Daily Brief

Lee Robinson discusses the future of Cursor and AI-native software development. Speaker: Lee Robinson — ML, Model Behavior, Cursor Model research and personality at Cursor. Previously Vercel. X: https://x.com/leerob LinkedIn: https://www.linkedin.com/in/leeerob/ GitHub: https://github.com/leerob Website: https://leerob.com Timestamps 0:37 - Introduction and recursive model improvement overview 1:55 - The two-loop training framework (inner and outer loops) 2:33 - Progress and success of Composer 2.5 4:31 - Improving the outer loop with user feedback 5:40 - Climbing the inner loop with high

Three agents click, type, and scroll through three different apps on one desktop at the same time, and the user's own mouse and keyboard never move. That's the live demo behind cua driver, a tool the team built in a single weekend after Codex shipped its own computer use model. Instead of taking over the hardware cursor, it talks straight to the accessibility layer underneath the operating system: UI Automation on Windows, AT SPI on Linux, AX on macOS. Those undocumented APIs let a click land on a background window or a keystroke reach a hidden one, so any number of agents can act without stea

Full episode: https://www.youtube.com/watch?v=OS1NZLgKM2c Me on twitter: https://x.com/dwarkesh_sp

Where the loss function for training LLMs comes from. Job opportunities aligned to this audience: https://3b1b.co/talent Early views and other perks for supporters: https://3b1b.co/support Home page: https://www.3blue1brown.com Manim animations by Aaron Gostein and Grant Sanderson NanoGPT animation by Clayton Rabideau 3d black-box model by Paul Dancstep Chess game and Lagrange Multiplier scenes by Nishad Deulkar Music by Vince Rubinetti Timestamps 0:00 - Language trees and zipping 3:02 - Recap optimal codes 5:20 - Defining cross-entropy 8:26 - Intuition and examples 12:59 - Application to l

Full episode: https://www.youtube.com/watch?v=OS1NZLgKM2c Me on twitter: https://x.com/dwarkesh_sp

Earlier this year, OpenAI ran Parameter Golf, a model-training competition that doubled as a hiring filter. Over 1,000 researchers competed to train the best small language model under a 16MB cap. The top contributor was the one candidate OpenAI couldn't hire. Our autonomous research agent Aiden finished with 7 merged records, more than twice as many as any other contributor, and ended up the most-cited participant in the community. This talk is about what those 22 days showed. I'll cover on high level how does it works and which of its ideas produced the records. But the part worth more than

Eve Bouffard from Y Combinator explores the concept of "Imagination Engineering"—the idea that with increasingly powerful AI models, the primary challenge for humans is no longer technical execution, but the ability to dream up bold, innovative ideas (0:13-0:59). Key themes and experiments shared: Thinking in Public: Inspired by Paul Graham, Eve describes an experiment where she shared her stream of consciousness in a dedicated Slack channel (Eve thoughts). She then used AI to aggregate these raw thoughts into a personal, interactive website (1:35-4:26). Software on Demand: Eve demonstrates

Garry Tan, President of Y Combinator, discusses how the rise of AI-native companies is revolutionizing organizational productivity, allowing lean teams to operate at a scale previously requiring hundreds or thousands of employees (0:52-1:07). Key Takeaways: The 400x Leverage: Tan reports a massive increase in output (approximately 400x) by shifting from writing individual lines of code to managing AI agents (2:32-2:35). Wiring the Work: He emphasizes that founders should treat AI not as a simple autocomplete tool, but as a workforce (3:58). He explains that core organizational components—lik

Andy Beam (CTO) and Rafa Gómez-Bombarelli (Co-founder & CSO of Physical Sciences) of Lila Sciences join us to talk about building scientific superintelligence. Andy makes the case that the internet is a spent resource ("we have but one internet. It's the fossil fuel. We fracked"), and that the next internet-scale dataset comes from running the scientific method as reinforcement learning, with the wet lab as verifier. Science becomes an "infinite token generator" — the lab isn't the product, the model is. The counterintuitive result: one general model trained on ~10 trillion experimentally-ver

Bunkerhill Health recently raised $55M to help build a true state-of-the-art AI platform for hospitals. In this episode of Founder Firesides, YC's Ankit Gupta sat down with their co-founder & CEO Nishith Khandwala to discuss how their tools dramatically speed up hospital operations, the cold email that landed them Cleveland Clinic as their first customer, and a future where even the most complicated surgeries are managed end-to-end by agents. https://www.bunkerhillhealth.com Chapters: 00:00 — $55M Series B Announcement 00:52 — What Bunker Hill Health Does 03:50 — Why It Takes Two Years to