Blog

Welcome to my blog! Here I share insights from my daily readings of AI research papers and write about interesting open-source projects I’m working on. I hope these posts are helpful to the community and spark interesting discussions.

AI Daily Paper Summary

I read AI and ML research papers daily and post concise summaries here. Each entry covers the key contributions, methods, and takeaways from recent papers across topics like deep learning, reinforcement learning, NLP, computer vision, and more.

Latest — July 24, 2026

1. AREX — A recursively self-improving deep-research agent that alternates evidence-gathering with constraint-by-constraint auditing, beating baselines on BrowseComp and Humanity’s Last Exam. arxiv.org/abs/2607.21461
2. ReferTrack — A refer-then-track paradigm for embodied visual tracking, hitting SOTA on EVT-Bench with real sim-to-real transfer on legged and humanoid robots. arxiv.org/abs/2607.20061
3. K12-KGraph — A curriculum-aligned knowledge graph from K-12 textbooks powering a new benchmark and training corpus for educational LLMs. arxiv.org/abs/2605.09635
4. Visual Contrastive Self-Distillation — A teacher-free on-policy self-distillation recipe that meaningfully lifts Qwen3-VL benchmark scores at no extra inference cost. arxiv.org/abs/2607.21556
5. Show, Don’t Tell — Proposes evaluating spatial cognition through generated pixels rather than text/coordinates, finding image-gen models competitive when allowed to “show” their answer. arxiv.org/abs/2607.21072
6. Tencent WorkBuddy Bench — A contamination-resistant, multi-domain coding-agent benchmark built from real commits/PRs rewritten as role-played tasks. arxiv.org/abs/2607.20911
7. NVIDIA Object-Oriented Agents (NOOA) — A Python framework where agents are plain objects whose “…” method bodies are completed at runtime by an LLM loop. arxiv.org/abs/2607.20709
8. Color Pass-Through — An end-to-end learned camera-display coupling that fixes long-standing capture-to-screen color fidelity mismatches. arxiv.org/abs/2607.12746
9. LLMs Get Lost in Evolving User Intent — Converts static benchmarks into multi-turn conversations with shifting intent, showing today’s LLMs don’t reliably track it. arxiv.org/abs/2607.20734
10. Structured Dynamics Model — Self-supervised video learning that disentangles camera motion from object motion in frozen ViT features. arxiv.org/abs/2607.21576

Browse all paper summaries →

Interesting Projects

In this section, I share outcomes, results, and lessons learned from open-source projects I’m working on. The goal is to give back to the community and document the journey of building and experimenting with different ideas.

Browse all project posts →