Terminal-Bench, SWE-bench, and OpenThoughts-TBLite by over 13 percentage points.

Code AgentsGRPOReTerm
0 likes · 9 min read
ReTerm: Code Agents That Learn to Recover from Their Own Failures
PaperAgent
PaperAgent
Nov 1, 2026 · Artificial Intelligence

CMU's Unified LLM Training: One Maximum Likelihood Objective for All Stages

CMU researchers propose a unified training paradigm where pre-training, mid-training, supervised fine-tuning, and reinforcement learning share a single maximum likelihood objective, using data mixing and scheduling instead of stage-specific algorithms, and show improved reasoning and continual learning on puzzles, math, and code benchmarks.

CMUContinual LearningLarge Language Models
0 likes · 9 min read
CMU's Unified LLM Training: One Maximum Likelihood Objective for All Stages
Meituan Technology Team
Meituan Technology Team
Nov 1, 2026 · Artificial Intelligence

LongCat-2.5: 1.6T Native Multimodal Model for Visual Coding & Long-Horizon Agents

LongCat-2.5 is a 1.6-trillion-parameter native multimodal mixture-of-experts model that activates 48B parameters per token, combining multi-level discrete visual tokens with large-scale long-horizon reinforcement learning to excel at visual coding, complex repository development, and creative 3D tasks, while achieving 86% higher training throughput on domestic AI accelerators and 43% faster inference via W4A4 quantization.

Long-Horizon AgentsLongCat-2.5Mixture of Experts
0 likes · 24 min read
LongCat-2.5: 1.6T Native Multimodal Model for Visual Coding & Long-Horizon Agents
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Oct 31, 2026 · Artificial Intelligence

World Models for Generalist Agents: From Sequential Decisions to Actionable Predictions

This article summarizes a paper introducing world models for generalist agents, framing them within sequential decision-making (POMDPs) and distinguishing prediction from action, covering control, estimation, policy learning, latent space dynamics, joint world-action modeling, and evaluation criteria for effective decision-making.

POMDPWorld Modelsgeneralist agents
0 likes · 32 min read
World Models for Generalist Agents: From Sequential Decisions to Actionable Predictions
SmartApe AI
SmartApe AI
Oct 31, 2026 · Artificial Intelligence

MIPU: New RL Method Bridges LLM Training-Inference Gap

Researchers from Tianjin University and Alibaba propose MIPU, a reinforcement learning method that addresses training-inference mismatch in LLMs by directly optimizing inference-time performance, achieving higher accuracy and more stable training on Qwen3 models across math benchmarks.

LLM trainingMIPI PrincipleMIPU
0 likes · 11 min read
MIPU: New RL Method Bridges LLM Training-Inference Gap
SmartApe AI
SmartApe AI
Oct 30, 2026 · Artificial Intelligence

ZwZ2: Single-Image Evidence Connection Boosts Multimodal Reasoning Across Tasks

Shanghai Jiao Tong University researchers introduce ZwZ2 and the EvidenceConnect framework, which synthesizes 20.4K single-image training samples requiring models to connect evidence across multiple regions, yielding 4B, 9B, and 27B models that surpass base models on 25 benchmarks and transfer gains to multi-image, spatial reasoning, and agent tasks.

Data SynthesisEvidenceConnectGRPO
0 likes · 14 min read
ZwZ2: Single-Image Evidence Connection Boosts Multimodal Reasoning Across Tasks
Machine Heart
Machine Heart
Oct 30, 2026 · Artificial Intelligence

ActionOPD: Distilling Agent Decisions, Not Tokens — Tsinghua's Action-Level Online Distillation

Tsinghua researchers propose ActionOPD, an action-level online distillation method that aligns student agents with teacher decisions at the semantic action level rather than token level, enabling black-box cross-family distillation and self-distillation with improved decision-making on benchmarks like τ²-bench.

ActionOPDagent distillationcross-family distillation
0 likes · 15 min read
ActionOPD: Distilling Agent Decisions, Not Tokens — Tsinghua's Action-Level Online Distillation
AI2ML AI to Machine Learning
AI2ML AI to Machine Learning
Oct 30, 2026 · Industry Insights

LLMs Automate Quant Research: From Factor Mining to Closed-Loop Portfolios

Cutting-edge AI projects now automate the most labor-intensive quant research steps — factor mining, synthesis, risk control, and portfolio management — using reinforcement learning, GFlowNets, and LLM-based closed-loop agents that target final portfolio performance, while independent audits reveal significant gaps between reported and replicable results.

AI in FinanceFactor InvestingGFlowNet
0 likes · 65 min read
LLMs Automate Quant Research: From Factor Mining to Closed-Loop Portfolios
SmartApe AI
SmartApe AI
Oct 30, 2026 · Artificial Intelligence

ClawLoop: Harness-Native Agent Self-Improvement Loop Without Stronger Teacher Models

Meituan and Peking University propose ClawLoop, a Harness-native framework closing the loop of task generation, model training, and diagnostic feedback, enabling agents to self-improve without stronger teacher models; experiments on Qwen3.6-27B show consistent gains across three benchmarks after three iterations, with SkillsBench rising from 46.3 to 57.2.

