Image created with Gemini. Image prompt: A photo-collage grid of slightly offset and overlapping rectangular panels snapping together to form one scene of a flat hand-drawn hand holding up a small tablet, where the tablet’s screen contains the same grid pattern repeated in miniature, set against a flat pool-turquoise sky and terracotta pink ground with hot yellow squiggle marks connecting the panels, rendered in flat saturated acrylic color with hard edges, no shading, high noon light and no shadows, tablet-drawn digital line.
// The Meta-Agent Challenge // How good are current agents at self-improving? This is a great paper covering some of the challenges. They propose the Meta-Agent Challenge (MAC), where they give a coding agent a sandbox, an evaluation API, and a time budget, then ask it to
https://x.com/omarsar0/status/2062919381777350914
Meta AI in the Ray-Bans has gotten very good. Walk down the street, ask a question about something you’re looking at, discuss it. Voice is natural and fast, and isn’t interrupted by environmental sounds. Phone stays in your pocket. Feels freeing and a direction AI will take.
https://x.com/_simonsmith/status/2064366553181626374
At least until (if?) rapid improvement stops, it seems less likely someone is going to catch the Big Three AI Labs. Microsoft and Meta released their models, which were fine, but not frontier. SpaceX also hasn’t regained its position. Chinese models are improving, but still lag.
https://x.com/emollick/status/2062737052479734205
EU Orders Meta To Stop Blocking Rival AI Chatbots On WhatsApp
https://www.engadget.com/2191213/eu-orders-meta-to-stop-blocking-rival-ai-chatbots-on-whatsapp/
Hyper-parameter tuning works really well! Since Keller asked very nicely – I took it as a challenge to find a minimal edit distance from his config to a working config with no further code modifications, which means vanilla Meta’s distributed_shampoo PyTorch package as is which
https://x.com/_arohan_/status/2064631528806908134
This feels purpose built to hobble Meta
https://x.com/kylebrussell/status/2064502244041511348
Ilya (sorry for name dropping) met with me in 2024 and said in a meeting that we can do better than Shampoo family (renamed to Muon and friends). Now I can say this is very true, there exist an optimizer that shows the same scale of improvements that shampoo shows over adamw on
https://x.com/_arohan_/status/2064036303021494418
Build was super fun! Here’s a video recap of my presentation. Check it out:
https://x.com/mustafasuleyman/status/2062891924453818418
Leave a Reply