🎉 Our paper “𝐻𝑜𝑤 𝑡𝑜 𝑇𝑟𝑎𝑖𝑛 𝑌𝑜𝑢𝑟 𝐿𝐿𝑀 𝑊𝑒𝑏 𝐴𝑔𝑒𝑛𝑡: 𝐴 𝑆𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐𝑎𝑙 𝐷𝑖𝑎𝑔𝑛𝑜𝑠𝑖𝑠” got an 𝐨𝐫𝐚𝐥 at next week’s 𝗜𝗖𝗠𝗟 𝗪𝗼𝗿𝗸𝘀𝗵𝗼𝗽 𝗼𝗻 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿 𝗨𝘀𝗲 𝗔𝗴𝗲𝗻𝘁𝘀! 🖥️🧠 We present the 𝐟𝐢𝐫𝐬𝐭 𝐥𝐚𝐫𝐠𝐞-𝐬𝐜𝐚𝐥𝐞

Thanks @_akhaliq for sharing our work! Excited to present our next generation of SVG models, now using Reinforcement Learning from Rendering Feedback (RLRF). 🧠 We think we cracked SVG generalization with this one. Go read the paper! https://t.co/Oa6lJrsjnX More details on

🚨🤯 Today Jensen Huang announced SLAM Lab's newest model on the @HelloKnowledge stage: Apriel‑Nemotron‑15B‑Thinker 🚨 A lean, mean reasoning machine punching way above its weight class 👊 Built by SLAM × NVIDIA. Smaller models, bigger impact. 🧵👇

1/ How do we evaluate agent vulnerabilities in situ, in dynamic environments, under realistic threat models? We present 🔥 DoomArena 🔥 — a plug-in framework for grounded security testing of AI agents. ✨Project : https://t.co/yOsZize8V1 📝Paper:

1/n Wish you could evaluate AI agents for security vulnerabilities in a realistic setting? Wish no more - today we release DoomArena, a framework that plugs in to YOUR agentic benchmark and enables injecting attacks consistent with any threat model YOU specify

🚨 SLAM Labs presents Apriel-5B! And it lands right in the green zone 🚨 Speed ⚡ + Accuracy 📈 + Efficiency 💸 This model punches above its weight, beating bigger LLMs while training on a fraction of the compute. Built with Fast-LLM, our in-house training stack. 🧵👇

🚀 Exciting news! Our work LitLLM has been accepted in TMLR! LitLLM helps researchers write literature reviews by combining keyword+embedding-based search, and LLM-powered reasoning to find relevant papers and generate high-quality reviews. https://t.co/ledPN4jEmP 🧵 (1/5)

🚀 Struggling with literature reviews? LitLLM can help! This AI-powered tool retrieves relevant papers, ranks them using LLMs, and structures comprehensive reviews in no time. Just input your abstract and let AI streamline your research! #LitLLM #AIforResearch

Can your AI keep up with dynamic attackers? In a paper to appear at #AISTATS2025 with @avibose22 @LaurentLessard and Maryam Fazel, we study robustness to learning algorithms to dynamic data poisoning attacks that can adapt attacks while observing the progress of learning

Internship opportunities in safe, secure and trustworthy AI at @ServiceNowRSRCH. For more context, check out this recent post about our Reliable & Secure AI Research Team: https://t.co/qg03W7hQ3v To apply and/or see more details:

Come multiply and amplify research talent at @ServiceNowRSRCH. The team here is curious, driven, fun, diverse and a bit off the wall. #ArtificialInteligence #AIAgents https://t.co/3wgTvMJLH7

New AI transparency and traceability framework from @linuxfoundation. Useful for folks working on #ResponsibleAI #trustworthyai

Launch of the Canadian #AI Safety Institue. Joins the US, UK and several other safety institutes working on challenge of evaluance, assurance, etc. #aisafety #trustworthyai

Good post about trustworthy AI and the risks of AI by @PhilMercure in @LP_LaPresse this morning.

Débattre des risques liés à l'IA dans un camp rustique à deux pas de la plage avec des chercheurs des principaux laboratoires frontaliers et institutions universitaires. Quelle expérience incroyable ce fut. https://t.co/Rc9ISDw3NS

Helpful overview of the broad array of AI standards initiatives that exist, lack of consistency in how they are applied, etc. #ArtificialIntelligence #AI #policy

Be the fire. Wish for the wind. That’s from Taleb about anti-fragile. Our AI systems need that in the form of adversarial and exploratory stressors so we learn how snd where to adapt and thrive.

Transparency is a key #trustworthyai and #responsibleai principle, but it can also create security risks -- e.g., being open about model confidence and explainability makes model inversion and evasion easier. Managing that tradeoff is tough.

What's the gap in performance and risks between major foundation models and fine-tuned versions of those models? We talk lots about evals on foundation models but less about evals of fine-tuned, instruction-tuned versions widely used. Both are key, but we need intel on the diff.

MLCommons released a new #genAI #safety benchmark and taxonomy recently. Good contribution, but like other benchmarks it creates an illusion of holistic eval. These eval tools remain far too simple to give a good read on overall security and safety. https://t.co/I5oHQwCG11