Agent Self-ImprovementClawLoopGRPO
0 likes · 11 min read
ClawLoop: Harness-Native Agent Self-Improvement Loop Without Stronger Teacher Models
Machine Heart
Machine Heart
Oct 29, 2026 · Artificial Intelligence

SimpleVLN: Native MLLMs Achieve Navigation SOTA Without Extra Modules

Fudan and NTU researchers propose SimpleVLN, a vision-language navigation framework that unlocks state-of-the-art performance by using native multimodal large language models without architectural changes, combining observation selection, action chunking, and a three-stage training pipeline for real-world robot deployment.

Action ChunkingDAggerMLLM
0 likes · 10 min read
SimpleVLN: Native MLLMs Achieve Navigation SOTA Without Extra Modules
Machine Heart
Machine Heart
Oct 29, 2026 · Artificial Intelligence

DiffusionART: 10x Faster Video Diffusion RL via Aggregation-First Reward Tilting

NUS and Kling AI researchers introduce DiffusionART, an online diffusion RL algorithm that eliminates forward-reverse consistency forces and uses a two-update scheme with trust-region masking to achieve 10-12x faster convergence than DiffusionNFT and Flow-GRPO-Fast on HunyuanVideo-13B, while supporting pairwise rewards with tiny group sizes.

Aggregation-FirstDiffusion RLDiffusionART
0 likes · 19 min read
DiffusionART: 10x Faster Video Diffusion RL via Aggregation-First Reward Tilting
Machine Heart
Machine Heart
Oct 28, 2026 · Artificial Intelligence

NV-CoT: Continuous Coordinates Boost Visual Chain-of-Thought (NeurIPS 2026 Oral)

NV-CoT replaces discrete tokenized bounding boxes with continuous numerical coordinate prediction via four lightweight linear heads and a Laplace policy for reinforcement learning, raising region localization IoU from 47.3 to 59.5 and improving high-resolution VQA accuracy on V* Bench and HR-Bench under both supervised fine-tuning and GRPO.

Continuous CoordinatesMultimodal Large ModelsNV-CoT
0 likes · 11 min read
NV-CoT: Continuous Coordinates Boost Visual Chain-of-Thought (NeurIPS 2026 Oral)
ITPUB
ITPUB
Oct 26, 2026 · Artificial Intelligence

Moonshot AI's Kimi K3.1 Pro Advances Intelligence Frontier While Curbing Cyberattack Capabilities

Moonshot AI releases Kimi K3.1 Pro, a 3-trillion-parameter model with 100k context, achieving near-top scores on coding, 3D creation, and financial benchmarks via massive RL post-training, while deliberately reducing cyberattack capabilities through targeted data removal, and planning open-weight release by November 4.

3D generationBenchmarkKimi K3.1 Pro
0 likes · 13 min read
Moonshot AI's Kimi K3.1 Pro Advances Intelligence Frontier While Curbing Cyberattack Capabilities
Machine Heart
Machine Heart
Oct 24, 2026 · Artificial Intelligence

VAST: Adaptive Horizon Selection for Long-Horizon RL via Value Stitching

Researchers from HKU and Tsinghua propose VAST, an offline RL framework that adaptively chooses time horizons for value updates by stitching variable-length trajectory segments, improving long-horizon decision-making across 50 OGBench tasks.

Adaptive Horizon SelectionLong-Horizon RLNeurIPS 2026
0 likes · 10 min read
VAST: Adaptive Horizon Selection for Long-Horizon RL via Value Stitching
Machine Heart
Machine Heart
Oct 23, 2026 · Artificial Intelligence

TRACE: Rollout-Guided FP4 Quantization Makes MoE RL Training Fast and Stable, 5.4× Speedup

Alibaba Qwen and Ohio State University propose TRACE, a rollout-guided quantization-aware training method that aligns training-side FP4 rounding with actual rollout quantization results, enabling stable FP4 W/A and KV cache rollout for MoE models with up to 5.4× decoding throughput gains and performance matching BF16 baselines.

Alibaba QwenFP4 quantizationMoE Models
0 likes · 9 min read
TRACE: Rollout-Guided FP4 Quantization Makes MoE RL Training Fast and Stable, 5.4× Speedup
SmartApe AI
SmartApe AI
Oct 23, 2026 · Artificial Intelligence

IQuest-Q1 Technical Deep Dive: 320B MoE Agent Model with 512K Context

IQuest Research releases technical report for IQuest-Q1, a 320B MoE model with 15B active parameters, 512K context window, recursive multi-token prediction, and agent-focused post-training, achieving strong benchmarks across coding, tool use, and reasoning tasks.

Mixture of ExpertsRecursive MTPagent training
0 likes · 16 min read
IQuest-Q1 Technical Deep Dive: 320B MoE Agent Model with 512K Context
Baidu Intelligent Cloud Tech Hub
Baidu Intelligent Cloud Tech Hub
Oct 22, 2026 · Artificial Intelligence

Baidu Baige Adapts RLinf v0.4: LEGO-style Robot RL, 2.37x Throughput, Sim-to-Real

Baidu Baige immediately adapted RLinf v0.4, an open-source reinforcement learning framework for embodied intelligence, providing pre-built images for three scenarios: Cosmos3 fine-tuning on LIBERO, Pi0.5 evaluation via ApxInf, and sim-to-real collaborative training using RL-Co, while detailing RLinf v0.4's modular robot interfaces, four new real-robot algorithms, expanded model support, realistic simulation benchmarks, and domestic chip compatibility, alongside Baige's full-chain AI Infra optimizations across data pipeline, training, simulation, and inference.

