give each pixel a random Pokemon type, and then battle pixels against their neighbors, updating each pixel with the winning type (using the Pokemon type chart). we quickly see areas of fire > water > grass > fire, electric sweeping over, ground frontiers taking over etc etc

To push the open source frontier for RL + LLMs, we need scalable, modular environments with real-world complexity, beyond math benchmarks. Today, we’re releasing *benchmax*. An open-source framework to build, run, & scale useful RL envs for LLM fine-tuning, with integrations to

We're announcing a $200M ceiling contract with the U.S. Department of Defense. Through hands-on prototyping and direct collaboration across the Department, we will help enhance U.S. national security and responsible AI deployment.

🚀 Meet Reka Research––agentic AI that 🤔 thinks → 🔎 searches → ✏️ cites across the open web and private docs to answer your questions. 🥇 State-of-the-art performance, available now via our API and Playground!

construction of a Julia set through repeated twists, reflections, and shifts. The twists correspond to the complex square root.

the mandelbrot set as a map of all Julia sets - version 2

a julia set fractal can be created with repeated twistings and translations of a starting image. The twistings double the detail in the picture, and correspond to complex square-roots

a ring built from the Julia sets that lie along the main cardioid of the Mandelbrot set - version 2

1/ Can codebase-specific RL push the frontier for code LLMs?. At @cgftlabs, we helped a client RL-tune Qwen-2.5-7B on their internal codebase for unit test creation, with coverage-guided GRPO. The result? It beats o4-mini & o3. Here’s how it works (link to full blog in bio) 🧵