@sequoia
We help the daring build legendary companies from idea to IPO and beyond.
"Is it Bitter Lesson-pilled?" — the most used, and most misused, phrase in AI right now. @RichardSSutton sums it up on the latest Training Data with @kjaved_ , @sonyatweetybird, @Alfred_Lin "Don't be distracted by human knowledge, as AI traditionally has been many times. Instead, focus on learning methods that will scale with computation, like search and like learning."
Apple: seq.vc/as5 Spotify: seq.vc/bm1 YouTube: seq.vc/26v
In Iain Banks' The Player of Games, a civilization chooses its emperor by playing a strategy game. That may be the best metaphor for AI right now: multiple players, one board, and everyone at the table believes the winner controls AGI – whatever that turns out to mean. @DavidCahn6 joined @Kantrowitz on @BigTechnology to discuss: - How each player is playing the board: Anthropic, OpenAI, XAI, Google, Meta, Microsoft, and the rest - Why the $200B question is now the $1.5T question - Why nothing short of AGI justifies the spend - Whether humans will end up worshiping AI and more.
Listen to the full episode: Apple: seq.vc/oj7 Spotify: seq.vc/zx5 YouTube: seq.vc/x7o
The unwritten rule of AI research: prove your architecture at small scale, then earn the compute to scale it. @MillionInt thinks that rule may have quietly killed the transformer's competitors. RL showed almost nothing interesting until it crossed a compute threshold – the capabilities simply don't appear below a baseline. If that's true of RL, it's probably true of other architectures too. Which means there could be entire families of ideas sitting in the discard pile, written off before they were ever given enough compute to show anything at all. That's the bet behind @coreautoai: run architecture search at the scale where the answers actually show up.
@MillionInt Full episode on podcast platforms: YouTube: seq.vc/qu1 Apple: seq.vc/r02 Spotify: seq.vc/v53
Building the most automated AI lab in the world doesn’t mean removing humans. @coreautoai co-founder @MillionInt's version of automated: give each researcher maximum agency. Walking gets you some distance. A bike gets you further. A car, much further. Farming by hand works a small plot; a machine works a vastly larger one. A single researcher can now move through ideas faster than entire teams could a few years ago. The choice every lab faces: retrofit old team structures around that, or build natively for it. They chose to build natively.
A Clay teammate told a stuck customer "I'm coming to your house" – and meant it. Kareem Amin on why genuine intent built the community, not a playbook. x.com/bhalligan/stat…
If everyone woke up sick tomorrow, coding-agent usage would collapse to near zero. Almost every token today starts with a human typing "hey, go do this." @FactoryAI co-founder @matanSF's prediction is that in 12–24 months, 90% of tokens will be asynchronous – droids autonomously spotting a customer signal and showing up with a first-pass fix, no human kicking them off. We're still in copilot mode. Agent-native is next.
Full episode here: YouTube: seq.vc/ezl Apple: seq.vc/l0c Spotify: seq.vc/sgy
New episode of Training Data with @FactoryAI's @matanSF YouTube: seq.vc/ezl Apple: seq.vc/l0c Spotify: seq.vc/sgy x.com/gradypb/status…
Open ecosystem or walled garden? For @anthropicai's @katelyn_lesse and @angjiang the answer is clear. "We actually aren't precious about “You should run these things on our infrastructure." In practice, that means self-hosted sandboxes, MCP tunnels that punch through your firewall to reach servers you control, and first-class partnerships with Modal, Vercel, Cloudflare, and Amazon's new MicroVMs. The strong opinions are about architecture – not whose servers it runs on.
"The weights are open. Our dependency is not." New essay on America's open-model paradox from @DeanMeyerrr and @Konstantine : x.com/DeanMeyerrr/st…
Today's AI models train once. We don't work that way. We learn continuously, forget what doesn't matter, and retain what does. That gap is what @dan_biderman and @realJessyLin are closing at @EngramLab. AI that never stops learning, with memory that lives inside the model instead of bolted on as an afterthought. In our latest Training Data episode we get into why memory is the next frontier: why the brain forgets on purpose, why RAG is a band-aid, and what becomes possible when a model is always training. 00:00 Introduction 00:59 Always Training Explained 01:51 Beyond Context Windows 03:29 Ngram Product Overview 04:34 Adapters And Training Signals 05:32 Internalize Vs Externalize 06:49 Compute And Token Savings 08:19 Teams First Then Individuals 08:51 Memorization Vs Understanding 12:47 Dreams And Offline Digestion 14:08 Training Beats Curation 15:19 Why Everyone Needs A Model 21:44 Bitter Lesson And Architecture 24:44 RAG Killer And KV Cache 31:38 Future Of Memory And Models
12 years ago a top scientist said AI would wipe out radiology. Computer vision spread through every scan – and radiology demand went UP. Hospitals hired more radiologists. @Nvidia’s Jensen Huang on why we keep confusing a job's tasks with its purpose. x.com/Konstantine/st…
Don't mistake task for purpose. @Nvidia's Jensen Huang in conversation with @Konstantine: "Typing is not the job of a software engineer. Coding is not their job. Solving problems is their job." Jensen says AI is creating more jobs, not destroying them. Why? AI lets us solve problems better than ever.
What does AI actually change for builders: It’s not just speed. It’s ambition. The old question was whether an idea was even possible. @GoogleDeepMind's @OfficialLoganK says he now catches himself wanting to push the same idea ten steps further. The hard part becomes resetting your own ceiling on ambition. What do you build when the only constraint left is how big you let yourself dream?
"So we could edit this set?" Done and done. New Training Data with @OfficialLoganK of @GoogleDeepMind and @sonyatweetybird shows some of what Gemini Omni can do.
Listen to the full episode: Spotify: seq.vc/abo Apple: seq.vc/2ls YouTube: seq.vc/32x
Listen to the full episode here on X, or wherever you get your podcasts: Spotify: seq.vc/b4u Apple: seq.vc/3xf YouTube: seq.vc/syf x.com/sonyatweetybir…
Meet @CameronLMcCord. He's done a lot of hard things: paid his way through MIT with ROTC while double-majoring in nuclear engineering and physics. Spent ~5 years on a nuclear submarine for the Navy. He once steered that submarine through a Nordic fjord in a blizzard, in the
View more from AI Ascent 2026 including talks with @karpathy, @demishassabis, @DrJimFan , @dmitri_dolgov, @matiii, @bcherny here: youtube.com/playlist?list=… x.com/sequoia/status…
@amspector100 @bfspector @flappyairplanes More content from AI Ascent 2026 is available here: seq.vc/goj