AI infrastructureBaidu BaigeCoRL 2026
0 likes · 16 min read
Baidu Baige Adapts RLinf v0.4: LEGO-style Robot RL, 2.37x Throughput, Sim-to-Real
Bilibili Tech
Bilibili Tech
Oct 19, 2026 · Artificial Intelligence

Index-Translate: Training a 150-Language Multimodal Translation Model Family

Bilibili's Index-Translate series introduces a family of multimodal translation models supporting 150 languages across three sizes, detailing their training recipe using WSD scheduling, multi-expert reinforcement learning with adversarial reward supervision, model merging and distillation, plus specialized models for speech translation, syllable-controlled dubbing, and long-document translation, with extensive benchmark results.

Index-Translatelong-document translationmodel merging
0 likes · 35 min read
Index-Translate: Training a 150-Language Multimodal Translation Model Family
SmartApe AI
SmartApe AI
Oct 19, 2026 · Artificial Intelligence

Terminal-RSI: 27B Agent Rivals Frontier Models via Recursive Self-Improvement

Tencent's Terminal-RSI framework uses recursive self-rewrite and policy refinement phases to train a Qwen3.8-27B terminal agent that achieves 86.52% on Terminal-Bench 2.1, surpassing GPT-6 Sol and Kimi K3, while demonstrating improved long-horizon task completion through credit assignment and reward design innovations.

Credit AssignmentLong-Horizon TasksQwen
0 likes · 13 min read
Terminal-RSI: 27B Agent Rivals Frontier Models via Recursive Self-Improvement
Machine Heart
Machine Heart
Oct 18, 2026 · Artificial Intelligence

DR-Rubric: Deep Research as Rubric for Open-Ended Reinforcement Learning

Xiaohongshu and Fudan University propose DR-Rubric, a framework where models perform deep research to gather evidence and synthesize atomic constraints as rubrics for reinforcement learning rewards, achieving competitive results on six benchmarks with only 1,000 training samples while revealing non-monotonic bootstrap dynamics and reward polarization as a stopping signal.

DR-RubricDeep ResearchGRPO
0 likes · 21 min read
DR-Rubric: Deep Research as Rubric for Open-Ended Reinforcement Learning
Rain's Path to Java Mastery
Rain's Path to Java Mastery
Oct 18, 2026 · Artificial Intelligence

ML Classification in One Diagram: Interview Answers & Java Implementation

This article explains the four machine learning paradigms—supervised, unsupervised, semi-supervised, and reinforcement learning—with a visual mind map, compares classification, regression, clustering, and dimensionality reduction tasks, provides a simulated interview Q&A covering K-Means pitfalls and core differences, and demonstrates a real-time text classification service using Java libraries Tribuo, ONNX, Spring Boot, and HanLP.

JavaSpring Bootinterview preparation
0 likes · 14 min read
ML Classification in One Diagram: Interview Answers & Java Implementation
Machine Heart
Machine Heart
Oct 18, 2026 · Artificial Intelligence

SPADER: Step-wise GRPO with Diversity-Aware Rewards Boosts Long-Tail Answer Discovery in Multi-Answer QA

SPADER introduces a step-wise GRPO framework with diversity-aware exploration rewards to improve long-tail answer coverage in multi-answer question answering, achieving state-of-the-art results across four benchmarks and two base models by refining credit assignment and incentivizing rare entity discovery.

Diversity-Aware RewardsEMNLP 2026GRPO
0 likes · 16 min read
SPADER: Step-wise GRPO with Diversity-Aware Rewards Boosts Long-Tail Answer Discovery in Multi-Answer QA
PaperAgent
PaperAgent
Oct 17, 2026 · Artificial Intelligence

DeepSeek-V4.1-Flash-1017 Open-Sourced: Self-Feedback RL Boosts Agent Performance

DeepSeek open-sources DeepSeek-V4.1-Flash-1017 with new RLSF (Reinforcement Learning with Self-Feedback) technique, showing across-the-board gains on 11 benchmarks, faster and shorter outputs, and demonstrating end-to-end project generation like 3D modeling and game development, all under MIT license.

AgentBenchmarkDeepSeek
0 likes · 9 min read
DeepSeek-V4.1-Flash-1017 Open-Sourced: Self-Feedback RL Boosts Agent Performance
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Oct 17, 2026 · Artificial Intelligence

12K GPU-Hour Reality Check: GRPO Outperforms 10 Self-Distillation Methods in Agentic RL

A large-scale empirical study comparing GRPO with 10 self-distillation methods across 4 tasks and 3 model sizes finds no method consistently beats GRPO; some methods show high final scores but suffer training instability, and compute-equivalent comparisons often reverse apparent gains.

GRPOReproducibilityagentic RL
0 likes · 10 min read
12K GPU-Hour Reality Check: GRPO Outperforms 10 Self-Distillation Methods in Agentic RL
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Oct 16, 2026 · Artificial Intelligence

A3 Framework: 14B CLI Agent Beats 235B Model via Structured Action Credit Assignment

Fudan and SJTU researchers open-source A3, a reinforcement learning framework that uses AST-based structured credit assignment to train a 14B CLI agent achieving 24.6% exact match on ShellOps, doubling baselines, and outperforming a 235B model on out-of-distribution tasks with minimal computational overhead.

