
My @iclr_conf 2025 Keynote on "Open-Endedness, World Models, and the Automation of Innovation" is now publicly available: iclr.cc/virtual/2025/invited…
✓Co-Founder and Chief Scientist of @Recursive_SI, Professor of AI @AI_UCL, PI @BOLD_Lab_AI, Fellow @ELLISforEurope. Ex @GoogleDeepMind @AIatMeta @CompSciOxford

My @iclr_conf 2025 Keynote on "Open-Endedness, World Models, and the Automation of Innovation" is now publicly available: iclr.cc/virtual/2025/invited…

NEW: Britain's recent higher-than-expected growth was the result of a tech sector surge and AI adoption, according to the FT. The UK is not quite the US, but it is benefitting from being the most tech-weighted large economy in Europe. G7 countries growth in Q3: 🇺🇸 +0.8% 🇬🇧 +0.7% 🇨🇦 +0.3% 🇩🇪 +0.2% 🇮🇹 0.0% 🇯🇵 -0.1% 🇫🇷 -0.1% Full piece in the FT: ft.com/content/76fec740-0825… Surging British tech sector helps drive UK economy

We @bold_lab_ai are hiring a large cohort of home and overseas students for Fall 2027 entry. If you’re excited about discovering the next paradigm of open-source AI and want to work on ambitious, exploratory, but scalable ML that empowers humanity, please apply by 1st of December 2026 AOE to maximise chances of funding: eng.ox.ac.uk/study/research-…

It was a lot of fun pointing our Eureka Machine at alien optimisers 👽👽👽, glad we could demonstrate some of the insanely wild creativity of AI there Even more fun contributing together with some old friends at @SakanaAILabs @RobertTLange ;)


Great to see @Recursive_SI in the top 10 in talent movement. We are actively hiring: recursive.com/join-us Join a fantastic team of ex OpenAI, Anthropic, DeepMind, and Meta members of technical staff.


Rethinking pre-pretraining (PPT) Allocating compute to synthetic PPT never outperforms pure natural-language training. In our compute-matched settings: even when natural language is scarce, repeated epochs over a 100M-token corpus beat every synthetic curriculum. 🧵

Today we are happy to share one of our first open research releases: Latent Sparse Attention (LSA), a step towards subquadratic attention that matches accuracy of full attention at ultra-long context. Sparse attention attends to only a few tokens per query, but it has to find the right ones. At 10M tokens, that search takes 98% of the attention layer's time. LSA makes sparse attention up to 22.7× faster and full attention 72.4x faster. With @kubicek_ales @caciolai @PengLiangzu @SukritSumant @Ben_Hoov @FraserGreenlee @RMBattleday @radim_urban @thoefler @jcrwhittington

I’ve had so much fun making bricky, and playing with it: a compiler for LEGO® models, built for AI agents. The agent writes a Python build script. bricky places every part, reports each mistake, and renders the model. Thanks to @omar_aflak for the idea of a LEGO® compiler.

Mindblowing chart from OBR. In 2029 real household disposable income per person expected to be just 1% higher than in 2019. What an unbelievably awful decade. It used to grow by 2%+ every year, not 1% in 10 years.

10 minutes with an iPhone, and this gorgeous craftsman house is now a 3D world you can walk through! It still blows my mind how easy Echo-3 makes this. Go capture your favorite space 🙂 spaitial.ai/echo-3

I got curious how Genie even started.. so I went back to the beginning 🧵 First, what is Genie? It's a world model. You give it a prompt, a photo, or even a sketch, and it builds a world you can walk around in and play. It came out of a 25-person @GoogleDeepMind research team, including Jack Parker-Holder and Tim Rocktäschel. Why Google cares: DeepMind has been building simulated worlds for over a decade to train AI agents. Endless worlds mean endless practice for agents and robots. Feb 2024: Genie 1. An 11B model trained on unlabeled internet videos. A sketch or a photo became a world you could play, frame by frame. Dec 2024: Genie 2. Full 3D worlds that stay consistent for up to a minute. Aug 2025: Genie 3. The first one you could explore in real time. 24 fps, 720p, and the world holds together for a few minutes. Oct 2026: Genie 4. Scenes that react, make sound, and people in them who move and speak. From a sketch to a living world in about 2.5 years 🤯

thanks for trying Flint! it was cool to see it trending #2 on @huggingface. glad also to hear so many people enjoying it. we're working on improving tool calling, broadening harnesses / inference support, and reducing thinking. anything else you'd like to see?

