
I gave Claude the ability to write music Here's the first song it wrote =)

I care about model performance and systems. There are unrealized capabilities for all those with the eyes to see!

I gave Claude the ability to write music Here's the first song it wrote =)


"How do stars form? To help answer this question, NASA used the James Webb Space Telescope to take the featured detailed image of the star-forming region NGC 7129 in infrared light. Star formation progresses when gravity pulls together massive amounts of gas, but soon things can get complicated when previously formed stars emit powerful jets and winds -- and some then explode. Opaque dust can form in the cool atmospheres of stars and, when dispersed, can obstruct visible-light images. Many stars can be seen to be forming in the featured image, with the young stars on the left, in the gold-colored gas, typically being more mature than the young stars on the right -- in the red region. Jets and winds likely cause the filamentary nature of the red-colored gas. Studying stars at different stages of development bolsters understanding about what physical processes are most important and when."




In our new paper we study sample efficiency across VLAs With enough demos, VLAs can achieve high success rates. But how many demos do they actually need to learn a new task? And how many training steps does it take to get there? We benchmark pi and GR00T to find out. 🧵👇







🦍 Introducing Memorilla: memory as a modality for LLMs Memorilla turns everything a frozen LLM needs to remember into just 16 memory embeddings ⚡ It uses up to 4,212× fewer tokens and is 27× faster than full context inference while more accurate 🎯 Pushes the Pareto frontier over RAG and agentic memory in personalization, long-document and knowledge QA A thread 🧵:

