Agent scaffolding matters as much as, or even more than, raw model capability for hard agentic tasks. In our latest research with @Meta, we show that carefully designed scaffolding achieve 54.3% (Claude Opus) and 52.7% (Claude Sonnet) on SWE-Bench-Pro, compared to a 52.0% Claude

What if small models could outperform giant agentic systems and monolithic LLMs - not by reasoning harder, but by orchestrating better? 🤔 Turns out they can, and we can do it with efficiency in mind. ❗️We’re releasing ToolOrchestra — a framework + model for training dedicated

A major breakthrough in reinforcement learning for robot training and the NeurIPS 2025 Best Paper. When training robots to walk, navigate, or manipulate objects, RL researchers have usually been using relatively shallow networks—typically 2-5 layer MLPs mapping sensor readings

The paper:

More agents is not always better 🤷‍♂️ Better coordination, aligned to the task, is what actually improves outcomes! Awesome paper from Google that shows that MAS do not scale like neural networks... What it means is that adding agents probably degrades performance unless the task

Train your own deep research agent in under a day Agent-R1 uses RL to teach agents proper tool use over multi-turn interactions, achieving 3x better performance than RAG We provide a step-by-step tutorial of the work: MDP extensions, action masking, and more. See below!

Very very impressive

https://t.co/FQe5bCBqW1

“You get many chances as long as you keep trying” — Peter Thiel

This is breathtaking 🙏

It took 9 years and 3 billion miles to get this shot. Pluto’s icy Mountains. https://t.co/jpj3kSYQoz

"Patience and aggressive opportunism is what you need. It's an odd combination, but it's what works best." — Charlie Munger

Agency > Intelligence I had this intuitively wrong for decades, I think due to a pervasive cultural veneration of intelligence, various entertainment/media, obsession with IQ etc. Agency is significantly more powerful and significantly more scarce. Are you hiring for agency? Are

Charlie Munger: “It's very counterproductive for an individual to feel like a victim, even if he is one. The best attitude is just to be cheerful about everything and keep plugging along.”

Paul Graham on why you shouldn’t write with AI: “In preindustrial times most people's jobs made them strong. Now if you want to be strong, you work out. So there are still strong people, but only those who choose to be. It will be the same with writing. 𝐓𝐡𝐞𝐫𝐞 𝐰𝐢𝐥𝐥

Marc Andreessen on product-market fit. “You don’t have a business until you have a product that a lot of people want.” “You can tell because the market’s pulling the product.” “Until you have that, time spent building the business around the product is pointless.” “Best case

Master quant of all master quants

Merry Christmas everyone near and far, and Happy New Year 🙏☘️

https://t.co/fxHYR6S29D

Jeff Bezos in a 2008 talk for the startup accelerator Y Combinator... One way to predict the future is to compare it to something similar from the past.