CLI agentsCredit AssignmentNeurIPS 2026
0 likes · 18 min read
A3 Framework: 14B CLI Agent Beats 235B Model via Structured Action Credit Assignment
Xiaohongshu Tech REDtech
Xiaohongshu Tech REDtech
Oct 15, 2026 · Artificial Intelligence

CompoWorld: Compositional Environment Scaling Boosts Agent Performance by 9.17 Points

Xiaohongshu's AllSpark team introduces CompoWorld, a compositional environment scaling framework that combines 448 reusable services with 10,130 tools to train agents on cross-application workflows, achieving a 9.17-point average improvement across eight benchmarks and tripling success rates on AutomationBench.

AllSparkAutomationBenchCompoWorld
0 likes · 12 min read
CompoWorld: Compositional Environment Scaling Boosts Agent Performance by 9.17 Points
Data Party THU
Data Party THU
Oct 14, 2026 · Artificial Intelligence

Optimal Transport Meets RL: A Survey from Distribution Matching to Robust Decisions

This 72-page survey explores how optimal transport provides geometric tools for comparing distributions in reinforcement learning, covering imitation learning, policy optimization, model-based RL, offline RL, robustness, and transfer, while analyzing design choices in cost functions, temporal handling, and computational trade-offs.

Offline RLOptimal TransportPolicy Optimization
0 likes · 35 min read
Optimal Transport Meets RL: A Survey from Distribution Matching to Robust Decisions
Radish, Keep Going!
Radish, Keep Going!
Oct 14, 2026 · Industry Insights

Boston Dynamics: From Demo King to $1.3B Loss — Why the Pioneer Couldn't Sell Robots

Boston Dynamics, once the viral robot demo leader, accumulated $1.3B losses over five years under Hyundai after three ownership changes, as its hydraulic hardware and hand-crafted control algorithms proved too costly and fragile against China's mass-produced electric actuators and reinforcement learning pipelines.

Boston DynamicsDARPASpot robot
0 likes · 13 min read
Boston Dynamics: From Demo King to $1.3B Loss — Why the Pioneer Couldn't Sell Robots
Machine Heart
Machine Heart
Oct 14, 2026 · Artificial Intelligence

Jev-LDE: One Local Edit Boosts 1-Shot ICL Accuracy by 7 Points

Tsinghua researchers propose Jev-LDE, a 1.7B model that performs a single local edit (Keep, Delete, Replace) on retrieved demonstrations for in-context learning, improving average 1-shot classification accuracy from 81.2% to 88.1% across multiple datasets and target LLMs with only 11% inference overhead.

Demonstration SelectionFew-Shot ClassificationGRPO
0 likes · 17 min read
Jev-LDE: One Local Edit Boosts 1-Shot ICL Accuracy by 7 Points

Karpathy: Only 5,000 People Worldwide Access True Frontier AI Systems

Andrej Karpathy reveals a steep AI capability funnel where roughly 5,000 people globally use the most advanced internal AI systems — capable of week-long agent collaboration, zero-day discovery, and scientific research — while 6 billion have never touched an LLM and most professionals see only limited chatbot utility.

AI access disparityAI agentsAndrej Karpathy
0 likes · 8 min read
Karpathy: Only 5,000 People Worldwide Access True Frontier AI Systems
Xiaohongshu Tech REDtech
Xiaohongshu Tech REDtech
Oct 13, 2026 · Artificial Intelligence

AdaptEvo: From Rule Matching to Evidence-Based Contextual Decisions in AI Agents

Xiaohongshu's AllSpark team and university researchers propose AdaptEvo, a framework combining confidence-adaptive reinforcement learning with evolving decision knowledge to teach AI agents evidence-based contextual decision-making for rule-governed tasks like content moderation, outperforming Gemini 3.8-Flash and GRPO on both current and changed rules.

AdaptEvoAgent learningCA-GRPO
0 likes · 14 min read
AdaptEvo: From Rule Matching to Evidence-Based Contextual Decisions in AI Agents
PaperAgent
PaperAgent
Oct 13, 2026 · Artificial Intelligence

AEWM: Direct Agent Correction via Editing World Models

Renmin University introduces AEWM, an editing world model that classifies and revises noisy agent actions in real time, improving long-horizon task performance across search, terminal, and software engineering benchmarks without fabricating tool feedback.

AEWMAgent CorrectionEditAct
0 likes · 8 min read
AEWM: Direct Agent Correction via Editing World Models
Amap Tech
Amap Tech
Oct 12, 2026 · Artificial Intelligence

Amap's 16 NeurIPS 2026 Papers: Multimodal LLMs, Spatial Intelligence & Visual Reasoning

Amap had 16 papers accepted at NeurIPS 2026, covering multimodal large models, spatial intelligence, visual reasoning, and reinforcement learning, with highlights including a decomposed verifier for generative world modeling, gradient projection for continual multimodal learning, map-free transit route generation, and benchmarks for coding agents and travel planning.

3D reconstructionEmbodied AIMultimodal Large Models
0 likes · 30 min read
Amap's 16 NeurIPS 2026 Papers: Multimodal LLMs, Spatial Intelligence & Visual Reasoning
SmartApe AI
SmartApe AI
Oct 12, 2026 · Artificial Intelligence

HybridCUA: Teaching AI Agents When to Click and When to Command

Zhejiang University's HybridCUA framework enables computer-use agents to autonomously orchestrate GUI interactions and CLI commands through a unified bash action format, mixed-trajectory training data, and dual-reward reinforcement learning, achieving 53.6% accuracy on OSWorld with 56% fewer steps.

