This is a major role with massive scope and impact. Help us build a science of post-training data curation.

🧵 Academic job market season is almost here! There's so much rarely discussed—nutrition, mental and physical health, uncertainty, and more. I'm sharing my statements, essential blogs, and personal lessons here, with more to come in the upcoming weeks! ⬇️ (1/N)

I'm psyched for my 2 *different* talks on Friday ACL 2025: 1.LLM Security (11:00): What does it mean for an AI agent to preserve privacy? 2.Workshop on Large Language Model Memorization (16:00): Emergent Misalignment thru the Lens of Non-verbatim Memorization (& phonetic to visual attacks!) Join us!

🌿Introducing MetaCLIP 2 🌿 📝: arxiv.org/abs/2507.22062 code, model: github.com/facebookresear… After four years of advancements in English-centric CLIP development, MetaCLIP 2 is now taking the next step: scaling CLIP to worldwide data. The effort addresses long-standing

🚨 Got a great idea for an AI + Security competition? SaTML Conference is now accepting proposals for its Competition Track! Showcase your challenge and engage the community. 👉 satml.org/call-for-compe… 🗓️ Deadline: Aug 6

From unlearning to zero-knowledge verification, Chhavi Yadav Chhavi Yadav is advancing the theoretical backbone of Trustworthy AI. Her paper “Cold Case: The Lost MNIST Digits” is just one example of her commitment to foundational research. DINQ sees you.👀

#mlrc_2025 happening in ~2 weeks! #reproducibility Join our discord channel discord.gg/bDAK6WUCjU to connect and network for the event!

I really would like AI to file my taxes, but it can't :(

Our open models are here. Both of them. openai.com/open-models

Reliability is the key missing piece to the vast adoption of agentic AI workflows: Excited to discuss the issues affecting reliability in the Reliable ML Workshop at NeurIPS 2025: - workflow consistency - robustness to adversarial inputs and domain shift - safety & security

ICLR 2026 will take place in 📍Rio de Janeiro, Brazil 📅 April 23–27, 2026 Save the date - see you in Rio! #ICLR2026

Why we should support exploratory research.

🏆 We're thrilled to announce that Meta FAIR’s Brain & AI team won 1st place at the prestigious Algonauts 2025 brain modeling competition. Their 1B parameter model, TRIBE (Trimodal Brain Encoder), is the first deep neural network trained to predict brain responses to stimuli

1/Pretraining is hitting a data wall; scaling raw web data alone leads to diminishing returns. Today DatologyAI shares BeyondWeb, our synthetic data approach & all the learnings from scaling it to trillions of tokens🧑🏼‍🍳 - 3B LLMs beat 8B models🚀 - Pareto frontier for performance

With 13,364 citations and an h-index of 49, Kamalika Chaudhuri Kamalika Chaudhuri stands as a luminary in Trustworthy AI. Her influential work on differential privacy continues to shape the foundations of robust, reliable, and ethical AI. DINQ salutes. 🫡

A classic paper, collab between AI at Meta , Google DeepMind , and NVIDIA AI Developer Language models keep personal facts in a measurable amount of “storage”. This study shows how to count that storage—and when models swap memorization for real learning. 📡 The Question Can we

🗣️Job alert: Our Brain and AI team at FAIR (AI at Meta) is looking for a software engineer with experience in 3D rendering in the browser: us.meta.talentnet.community/jobs/ccb22e48-… Please RT 🙏

📢 Late-Breaking News! Travel grants for accepted abstracts at NeurIPS (San Diego, Mexico City, Copenhagen). 👩‍💻 Submit your abstracts: openreview.net/group?id=NeurI… ⏰ Deadline: Sept 07 More details: tinyurl.com/wiml2025cfp. #womeninstem #WomenInML #MachineLearning #WiML #NeurIPS

See the speakers for 2025-26 #KempnerInstitute Seminar Series! bit.ly/46xphKq @moira_dillon Kevin Ellis Tatsunori Hashimoto Hannah Choi Sanjeev Arora Sherry Yang Kamalika Chaudhuri Kyunghyun Cho Stephanie Chan Stella Biderman Azalia Mirhoseini Danqi Chen Tatiana Engel Lenka Zdeborova

by the way. recently wrote a paper on this! for transformers, the number is about 3.6 bits-per-parameter so you would need 25GB ÷ 3.6 bits ≈ 56.9B parameters to exactly memorize Wikipedia that’s a pretty big model actually