Do language models have an internal world model? A sense of time? At multiple spatiotemporal scales? In a new paper with Max Tegmark we provide evidence that they do by finding a literal map of the world inside the activations of Llama-2!

Similar experiments have obviously been led post-word2vec and we get almost the same design on a map of Germany on a 2018 course from Jeremy Howard (actually highly recommended if you want to know more about embedding spatialization)

Unloading trucks with robots! 📦 Bastian Solutions has developed a mobile robotic system that automates high-volume, floor-level unloading of trailers. The robot seamlessly drives itself in and out of trailers and docks without any fixed infrastructure. This system picks up

For the first time, a humanoid robot can fold laundry using a neural net We made no changes to the Helix architecture, only new data

How to use @Grok Imagine to turn your photos into videos

Nvidia is the leading open model/ecosystem lab in the U.S. in the last 6months.

Robotics without #AI? Directionless. AI without #robotics? Disembodied. 🤖 The next #industrial transformation needs both. Find out why in our Future of Robotics 2035 report: hxgn.biz/4nL1UUt Lukas Ziegler

That's what a French engineer (Julien Allali) is building for fun (spotted by Remi Fabre) 🤯🤯🤯 Things like that are making me excited about the power of open-source and collaboration for robotics that Pollen Robotics & Hugging Face can unleash! Let's build the future!

Python is wild for finance. I built a sector-based hierarchical risk parity portfolio in 4 lines of Python code. This is how: 🧵

🚨 Open Model Leaderboard Update New open models entered the Text Arena, and the rankings by provider have reshuffled for August. - Qwen-3-235b-a22b-instruct from Qwen takes the crown 🏆 - GLM-4.5 from Z.ai and gpt-oss-120b by @openAI debut in the top 10! All the

Today, we published new details about the advanced Eyetracking technology from our computer vision team for Bigscreen Beyond 2e! Unlike traditional computer vision implementations, we trained a unique AI model of the human eye in 3D space. Machine learning with cloud-based GPUs

The key is that among those top 10 models, Google's Gemma and Mistral Small are the only ones who can fit/run on consumer hardware. A single RTX 4090 is enough. All the other models have >100b parameters and requires very expensive professional hardware. This changes the

Solution: Thunderbolt Bridge subnet needs to be 255.255.255.0

Today was my last day at xAI, the company that I helped start with Elon Musk in 2023. I still remember the day I first met Elon, we talked for hours about AI and what the future might hold. We both felt that a new AI company with a different kind of mission was needed. Building

For anyone still running into issues with gpt-oss models, this command in vLLM for the 120b model delivers outstanding results. I’m using it for synthetic medical dataset generation.

The future of package delivery is self-driving vehicles paired with humanoids that can perceive, reason and navigate within spaces designed for human interaction. Also, drones.

Coming to MLX :)

To recap, instead of producing independent chunk embeddings, contextualized chunk embedding models like voyage-context-3 process the entire doc in a single pass to embed each chunk. This leads to document-aware chunk embeddings that generate semantically aware retrieval in RAG.

ABPCs can initiate de novo antler generation, which is the only mammalian organ to fully regenerate annually in adulthood Unlike conventional MSCs (bone marrow, adipose or umbilical cord origins) showing senescence after 10–15 culture cycles, ABPCs maintain robust proliferative

We launched a Claude Code learning mode! Claude Code not only makes you more productive, but can now also help you get better at coding. Whether you're a CS student or seasoned programmer, it will push you to think deeper about the code you're generating.