AI agentsComputer-Use AgentsGUI-CLI orchestration
0 likes · 11 min read
HybridCUA: Teaching AI Agents When to Click and When to Command
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Oct 11, 2026 · Artificial Intelligence

DeepSeek's Cost Curve Chronicle: How Token Economics Rewrote LLM Architecture

This article chronicles DeepSeek's paper timeline from 2023 to 2026, analyzing how each model generation optimized training compute, KV cache, and throughput via MoE, MLA, sparse attention, and conditional memory, shifting long-context competition from maximum length to per-token byte cost.

DeepSeekGRPOKV cache optimization
0 likes · 37 min read
DeepSeek's Cost Curve Chronicle: How Token Economics Rewrote LLM Architecture
PaperAgent
PaperAgent
Oct 11, 2026 · Artificial Intelligence

Two ACL 2026 Papers Solve Overthinking and Skill Forgetting in LLMs

This article summarizes two high-impact ACL 2026 papers: DRP reduces reasoning token usage by 64% while improving accuracy via skill-aware step pruning, and SAGE doubles skill reuse success rates in agents through sequential rollout and skill-integrated rewards.

ACL 2026DRPLLM reasoning
0 likes · 9 min read
Two ACL 2026 Papers Solve Overthinking and Skill Forgetting in LLMs
Machine Heart
Machine Heart
Oct 10, 2026 · Artificial Intelligence

Tian Yuandong: Co-Authoring a Paper with GPT-5 Made Me Realize My Job Could Vanish in 5 Years

Meta FAIR veteran and Recursive Superintelligence co-founder Tian Yuandong recounts his journey from AlexNet debates to Go AI, Coconut, and grokking research, revealing how co-authoring a paper with GPT-5 boosted efficiency 6-10x and led him to predict AI will replace researchers within five years, while arguing humans must still initiate and verify ideas.

AI research automationCoconutGPT-5 collaboration
0 likes · 15 min read
Tian Yuandong: Co-Authoring a Paper with GPT-5 Made Me Realize My Job Could Vanish in 5 Years
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
Oct 9, 2026 · Artificial Intelligence

Why Your 93% Agent Success Rate Collapses Under Scrutiny: The Evaluation Gap Between Saying and Doing

The article details a rigorous agent evaluation framework for after-sales operations using GRPO, revealing why a 93% overall success rate masks critical failures in specific scenarios and safety guardrails, and how to structure evaluations to withstand interviewer scrutiny.

Agent EvaluationBucket AnalysisGRPO
0 likes · 18 min read
Why Your 93% Agent Success Rate Collapses Under Scrutiny: The Evaluation Gap Between Saying and Doing
Machine Heart
Machine Heart
Oct 9, 2026 · Artificial Intelligence

VideoGen-Agent: 8B Agent Autonomously Calls Tools to Boost Video Generation

VideoGen-Agent, a Qwen3-VL-8B-based multimodal agent, learns via supervised fine-tuning and reinforcement learning to autonomously invoke retrieval, simulation, and verification tools for video generation, scoring 75.6 on VABench (19.1 points above base) and 86.1 with upgraded tools, while humans prefer its outputs 84.3% over Seedance 2.0.

Multimodal AgentsQwen3-VL-8BVABench
0 likes · 20 min read
VideoGen-Agent: 8B Agent Autonomously Calls Tools to Boost Video Generation
SmartApe AI
SmartApe AI
Oct 8, 2026 · Artificial Intelligence

Open-Source AI Social Coach Uses RL-Scheduled Scenarios and 3D Simulation for EQ Training

SocialCoach, an open-source AI system from HKUST(GZ) and USTC, simulates high-pressure social scenarios with timed responses, quotes users' exact words for post-practice review, and uses a Qwen3-8B reinforcement-learning scheduler to personalize practice sequences, achieving 4.22/5 goal-match score in evaluations.

AI coachingHuman-Computer InteractionQwen3-8B
0 likes · 7 min read
Open-Source AI Social Coach Uses RL-Scheduled Scenarios and 3D Simulation for EQ Training
Ops Development & AI Practice
Ops Development & AI Practice
Oct 8, 2026 · Artificial Intelligence

Why Xiaomi's MiMo Improved So Fast: Autonomous Training, Open-Source Misconceptions, and Talent Value Assessment

The article analyzes Xiaomi's MiMo model progress, clarifying its autonomous training foundation, distinguishing open-source reuse types, examining RL training costs, and evaluating talent impact versus commercial returns, arguing that technical output and organizational capability matter more than hype.

AI model evaluationLLM trainingMiMo
0 likes · 13 min read
Why Xiaomi's MiMo Improved So Fast: Autonomous Training, Open-Source Misconceptions, and Talent Value Assessment
Data Party THU
Data Party THU
Oct 3, 2026 · Artificial Intelligence

Agentic Visual Generation: A Five-Level Hierarchy of Controller Decision-Making

This survey introduces a five-level hierarchy (L0–L4) for agentic visual generation systems, classifying them by how far the controller can influence generation decisions—from fixed support to cross-task experience reuse—and proposes a level-conditioned evaluation protocol to isolate causal contributions of control scope.

