AI Research

Liquid AI just dropped an experimental model, LFM2-2.6B-Exp, that uses pure reinforcement learning to boost its skills. It's a tiny 2.6B model that's now punching way above its weight class, even outperforming models hundreds of times its size.

NVIDIA's AI research team just dropped something wild: NitroGen, a foundation model that learns to play video games by watching 40,000 hours of gameplay videos from the internet. Here's how it works and why it's a big deal.

Ever seen a demo of an AI agent that can do anything, only for it to fall apart when you try it? A new paper from Stanford and Harvard finally explains why this happens and lays out a clear path to building AI that actually works.

Ever wonder how AI could understand the world like we do? Meta's new PE-AV model is a huge step forward, learning to connect sights, sounds, and text in a single, unified brain. Let's break down how it works and why it's a big deal for the future of AI.

A surprising new AI model from Korea is outperforming giants. More importantly, the startup behind it just published their playbook, revealing four critical lessons for any company trying to build its own LLM.

For the first time, AI can analyze the nuances of language as well as a human expert. This isn't just about grammar; it's about a 'gut feeling' for what sounds right. Let's talk about what this new metalinguistic ability means for all of us.

Ever wondered how an AI *really* thinks? OpenAI just released a new set of tools called 'circuit-sparsity' that lets us peek inside the black box, revealing the tiny, understandable circuits that power complex models. Let's break down what this means.

Ever feel like AI models have the memory of a goldfish? Google Research is tackling this with Titans and MIRAS, new approaches that give AI a persistent, long-term memory, allowing them to process millions of tokens without forgetting key details.

Researchers from Apple just dropped a new AI model called CLaRa, and it tackles one of the biggest headaches in Retrieval-Augmented Generation (RAG). It compresses documents by up to 128x and teaches the AI's two halves to finally work together.

Ever feel overwhelmed by research papers? Let's walk through how to build a simple but powerful AI agent that can read literature, form hypotheses, and even write scientific reports for you.

Five years ago, Google DeepMind's AlphaFold solved a 50-year-old biological puzzle and won a Nobel Prize. We talked to its co-creator, John Jumper, to find out what's happened since the hype died down and what's next for the AI that changed science.

Reinforcement learning for large models is often stuck in a traffic jam, with a few slow tasks holding up powerful GPUs. Discover how Moonshot AI's new 'Seer' system acts like a smart traffic controller to boost throughput by up to 97%.