ACL ARR and EMNLP 2025 does not notify co-authors when one of them fails to complete their reviewing duty, unlike NeurIPS or ACM MM, where all co-authors are informed and can follow up.

Without such notice, first-author students, who bear the heaviest consequences, are punished for circumstances completely outside their control.

Neither the conference nor ARR informs the first author which co-author has been assigned reviews or who has failed to complete them. This is especially unfair to first authors who have diligently fulfilled their own reviewing obligations.

Sound on.

We recently organized #Agents4Science, the 1st conference where LLMs are both authors and reviewers🤖 It was an open experiment to assess how well AI can lead research and review papers. Today we report what we learned in Nature Biotechnology Highlights in 🧵

Can robots truly "think before acting"? 🤖 The conflict between sequential reasoning and high-speed motor control has finally been resolved. Today, we highlight a new multimodal breakthrough from OpenBMB and THUNLP: DeepThinkVLA —— a unified architecture that aligns reasoning

✨Introducing ProfBench. LLM eval shouldn’t be limited to math/code/short QA. Real work is: read professional docs → synthesize → produce long-form reports. ProfBench is a rubric-based benchmark written by domain experts (PhD/MBA) across 4 professional domains: Physics /

This super detailed tutorial by Pau Labarta Bajo is pure gold 🪙 "Fine-tuning a Small Language Model for browser control with GRPO and OpenEnv" LFM2-350M (Liquid AI) + BrowserGym (OpenEnv) + GRPO (TRL) for learning browser control 🤝 paulabartabajo.substack.com/p/fine-tuning-…

🥇 Nemotron-Orchestrator-8B takes #1 on GAIA benchmark! An 8B orchestrator coordinating intelligent tools and agents, trained via end-to-end RL, demonstraing small language models are shaping the future of agentic AI 🚀 huggingface.co/spaces/gaia-be…

x.com/i/article/2025…