L0-L4 taxonomyagentic visual generationcontroller hierarchy
0 likes · 35 min read
Agentic Visual Generation: A Five-Level Hierarchy of Controller Decision-Making
PaperAgent
PaperAgent
Oct 1, 2026 · Artificial Intelligence

ScienceBuddy: Open-Source AI Research Agent with Recursive Self-Improvement

ScienceBuddy is an open-source AI research agent that uses a recursive dual-loop architecture to continuously improve its harness and model weights from real user interactions, achieving 42.2% to 73.3% accuracy gains on scientific benchmarks across genomics, molecular biology, and pharmacology domains.

AI agentHarness AdaptationLAB-Bench
0 likes · 8 min read
ScienceBuddy: Open-Source AI Research Agent with Recursive Self-Improvement
Data Party THU
Data Party THU
Sep 29, 2026 · Industry Insights

Agility's Digit Humanoids Deploy in Warehouses, Target $2.5B SPAC IPO

Agility Robotics' Digit humanoid robots are already moving boxes in warehouses for customers like Amazon and Spanx, using learning-from-demonstration and large-scale simulation training to achieve adaptive manipulation, with the company claiming 20% lower operating costs than human labor over five years and preparing a $2.5 billion SPAC listing.

Agility RoboticsDigitHumanoid Robots
0 likes · 7 min read
Agility's Digit Humanoids Deploy in Warehouses, Target $2.5B SPAC IPO
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
Sep 29, 2026 · Interview Experience

GRPO Interview Mastery: From Critic-Free Design to Collapse Detection & Reward Hacking

This article breaks down six high-frequency GRPO interview questions from top Chinese tech companies, covering GRPO vs PPO trade-offs, group-relative advantage calculation with concrete numbers, handling all-correct/all-wrong sample groups, KL constraint mechanics, convergence monitoring priorities, and reward hacking detection via shadow evaluation.

Advantage EstimationGRPOKL Divergence
0 likes · 19 min read
GRPO Interview Mastery: From Critic-Free Design to Collapse Detection & Reward Hacking
Machine Heart
Machine Heart
Sep 28, 2026 · Artificial Intelligence

OmniVChat: Teaching Models Native Video Calls via Synthetic Data Generation

OmniVChat introduces a synthetic data pipeline (OmniVChat-Studio), benchmark (OmniVChat-Bench), and RL reward design (OmniVChat-RL) to train multimodal models for native audio-visual dialogue, achieving strong generalization from synthetic to real human interactions.

BenchmarkOmniVChatQwen-Omni
0 likes · 16 min read
OmniVChat: Teaching Models Native Video Calls via Synthetic Data Generation
Machine Heart
Machine Heart
Sep 25, 2026 · Artificial Intelligence

ULTRA: Unified Control for Humanoid Tracking & Goal-Driven Manipulation

ULTRA unifies motion tracking and sparse goal control in a single policy for humanoid robots, combining physics-driven motion retargeting, policy distillation, and RL fine-tuning, validated on Unitree G1 with first-person point cloud perception.

Sim-to-RealUnitree G1humanoid robotics
0 likes · 9 min read
ULTRA: Unified Control for Humanoid Tracking & Goal-Driven Manipulation
TonyBai
TonyBai
Sep 20, 2026 · Industry Insights

Jev's 'Breakthrough' Challenged: Researcher Claims Prior Art with Open-Source Laya

TypeSafe AI's Jev model, marketed as a breakthrough non-autoregressive decision model, faces controversy after independent researcher Nandakishor Mukkunnoth reveals he published similar work a year earlier and released open-source Laya outperforming Jev on benchmarks, sparking debate on Hacker News about priority, open source vs proprietary AI, and attention asymmetry.

Hacker NewsLayaTypeSafe AI
0 likes · 17 min read
Jev's 'Breakthrough' Challenged: Researcher Claims Prior Art with Open-Source Laya
Machine Heart
Machine Heart
Sep 20, 2026 · Artificial Intelligence

HSImul3R: Physics-in-the-Loop Reconstruction Turns Human Videos into Robot Skills

HSImul3R introduces a physics-in-the-loop framework that reconstructs simulation-ready human-scene interactions from sparse views, using scene-targeted reinforcement learning and direct simulation reward optimization to achieve stable physical interactions, validated on HSIBench and deployed on Unitree G1 robot.

3D reconstructionECCV 2026HSIBench
0 likes · 11 min read
HSImul3R: Physics-in-the-Loop Reconstruction Turns Human Videos into Robot Skills
Machine Heart
Machine Heart
Sep 20, 2026 · Artificial Intelligence

VBVR-Pro: 300 Visual Reasoning Tasks, Verifiable Rewards, and a Unified Training Framework

VBVR-Pro introduces a comprehensive framework for native visual reasoning, featuring 300 tasks, 1.25M training samples in video and interleaved formats, verifiable scorers for 100 tasks, benchmarking of 30+ models, and demonstration that verifiable rewards enable reinforcement learning to improve visual reasoning capabilities.

BenchmarkMultimodal ModelsVideo Generation
0 likes · 12 min read
VBVR-Pro: 300 Visual Reasoning Tasks, Verifiable Rewards, and a Unified Training Framework
UCloud Tech
UCloud Tech
Sep 16, 2026 · Artificial Intelligence

Train a $399 Microduck Robot with Reinforcement Learning on Cloud GPUs

This guide walks through training custom locomotion policies for the $399 Microduck bipedal robot using reinforcement learning in MuJoCo simulation on UCloud GPU instances, covering environment setup, walking and running policy training, ONNX export, simulation validation, and Hugging Face deployment.

