Hanqi Yan

@yan_hanqi

Lecturer (assistant professor) @kclinformatics Interpretable | Reliable Language models.

London, United Kingdom

hanqi-qi.github.io

Joined September 2021

547 Following

710 Followers

Replies Reposts Media

Hanqi Yan
@yan_hanqi
Oct 18
🎉 Excited to share two recent works in collaboration with the Data Intelligence and Social Computing (DISC) Group at @FudanUniversity 📌 Paper 1: General Decision Models: Benchmarking and Insights Beyond Jev 🔗 arxiv.org/abs/2610.03935 How well do today’s LLM-based agents
arxiv.org
General Decision Models: Benchmarking and Insights Beyond Jev
General decision models, such as Jev, have recently emerged as efficient alternatives to LLMs for structured judgment and selection. But what kinds of decisions can these models reliably make, and...
1
Hanqi Yan
@yan_hanqi
Oct 18
🎉 Better late than never — a milestone worth celebrating regardless of when its shared! 100% acceptance rate across context compression, agent memory and mechanistic interpretability. 🏆 📌 Paper 1: Fix the Structural Bottleneck: Context Compression via Explicit Information
arxiv.org
Fix the Structural Bottleneck: Context Compression via Explicit...
Long-context LLM agents often struggle with growing token, memory, and latency costs, making efficient context compression essential for practical deployment. Existing LLM-as-a-compressor methods...
3
Hanqi Yan
@yan_hanqi
Jun 5
We asked: can LLMs hack society the same way they hack reward functions during RL training? The answer is yes. 🧵 Introducing SocioHack — 72 real-world societal environments where LLMs discover legal loopholes, exploit regulatory gaps, and stay technically compliant while
Wei Liu
@thinkwee2767
Jun 4
Large Language Models Hack Rewards, and Society arxiv.org/abs/2606.04075
8
Hanqi Yan
@yan_hanqi
Apr 28
🎉 @kclnlp led by @yulanhe is recruiting a 2-year postdoc in NLP/LLM! 🤖 🔬 Faithful reasoning, meta-reasoning with cognitive consideration to build reliable and impactful systems for high-stakes educational assessment. 🎓 🤝 Collaborated with AQA. 📅 Application deadline:
lnkd.in
LinkedIn
This link will take you to a page that’s not on LinkedIn
Hanqi Yan
@yan_hanqi
Apr 23
RAG alone isn't enough for agent memory 🧠 Our work xMemory tackles this through decoupling & aggregation — and it's been picked up by Claude Memory, PageIndex, and OpenClaw ⚡ Glad the direction is resonating 🌊 🔗 arxiv.org/abs/2602.02007 led by @HZhanghao
KCL NLP
@kclnlp
Apr 23
Our work on agent memory, xMemory (Beyond RAG for Agent Memory: Retrieval by Decoupling and Aggregation), has been discussed by projects such as Claude Memory, PageIndex, and OpenClaw. Glad to see this direction resonating. @HZhanghao @yulanhe @LinGui_KCL @dair_ai @VentureBeat

Hanqi Yan

Hanqi Yan’s posts