Read the paper. Follow the mechanism.

Persona Dosing: Calibrated Activation Steering for Graded Trait Control

Persona Dosing: Calibrated Activation Steering for Graded Trait Control · v1

Read and watch →

Self-Evolving Agents via Likelihood-Guided Tool-Space Optimization

Self-Evolving Agents via Likelihood-Guided Tool-Space Optimization · v1

Read and watch →

xDailyBench: Benchmarking LLMs on Professional Consultation for Real-Life Problems

xDailyBench: Benchmarking LLMs on Professional Consultation for Real-Life Problems · v1

Read and watch →

HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?

HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness? · v1

Read and watch →

Dr. Claw: An AI Scientist Workspace for Vibe Research

Dr. Claw: An AI Scientist Workspace for Vibe Research · v1

Read and watch →

Aspire: Can Models Self-Evolve from Vague Goals?

Aspire: Can Models Self-Evolve from Vague Goals? · v1

Read and watch →

S3Gym: Can LLMs Turn Self-Testing and Self-Judging into Self-Improvement?

S3Gym: Can LLMs Turn Self-Testing and Self-Judging into Self-Improvement? · v1

Read and watch →

WebWorld: The Browser as a World Model for Self-Improving Web Code

WebWorld: The Browser as a World Model for Self-Improving Web Code · v1

Read and watch →

VGI-Bench: Probing Visual Intelligence in Video Generation Models

VGI-Bench: Probing Visual Intelligence in Video Generation Models · v3

Read and watch →

StartupBench: Benchmarking General-Purpose Agents on Market-Validated End-to-End Workflows

StartupBench: Benchmarking General-Purpose Agents on Market-Validated End-to-End Workflows · v1

Read and watch →

MedClaw: Heuristic Agent Harness for Long-Horizon Surgical Video Reasoning

MedClaw: Heuristic Agent Harness for Long-Horizon Surgical Video Reasoning · v1

Read and watch →

Distributional Matching for Vector Quantization: A Unified Theoretical and Empirical Framework

Distributional Matching for Vector Quantization: A Unified Theoretical and Empirical Framework · v2

Enhancing Vector Quantization with Distributional Matching: A Theoretical and Empirical Study · v1

Read and watch →

Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models

Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models · v3

Read and watch →

LP-SFT: Keeping Plausible Alternatives Alive in Supervised Fine-Tuning

LP-SFT: Keeping Plausible Alternatives Alive in Supervised Fine-Tuning · v4

Read and watch →

Learning from the Self-future: On-policy Self-distillation for dLLMs

Learning from the Self-future: On-policy Self-distillation for dLLMs · v2

Read and watch →

CompRank: Efficient LLM Reranking via Token-Level Compression and Decoding-Free Scoring

CompRank: Efficient LLM Reranking via Token-Level Compression and Decoding-Free Scoring · v1

Read and watch →

Workflow-GYM: Towards Long-Horizon Evaluation of Computer-use Agentic Tasks in Real-World Professional Fields

Workflow-GYM: Towards Long-Horizon Evaluation of Computer-use Agentic Tasks in Real-World Professional Fields · v4

Read and watch →

OpenSkill: Open-World Self-Evolution for LLM Agents

OpenSkill: Open-World Self-Evolution for LLM Agents · v1

Read and watch →

Where, What, Why, and Importance: Structured Defect Grounding for Text-to-Image Feedback

Where, What, Why, and Importance: Structured Defect Grounding for Text-to-Image Feedback · v2

Read and watch →

MIRA: Mid-training Rubric Anchoring for Source-Aware Data Selection

MIRA: Mid-training Rubric Anchoring for Source-Aware Data Selection · v2

Read and watch →

RewardHarness: Learning Human Preferences for Image Editing with Only 100 Demonstrations

RewardHarness: Learning Human Preferences for Image Editing with Only 100 Demonstrations · COLM 2026

RewardHarness: Self-Evolving Agentic Post-Training · Related earlier version, arXiv v1

Read and watch →

ClawBench: Can AI Agents Complete Everyday Online Tasks?

ClawBench: Can AI Agents Complete Everyday Online Tasks? · v2

Read and watch →

Watch Before You Answer: Learning from Visually Grounded Post-Training

Watch Before You Answer: Learning from Visually Grounded Post-Training · v1

Read and watch →

Dr. Bench: A Multidimensional Evaluation for Deep Research Agents, from Answers to Reports

Dr. Bench: A Multidimensional Evaluation for Deep Research Agents, from Answers to Reports · v2

Read and watch →

VideoScore2: Think before You Score in Generative Video Evaluation

VideoScore2: Think before You Score in Generative Video Evaluation · v1

Read and watch →

IntentionReasoner: Facilitating Adaptive LLM Safeguards through Intent Reasoning and Selective Query Refinement

IntentionReasoner: Facilitating Adaptive LLM Safeguards through Intent Reasoning and Selective Query Refinement · v1

Read and watch →

WikiGap: Promoting Epistemic Equity by Surfacing Knowledge Gaps Between English Wikipedia and other Language Editions

WikiGap: Promoting Epistemic Equity by Surfacing Knowledge Gaps Between English Wikipedia and other Language Editions · v4

Read and watch →

StructEval: Benchmarking LLMs’ Capabilities to Generate Structural Outputs

StructEval: Benchmarking LLMs’ Capabilities to Generate Structural Outputs · v3

Read and watch →