GPU Cloud TrainingHugging FaceMicroduck
0 likes · 12 min read
Train a $399 Microduck Robot with Reinforcement Learning on Cloud GPUs
java1234
java1234
Sep 16, 2026 · Artificial Intelligence

Why LLM Post-Training Got Hard: 5 Paradigm Shifts in 6 Months

This article analyzes five major paradigm shifts in large model post-training over the past six months, covering expert distillation, online distillation as RL alternative, RLVR refinements, SFT-RL distribution mismatch, and data quality as irreducible constraint, with specific papers and metrics.

Large Language ModelsRLVRTraining Dynamics
0 likes · 11 min read
Why LLM Post-Training Got Hard: 5 Paradigm Shifts in 6 Months
Machine Heart
Machine Heart
Sep 15, 2026 · Artificial Intelligence

REAL: Embodied Agents Navigate Open Worlds Without Oracle Perception or Perfect Instructions

Researchers from Shanghai Jiao Tong University and Shanghai AI Lab introduce REAL, an ECCV 2026 framework that enables embodied agents to actively explore unknown environments, disambiguate vague user instructions through dialogue, and execute mobile manipulation tasks via a unified MCP tool interface, achieving 78.3% real-world success on a dual-arm robot.

Active ExplorationECCV 2026Embodied AI
0 likes · 19 min read
REAL: Embodied Agents Navigate Open Worlds Without Oracle Perception or Perfect Instructions
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Sep 14, 2026 · Artificial Intelligence

Astar: Alibaba & Zhejiang Univ's AI That Guides AI Evolution, Beating Human Experts 100x Faster

Alibaba and Zhejiang University's Astar learns from AI systems' own Git history to propose evolution strategies, achieving 54–68% single-shot success rates versus 31% for GPT-5.5 and 32% for human experts, and delivering 23.6% offline HitRate and 4.86% GMV gains in Lazada's ad system through 20 fully automated iterations.

AI evolutionAlibabaAstar
0 likes · 15 min read
Astar: Alibaba & Zhejiang Univ's AI That Guides AI Evolution, Beating Human Experts 100x Faster
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Sep 13, 2026 · Artificial Intelligence

PhyFilter: Physics-Based Filtering Enables Robot Generalization Beyond Data Scaling

Researchers from Beihang University and Nanyang Technological University propose PhyFilter, a lightweight plug-and-play module that uses physical differential structures and real-time feedback to correct learning residuals, allowing robots to generalize across unseen terrains and disturbances without massive datasets.

PhyFilterSim-to-Realacceleration estimation
0 likes · 17 min read
PhyFilter: Physics-Based Filtering Enables Robot Generalization Beyond Data Scaling
Java Tech Enthusiast
Java Tech Enthusiast
Sep 13, 2026 · Databases

MySQL 9.0 GA: JSON Multi-Valued Indexes, Parallel Query V2, and AI Optimizer Deep Dive

MySQL 9.0 introduces three major features: native JSON multi-valued indexes delivering 370x faster array queries, Parallel Query V2 with work-stealing achieving 7-8x speedups on analytical workloads, and an ML-based AI query optimizer that cuts index misselection from 12% to 2% and boosts complex query performance by 25-40%.

AI query optimizerDatabase PerformanceJSON multi-valued index
0 likes · 11 min read
MySQL 9.0 GA: JSON Multi-Valued Indexes, Parallel Query V2, and AI Optimizer Deep Dive
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Sep 12, 2026 · Artificial Intelligence

SafeEvolve: Co-Evolving Harness and Policy for Self-Improving Agent Safety

SafeEvolve introduces a co-evolution framework where Agent Harness and Policy jointly learn from execution trajectories, reducing attack success rates to 0.79% on AgentDojo and 2.42% on Qwen3-4B while improving task utility, enabling continuous safety improvement from real-world experience.

AI safetyAgent SafetyHarness-Policy Co-Evolution
0 likes · 9 min read
SafeEvolve: Co-Evolving Harness and Policy for Self-Improving Agent Safety
DataFunTalk
DataFunTalk
Sep 10, 2026 · Artificial Intelligence

CAPO-SLT: Confidence-Aware RL Fixes Fluent-but-Wrong Sign Language Translation

vivo AI Lab introduces CAPO-SLT, a confidence-aware policy optimization method that stabilizes sign language translation by giving each token a dynamic clipping bound based on old-policy confidence, achieving state-of-the-art results on Chinese and American sign language benchmarks using only pose input.

CSL-DailyEMNLP 2026How2Sign
0 likes · 14 min read
CAPO-SLT: Confidence-Aware RL Fixes Fluent-but-Wrong Sign Language Translation
vivo Internet Technology
vivo Internet Technology
Sep 9, 2026 · Artificial Intelligence

SmartPhotoCrafter: Think-First Photo Enhancement Unifies Restoration & Retouching

SmartPhotoCrafter introduces a unified reasoning-to-generation framework for automatic photo enhancement: an Image Critic analyzes quality via chain-of-thought reasoning, then a Photographic Artist performs high-fidelity restoration and aesthetic retouching, unifying dehazing, deblurring, and color grading without explicit user instructions, trained via a three-stage strategy with multi-level rewards.

