Highlights from recent evaluations (insurance underwriting & more): Surprising error modes in complex reasoning Trade-offs between tool use & efficiency Beyond accuracy: deeper evaluation with Snorkel Evaluate Full leaderboards →
GENERATIVE AI
-
Small Specialized AI Models Represent Future Direction
By
–
been saying this since October last year btw Small Specialized Models are the future
-
YouTube Shorts launches Veo 3 and Lyria 2 AI video generation
By
–
Fun new features in @YouTube Shorts: Veo 3 will generate video clips with integrated audio from a single text prompt, and Lyria 2 powers ‘Speech to song’ which can turn video dialogue into a soundtrack! https://t.co/El6XDwLZH2
— Demis Hassabis (@demishassabis) 16 septembre 2025Fun new features in @YouTube Shorts: Veo 3 will generate video clips with integrated audio from a single text prompt, and Lyria 2 powers ‘Speech to song’ which can turn video dialogue into a soundtrack!
-
Codex: Imagination as the Only Limit
By
–
Codex makes it feel like the only limit is your imagination. 👾🎮 https://t.co/ylMnHkx3gC
— Romain Huet (@romainhuet) 16 septembre 2025Codex makes it feel like the only limit is your imagination.
-

MoE GPU Communication Overhead Analysis Technical Deep Dive
By
–
Calling all Mathletes, this one is for you. We’ve been asked to show the math behind our MoE claims. So we did. Our analysis confirms: On GPUs, expert parallelism creates severe communication overheads that dwarf computation and make MoE training painfully slow. At
-
Claude Update Impact on Developer Understanding and Code Workflow
By
–
Felt it too since the last Claude update. My daily work (via Cursor) significantly changed… Hardest part these days is making the executive decision: do I even need to understand how it works? For throw-away scripts or tools with verifiable outputs it's increasingly tough!
-
Open-Ended Quality Diversity AI-Generating Algorithms Foundation Models
By
–
Excited to host @jeffclune this Friday in our AI4Science Community! Open-Ended, Quality Diversity, and AI-Generating Algorithms in the Era of Foundation Models Join in the link below!
-

Meta FAIR Releases New Test-Time Compute Scaling Method for LLMs
By
–
Meta FAIR just released a brand new test-time compute scaling method! With their new LLM solution aggregator trained with RLVR, they showed that majority voting or reward model ranking aren’t actually the most efficient. This simple yet robust method outperforms RM baselines.
-
YouTube Launches Veo 3 AI Video Tool for Creators
By
–
We’re building AI tools to support creators throughout the creative process and redefine what's possible on @Youtube
. Here are a few of our favorite AI features announced at this year’s #MadeonYoutube — Veo 3 is coming to YouTube Shorts, making it possible to create short -
Abraham AI Project Reaches Final Stage with New Details
By
–
To learn more about Abraham, this thread summarizes its goals and has links to the original paper and articles I wrote about it. In the next few weeks, we'll release more details about the next (and final) stage of this project!