🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

The data team has made a lot of improvements to the data since Llama 3 and we’ve worked meticulously to build a datamix to ensure the highest model quality. Llama 4 Scout was trained on 40 trillion tokens of image and text data.

Large Audio Models should be the foundation models for voice assistants, but most benchmarks focus on chat & audio analysis skills. Read about our big team effort to develop a set of benchmarks to cover all the capabilities a model needs to support a great voice assistant!

40% with just 1 try per task: SWE-agent-LM-32B is the new #1 open source model on SWE-bench Verified. We built it by synthesizing a ton of agentic training data from 100+ Python repos. Today we’re open-sourcing the toolkit that made it happen: SWE-smith.

It feels worth conference organizers running a study to see if this significantly impacts reviewer scores. I hope things like this are placebos, but if not we need to seriously consider whether existing peer-review processes for big ML conferences are providing value.

Saturday, Tel Aviv, the protest gets bigger. Nearly 1000 Israelis holding pictures of children killed in Gaza in a silent protest in city center. Many more asked to join

Here are some of the benefits of federally funded research: publicusaresearchbenefits.com

Please reshare! Important piece on American Science, Health, and Technology funding. Medical and Scientific Research Makes America Great | TIME time.com/7286127/medica…

For trying to understanding LMs deeply, EleutherAI’s Pythia has been an invaluable resource: 16 LMs (70M to 12B parameters) trained on the same data (The Pile) in the same order, with intermediate checkpoints. It’s been two years and it’s time for a refresh.

"Ending a $5 million research study when it is 80% complete does not save $1 million, it wastes $4 million." The Bethesda Declaration: Call for NIH & HHS Leadership to Deliver on Promises of Academic Freedom & Scientific Excellence. Read and support here: standupforscience.net/bethesda-decla…

New #ACL2025NLP Paper! 🎉 Curious what AI thinks about YOU? We interact with AI every day, offering all kinds of feedback, both implicit ✏️ and explicit 👍.  What if we used this feedback to personalize your AI assistant to you? Introducing SynthesizeMe! An approach for

🚨 70 million US workers are about to face their biggest workplace transmission due to AI agents. But nobody asks them what they want. While AI races to automate everything, we took a different approach: auditing what workers want vs. what AI can do across the US workforce.🧵

If they can handcuff a U.S. Senator for asking a question, imagine what they will do to you.

Receiving the Turing Award, Andrew Barto thanks the federal government (especially AFOSR) for supporting work on reinforcement learning before it was popular, and says without it “none of this would have ever happened”. RL is now at the heart of training/finetuning LLMs,

One of the pioneers of reinforcement learning (Andrew Barto), which powers AlphaGo, ChatGPT, and robot motor skill learning to name a few, mentions the key role U.S. National Science Foundation funding played in the development of that work. At the ceremony with him receiving the Turing Award.