This paper from Tsinghua University and Shanghai Jiao Tong University received perfect scores (6, 6, 6, 6) at NeurIPS 2025! It aims to answer a key question: Does reinforcement learning really make large language models better reasoners? The authors study Reinforcement Learning

Announcing we’ve raised $7M to build terraforming robots at Terranova. We are lifting cities out of flood zones using underground injection. Flooding is a multi-trillion dollar U.S. national security problem and this funding will get us to market with the first solution. Led by

There's good news EVERYWHERE: AI solved a 25-year Crohn's disease mystery... Researchers at UC San Diego discovered a broken bond between 2 gut proteins that drive chronic inflammation.  AI's potential to heal is massive.

Grok upgrades

🚨 ElevenLabs just killed every transcription tool on the market. They dropped Scribe v2 Realtime and it's not just another speech-to-text model. This thing processes audio in real-time with zero lag, handles multiple speakers, and actually understands context not just words.

Kalpa Labs (Kalpa Labs (YC F25)) is training one generalist model for all speech tasks. Steer it like an LLM, give long system prompts & make use of native contextual awareness. Congrats on the launch, Prashant Shishodia and @jhagauti! ycombinator.com/launches/Op4-k…

Introducing Cerebras for Nations, our global initiative to advance and scale sovereign AI. How it works: 1️⃣ We will build world-class AI supercomputers with our WSE-3 chips and CS-3 systems 2️⃣ Co-develop state-of-the-art models and deploy with the world’s fastest inference 3️⃣

To kick off ElevenLabs Summit we are releasing the best speech-to-text real-time model: Scribe v2 Realtime. - <150ms latency - incredible across all languages - 93.5% accuracy on FLEURS for top 30 languages And of course available in Agents Platform straight away too.

Sanctum lets you ship features to simulated users before real ones. It builds AI agents from your session recordings that explore new changes in test environments—surfacing bugs, UX friction, and performance issues, helping you predict how users will react.

🤫 Something's been brewing in stealth. Our SDK team's side project, codenamed W&B LEET, is being unleashed. We are releasing a full Terminal UI (TUI) for live, interactive W&B monitoring right in your terminal. No browser, no internet, no problem.

Introducing captions.events, an open-source template for broadcasting realtime transcripts. • Generate live transcripts with Scribe • Broadcast to many using Supabase • Translate with Chrome’s built-in AI Demo and open source template below.

Introducing ElevenLabs Iconic Marketplace - a curated platform for licensing legendary voices. As part of this launch, we’re honored to announce our partnership with Sir Michael Caine, whose voice joins the Iconic Marketplace and ElevenReader.

World Labs is now open to everyone ✨ Generate persistent 3D worlds from an image, video, or text prompt. It's easy for consumers to use - but there are some impressive editing & stitching features for 3D experts. I've had SO much fun with it. 👏 to World Labs team!

Really slick. I expect to see this autocomplete paradigm everywhere in AI interface design over the coming months.

"If you had secret microphones in robotics labs across the country right now you'd just be hearing 'holy shit!'" A lot of the smartest founders, engineers, and investors are investing time and money in robotics right now because they can finally see the path.

The first compelling Grok Imagine 30 min show is probably ~3 months away

Finally, the moment you’ve all been waiting for 🥁: Rolling out today, you can now create your own custom video overview styles by typing in a prompt in the customization box. Drop your favorite examples below (our expectations are SKY HIGH).

I've spent my whole life trying to make absence mean something. When I was five, the piano bench became my world. Practicing eight, ten hours a day wasn't punishment; I learned early to speak love through sacrifice. Every practice session was also a choice to say no. No to

We’ve developed a new way to train small AI models with internal mechanisms that are easier for humans to understand. Language models like the ones behind ChatGPT have complex, sometimes surprising structures, and we don’t yet fully understand how they work. This approach

This is a product I have dreamed about for years-- an events system for the entire internet. It brings some of the design principles I love about crypto (observability, arbitrage, composability) to the much larger offchain world. I'm monitoring: > a product (1996 runway