🚨 Postdoc Hiring:.I am looking for a postdoc to work on rigorously evaluating and advancing the capabilities and safety of computer-use agents (CUAs), co-advised with @ysu_nlp @osunlp. We welcome strong applicants with experience in CUAs, long-horizon reasoning/planning,.

RT @jxwuyi: 🔍We introduce ASearcher, a search agent trained by end2end RL.Large-scale (up to 128 turns) RL with AReaL unlocks Long-Horizon….

More details about Mind2Web 2: .Explore examples here:

RT @_zifan_wang: 🧵 (6/9) Our findings suggest that traditional capability benchmarks may not adequately assess search-base LLM agents. We r….

Glad that @scale_AI team specifically studies this "Search-Time Data Contamination" (STC) problem of existing agentic search benchmarks, where an agent might just retrieve a source that directly leaks the answer - no actual reasoning or complex web navigation is needed. In.

Glad that @scale_AI team specifically studies this "Search-Time Data Contamination" (STC) problem of existing agentic search benchmarks, where an agent might just retrieve a source that directly leaks the answer - no actual reasoning or complex web navigation is needed. In.

RT @xiangyue96: 20 months after our multimodal reasoning benchmark MMMU ( release, both frontier and open models ar….

RT @abeirami: The main ingredient that led to GRPO's performance leap is the calibration of the reward/value via multiple rollouts per prom….

RT @sayashk: How does GPT-5 compare against Claude Opus 4.1 on agentic tasks? . Since their release, we have been evaluating these models o….

RT @CaimingXiong: 🚀 Computer-using agents represent a powerful new paradigm for human-computer interaction. Over the past year, we’ve explo….

RT @AnthropicAI: New Anthropic research: Building and evaluating alignment auditing agents. We developed three AI agents to autonomously c….

The System Two Safety idea is very interesting. In environments that may have maliciously injected instructions, deliberative reasoning seems particularly important: . Are the instructions in the current context out of place or suspicious, e.g., see Indirect Prompt Injection

RT @AbrahamOwos: Yaaaay! our paper got a best social impact award at ACL!!!🎉🎉. I couldn't attend the conference sadly.

RT @nouhadziri: Come join us tomorrow at the 1st LLM agents workshop in ACL (REALM); amazing talks and oral presentations are ahead, with a….

RT @boyuan__zheng: Remember “Son of Anton” from the Silicon Valley show(@SiliconHBO)? The experimental AI that “efficiently” orders 4,000 l….

Check out our WebGuard led by @boyuan__zheng: the first large-scale dataset for training and evaluating guardrails to detect consequential web agent actions (actions with significant or irreversible consequences): 📊 4,939 human-labeled actions.📷 193 websites across 22 domains.

RT @scale_AI: WebGuard turns abstract safety into something measurable. Captured from 193 different websites across 22 diverse domains, it'….

RT @scale_AI: As AI agents start taking real actions online, how do we prevent unintended harm?. We teamed up with @OhioState and @UCBerkel….

RT @ysu_nlp: Safety is one of the biggest blockers for computer use agents: how can I trust an agent won’t accidentally do something conseq….

RT @AnaisHowland18: WebGuard is a big step for web-agent safety: ~5k human-tagged actions across 193 sites. Frontier LLMs hit <60% on high-….

RT @niloofar_mire: I’m gonna be recruiting students thru both @LTIatCMU (NLP) and @CMU_EPP (Engineering and Public Policy) for fall 2026!….