Now you know there's another dude just discussed AI Safety and Security with both sides ;) #NeurIPS2023 [📸 With legendaries @ylecun and Yoshua Bengio]

🫂Dear Metamates impacted today—beyond safety, our foundation model team also has opening for native multimodal roles (text/video/audio, OCR, and more):

BTW, We’re growing a safer foundation model research stack at @tiktok_us —safety pretraining, RLHF/RLAIF, evals. 🚨Intern + FTE roles. 📨DMs open.

"I'm excellent, grandma" -- Sora 2 Pro > a viral and unhinged vine post

Can someone fact-check this for me: why does every Sora 2 vid featuring @sama have the same Adidas sneakers ? Is it just me or is this… consistent? 😂

[10/12] To make matters worse, soon AI agents will know you better than your friends. Will they give you uncomfortable truths? Or keep validating you so you’ll never leave?

🧵 New paper from @AISecurityInst x @AiEleuther that I led with Kyle O’Brien: Open-weight LLM safety is both important & neglected. But we show that filtering dual-use knowledge from pre-training data improves tamper resistance *>10x* over post-training baselines.

gpt-oss is overfit on refusal safety — sky-high over-refusal, tanked helpfulness on my bench. gpt-5 swings the other way: max helpfulness, refusals softened with follow-ups even on harmful asks. What do you see in this pattern?

who's making these graphs

'Update Federal procurement guidelines to ensure that the government only contracts with frontier large language model (LLM) developers who ensure that their systems are objective and free from top-down ideological bias.' there is an executive order on this arriving today.

The AI action plan has been released.

🎉 Thrilled to be presenting my first paper at @icmlconf! "Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning" We introduce ACTOR—a lightweight, activation-based training method that reduces over-refusal without

🔹 AI alignment really needs interdisciplinary work! 🔹 See my talk on "how to humanize AI to persuade them for jailbreaking":

I'll never forget this model as well as the relationship between pretrain, SFT and RLHF.

Sparsity can make your LoRA fine-tuning go brrr 💨 Announcing SparseLoRA (ICML 2025): up to 1.6-1.9x faster LLM fine-tuning (2.2x less FLOPs) via contextual sparsity, while maintaining performance on tasks like math, coding, chat, and ARC-AGI 🤯 🧵1/ https://t.co/ilZgkfj78J

🌉 Bridging Offline & Online RL for LLMs 🌉 📝: https://t.co/G12TS6Z84n New paper shows on verifiable & non-verifiable tasks: - Online DPO & GRPO give similar performance. - Semi-online (iterative) DPO with sync every s steps (more efficient!) works very well also. - Offline DPO

1/ 🔥 AI agents are reaching a breakthrough moment in cybersecurity. In our latest work: 🔓 CyberGym: AI agents discovered 15 zero-days in major open-source projects 💰 BountyBench: AI agents solved real-world bug bounty tasks worth tens of thousands of dollars 🤖

AIR-Bench is a Spotlight @iclr_conf 2025! Catch our poster on Fri, Apr 26, 10 a.m.–12:30 p.m. SGT (Poster Session 5). Sadly, I won’t be there in person (visa woes, again), but the insights—and our incredible team—will be with you in Singapore. Go say hi 👋

🚀 Really excited to launch #AgentX competition hosted by @BerkeleyRDI @UCBerkeley alongside our LLM Agents MOOC series (a global community of 22k+ learners & growing fast). Whether you're building the next disruptive AI startup or pushing the research frontier, AgentX is your

1/14: If sparse autoencoders work, they should give us interpretable classifiers that help with probing in difficult regimes (e.g. data scarcity). But we find that SAE probes consistently underperform! Our takeaway: mech interp should use stronger baselines to measure progress 🧵

Excited to share new work from my internship @GoogleAI ! Curious as to how we should measure the similarity between examples in pretraining datasets? We study the role of similarity in pretraining 1.7B parameter language models on the Pile. arxiv: https://t.co/iyS3Fxtx9a 1/🧵