
Intern @UW | Undergraduate @UniversityLeeds | NLP | AI | Agent
Joined June 2026
- Tweets10
- Following140
- Followers23
- Likes341
Han Luo retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Agentic Abstention has been accepted to #NeurIPS2026.
We introduce a consolidated benchmark for evaluating whether agents know when to stop rather than continue acting when a task is underspecified or cannot be completed in the current environment.
It was amazing to co-lead this project with @bingbingwen1, with the guidance and support of @lucyluwang 🌟
🎉Excited to share our preprint:
Agentic Abstention: Do Agents Know When to Stop Instead of Act?
LLM agents are increasingly capable of using tools, browsing the web, and interacting with external environments. But when should an agent stop acting instead of taking more actions? 🧵1/n
🧬 We also introduce CONVOLVE, a context engineering method that distills full interaction trajectories into reusable stopping rules, that help to improve timely abstention without requiring parameter updates.
1
1
72
🌟Special thanks to @bingbingwen1 and @lucyluwang for guidance and support throughout the entire project!!!
📖 Paper: arxiv.org/abs/2606.28733
🔗 Project page: lhannnn.github.io/agentic-ab…
1
41
Han Luo retweeted
Agentic Abstention
Do LLM agents know when to stop instead of act?
We evaluate 13 agent systems across 28K+ tasks in web, terminal, and QA environments.
Most agents abstain too late, if they abstain at all.
CONVOLVE improves timely abstention from 26.7% to 57.4% without any model fine-tuning.