Chain-of-ThoughtSmartPhotoCrafterautomatic photo enhancement
0 likes · 13 min read
SmartPhotoCrafter: Think-First Photo Enhancement Unifies Restoration & Retouching
Huawei Cloud Developer Alliance
Huawei Cloud Developer Alliance
Sep 8, 2026 · Artificial Intelligence

AReaL v1.0.5 LoRA RL: Low-Rank Adaptation for Accessible Large Model RL on Ascend

This article details AReaL-Ascend v1.0.5's LoRA RL capabilities, explaining how low-rank adaptation reduces memory overhead for large model reinforcement learning, describing two Megatron LoRA weight update modes (adapter sync vs. merge), and covering cross-node LoRA RL, MoE support, XCCL communication, and Qwen3.6-27B examples for practical deployment.

AReaLAscend NPULoRA
0 likes · 6 min read
AReaL v1.0.5 LoRA RL: Low-Rank Adaptation for Accessible Large Model RL on Ascend
Machine Heart
Machine Heart
Sep 8, 2026 · Artificial Intelligence

Behavior Consistency Beats State Consistency in Text World Models for Agents

The paper introduces BehR, a behavior consistency reward for training text-based world models, showing that optimizing for agent decision alignment rather than text fidelity improves trajectory-level consistency across 16 configurations, reduces false positives in offline evaluation from 42.5% to 9.5%, and enhances lookahead planning for weaker agents.

Behavior ConsistencyEMNLP 2026GRPO
0 likes · 11 min read
Behavior Consistency Beats State Consistency in Text World Models for Agents
Huawei Cloud Developer Alliance
Huawei Cloud Developer Alliance
Sep 6, 2026 · Artificial Intelligence

AReaL v1.0.5: Colocated Training/Inference & Multi-Teacher On-Policy Distillation for Efficient RL

AReaL-Ascend v1.0.5 introduces two major innovations: colocated training and inference on shared NPUs via Ray scheduling and AWEX IPC zero-copy, and Multi-Teacher On-Policy Distillation (MOPD) that fuses domain expert models into a single student using token-level teacher signals, demonstrated on Qwen3.6-27B with GRPO.

AReaLAWEXAscend NPU
0 likes · 12 min read
AReaL v1.0.5: Colocated Training/Inference & Multi-Teacher On-Policy Distillation for Efficient RL
Alibaba Cloud Native
Alibaba Cloud Native
Sep 5, 2026 · Artificial Intelligence

DeepSeek Harness: Architecting Enterprise Evolution via Post-Training Design

This article analyzes DeepSeek Harness from a post-training perspective, revealing how its architecture bakes enterprise evolution into environment shaping and trajectory sedimentation, validated by a self-evolution POC that identifies interface design, contract shape, and feedback quality as critical bottlenecks.

DeepSeek HarnessPoC validationagent architecture
0 likes · 37 min read
DeepSeek Harness: Architecting Enterprise Evolution via Post-Training Design
Alimama Tech
Alimama Tech
Sep 3, 2026 · Artificial Intelligence

MOON 3.0: Attribute Reasoning for E-commerce Multimodal Representation (ACM MM'26)

MOON 3.0 shifts e-commerce multimodal representation from direct encoding to explicit attribute reasoning, using Multi-head Modality Fusion, joint contrastive and reinforcement learning, and fine-grained residual enhancement to improve fine-grained retrieval and classification across multiple benchmarks.

Contrastive LearningMLLMMOON 3.0
0 likes · 24 min read
MOON 3.0: Attribute Reasoning for E-commerce Multimodal Representation (ACM MM'26)
Machine Heart
Machine Heart
Sep 3, 2026 · Artificial Intelligence

EASE: Teaching Multimodal RL Where to Look, Not Just What to Answer

EASE introduces evidence-anchored spatial attention supervision to multimodal reinforcement learning with verifiable rewards, using annotated evidence bounding boxes to guide model attention toward relevant image regions during training, improving accuracy on visual reasoning benchmarks without inference overhead.

Attention MechanismEMNLP 2026Evidence Grounding
0 likes · 13 min read
EASE: Teaching Multimodal RL Where to Look, Not Just What to Answer
Machine Heart
Machine Heart
Sep 2, 2026 · Artificial Intelligence

Facet-0 Enables Robots to See, Insert Precisely, and Recover from Errors in Precise Assembly

The NTU PINE Lab introduces Facet-0, a multimodal robot foundation model that achieves 82% success in five real computer‑assembly tasks with 0.5 mm placement accuracy, reduces human intervention from 47% to 24%, and learns to recover from contact failures using a force‑synchronized dataset and reinforcement‑learning post‑training.

ManuFacet-1KNTURobotics
0 likes · 11 min read
Facet-0 Enables Robots to See, Insert Precisely, and Recover from Errors in Precise Assembly
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Sep 2, 2026 · Artificial Intelligence

How UniSteer Boosted VLA Success from 20% to 90% in Just 66 Minutes

UniSteer introduces a noise‑inversion interface that lets human corrections directly train a lightweight noise actor, enabling a Vision‑Language‑Action model to improve real‑world task success from 20% to 90% within 66 minutes and outperforming DSRL and DAgger baselines.

Human-Guided RLNoise InversionRobotics
0 likes · 15 min read
How UniSteer Boosted VLA Success from 20% to 90% in Just 66 Minutes