BREAKING NEWS: We just launched our newest products @eightsleep ✅Pod 5 ✅Hydro-powered Blanket for 2x thermal coverage ✅Base with zero-gravity elevation & surround-sound w/ @hubermanlab content ✅AI that monitors your health while you sleep Get yours👉eightsleep.com

Last week we hosted our 3rd annual @sequoia AI Ascent, bringing together 150+ of the top founders and researchers in AI. Our star-studded speakers dropped hot takes and sparked inspiration for a standing-room-only audience: - jensen huang on token-generating AI factories as the new industrial infrastructure - @sama on the timeline to agents/science/robots and ChatGPT as Operating System - @JeffDean on how he's been trying vibes coding and his prediction on the timeline to an AI junior engineer (sooner than you might think!) - @btaylor on becoming Facebook CTO at age 29, pricing/packaging agents, and advice for application layer founders - @DrJimFan of NVIDIA on why simulation is the key to robots passing the physical Turing test - @ChaseLochmiller of Crusoe on the moving industrial bottlenecks to scaling AI: power, steel, real estate - @mikeyk of Anthropic on MCP, Claude Code, and on his framework for product-building when the models themselves are changing so quickly - @hwchase17 of Langchain on ambient agents and his vision for an "agent inbox" - @danintheory of OpenAI on how RL is going from the "cherry on top" of the pre-training cake to the main course, and what questions einstein-v1907-super-high might be able to answer by scaling reasoning compute - @gradypb on the new physics of distribution, the Leone merchandising cycle applied to AI foundation models vs application startups, and his advice to startups to "run like heck" - @Konstantine on the agent economy and stochastic mindset And many more. Thanks to our community for making this the best, most buzzy, content-rich AI Ascent yet 💚 Detailed session recaps to follow -- and replay videos dropping over the course of the week.

@DrJimFan of @nvidia: * A new Physical Turing Test: when my trusty household robot can clean up the mess at home, that's it, that's the turing moment for robotics * Why is it so hard to solve the physical Turing test? Data. LLM researchers have it easy -- they have internet scale data. Robotics researchers have it hard -- you can't scrape enough data from the internet, you have to collect it yourself. It’s a slow and painful process. It doesn’t scale. It’s at most 24h/robot/day. * What to do? Enter simulation. Leave the physical world. Simulate 10,000 environments. Vary the gravity, the weights, the parameters. * Imagine a neural net could control a robot to solve 1M virtual worlds, then it could probably solve 1M+1th world. That 1M+1th world would be the physical world. In other words, the physical world is in distribution of our training data. You can train in simulation and transfer to the real world. * You instantiate a robot in dream space, and it can interact with everything everywhere all at once. Neural world models, with 2.0, will scale exponentially with compute. And a neural net will outperform a classical graphics engine. That plus classical simulation are our nuclear power for scaling. * And one day you’ll come home to a clean dinner, candlelit diner, and partner smiling at you instead of yelling at you for not doing laundry. And those robots fade into the background, like ambient intelligence. And you won’t even notice the day we pass the physical Turing test. :-)

Turn GitHub repositories into conversations👀