hyunji amy lee

@hyunji_amy_lee

postdoc @unc_ai_group w/ @mohitban47. PhD @kaist_ai. Previously: @allen_ai @adobe @MSFTResearch.

following
followers

LookAhead Defense aims to prevent fine-tuning from activating dormant poisoning in AI agents

Its creators say benign fine-tuning can trigger harmful agent actions, including leaking private information through email tools.

10 Sources, 7h ago, updated 7h ago

Hybrid language models reportedly rely heavily on attention, even after supervised fine-tuning

The author reports that encouraging recurrent memory use improved performance, especially on long-context tasks.

8 Sources, 14d ago, updated 14d ago

MINTEval benchmark for AI memory systems accepted to NeurIPS 2026

MINTEval’s creators report that seven representative systems averaged 27.9% accuracy on its changing-context tasks.

4 Sources, 19d ago, updated 19d ago

Adaptive-View Self-Distillation accepted to NeurIPS 2026

A researcher behind AVSD says it combines multiple teacher views, including hints, solutions and execution outputs, to improve token-level learning signals for LLM math and code reasoning.

5 Sources, 20d ago, updated 20d ago

Featured in

LookAhead Defense aims to prevent fine-tuning from activating dormant poisoning in AI agents

Hybrid language models reportedly rely heavily on attention, even after supervised fine-tuning

MINTEval benchmark for AI memory systems accepted to NeurIPS 2026

Adaptive-View Self-Distillation accepted to NeurIPS 2026

Featured in

LookAhead Defense aims to prevent fine-tuning from activating dormant poisoning in AI agents

Hybrid language models reportedly rely heavily on attention, even after supervised fine-tuning

MINTEval benchmark for AI memory systems accepted to NeurIPS 2026

Adaptive-View Self-Distillation accepted to NeurIPS 2026