Wei Xu shares talks on multilingual LLMs and rollout diversity for GRPO-style RL
cocoweixu · x · 2026-09-28
Researcher Wei Xu (cocoweixu) recapped her recent talks covering two themes:
- Multilingual and multicultural LLMs: how models handle diverse languages and cultural contexts.
- Reinforcement learning for diversity: showing how rollout diversity improves GRPO-style RL training.
She is also preparing a new talk for a NeurIPS 2026 workshop, titled "TL;DR for LLMs: Agent Meets Checklist on Long-Context Tasks," focusing on agents and checklists for long-context tasks.
More from Research
- Stanford study: GPT-4 alone outdiagnoses doctors using GPT-4 — jonc101x · 2026-09-28
- Meta's TRIBE v2: Tri-modal foundation model predicts human brain activity from 1,000+ hours of fMRI — burny_tech · 2026-09-28
- Laya replaces LLM-as-a-judge with a 322M decision engine — 26,639 stars in 9 days — AIFrontierReads · 2026-09-28
- Reconstructive Identity: LLMs Link Weak Signals to Deanonymize at 68% Recall — AmuzedX · 2026-09-28
- Spectral deflation framework improves Muon: consistent validation loss gains in GPT-2 pretraining — hankyang94 · 2026-09-28
- CMU's DeformX trains robots to whip ropes in sim — UR5e knocks apple off a head with 0cm error — DJiafei · 2026-09-28