MTS (@MTSlive)

DeepMind's Andrew Trask on why the scaling laws are pushing AI from one big model toward a protocol: "The zoomed-out picture is that in the end, AI is gonna be a protocol instead of a program. We're seeing that evolution start to really gather steam as the scaling laws constrain how much data, compute, and talent one company can bring together." "When you combine models from multiple different providers, you're implicitly combining the data, compute, and talent that they trained on. So you can get better, faster models for a lower price, which is pretty crazy when you think about it." "If you want the absolute most accurate model, ensembling the top models is always going to win, you'll get higher scores than any single model. And if you want the best accuracy relative to any unit of price, ensembling some open and closed models is likely gonna own that Pareto frontier quietly for a while." "It won't be until there's a leaderboard that widely recognizes ensembles as comparable to individual models that we start to really see it. Then in 12 to 18 months, that saturates a bunch of benchmarks across the space, and the harnesses pick it up, routing you to the best combinations of models on the fly. That's when the market really starts to change in terms of how people buy intelligence." @iamtrask @openminedorg

Tim Roughgarden, mathematician at Institute for Advanced Study and head of research at a16z Crypto, on quantum computing, blockchain as a laboratory for social science, and AI transforming mathematics right now the way it transformed coding six months ago. 1:35 — The story from Turing to 2026 3:43 — Quantum was "25 years away" for 25 years, then something changed 4:25 — Q-Day prediction + Shor's algorithm + how to spot a quantum agenda 6:37 — Positive quantum applications + we're not doomed on cryptography 9:16 — Post-quantum migration is a coordination problem 12:22 — Blockchain as a laboratory for social science 15:50 — The sociology big bang analogy 17:37 — Computation is a law of nature 19:56 — AI is transforming mathematics right now 25:54 — Start with what you want to build @Tim_Roughgarden

.@tekbog and @basedjensen on why they think Anthropic's cyber risk argument against open source AI is actually just about monopoly: @tekbog: "I'm kind of skeptical. Yeah, they just want the market and the monopoly. They want everyone to be fearful and they're pumping their numbers. That's my personal opinion of course." "I think it's good that these models can be weaponized to some degree because a lot of cybersecurity is really bad. It's a very good wake up call to systems engineers to get your sh*t together." @basedjensen: "It's a double-edged kind of thing. Yes, there is some risks. But I think there's a bigger risk having a small group of people having access to this while the defenders do not have anything." "If the defenders can actually come in and scan the code and quickly patch things up, that would mitigate the risks much better than having Glasswing or a small Chinese set of companies accessing these cyber capabilities." "It's rather better to democratize and enable everybody than to hide the capabilities. At some point or the other, it's going to leak. It's better to be patched up and take the pain now than later."

Fireworks CEO Lin Qiao on why building an application is no longer a moat and every company needs to own their own intelligence: "In the past, application itself is a moat. If you come up with an idea, implement it, launch and scale in a big way, that's moat. That's hard to do." "Today it's not hard to do. In the past, you needed tens of very strong product engineers, PMs, multiple quarters to get there. Today, one person, multiple weeks can get there. That's no longer a moat." "What is remaining as a moat is the unique product insights of what your user intent is, why they engage, what their preferences are, and how to run your business. This is your proprietary information. That's where individual companies are innovating." "Every single company should own this extremely strategic resource that is intelligence." @lqiao @FireworksAI_HQ

SITUATION EXPLAINED: Japan just formed a 44-company AI consortium aimed squarely at robotics. • Noetra is a new Japanese sovereign AI consortium anchored by Sony, SoftBank, NEC, and Honda, with 44 companies and organizations investing in total, spanning manufacturing, finance, and logistics • Backed by $6.1 billion in Japanese government funding over five years • Its data center is slated to go online in two years, with a reported order of 27,500 Nvidia Rubin GPUs, small next to Microsoft's orders, but notable for a single national consortium • The target: 10 million AI-equipped robots deployed across the country by 2040 • Plans to release a first foundation model by March 2027, purpose-built for robotics and physical AI applications • Minister Ryosei Akazawa wore a leather jacket and black T-shirt to the announcement, visibly echoing Jensen Huang's signature look • The reason for the robotics focus: Japan's median age is 50.2, meaning half the country's population is over 50 @schisofrenia: "They need robots 'cause they don't have population." @theojaffee: "It's really a problem. It's gonna be a problem for everywhere in the world, but it's gonna be a problem for Japan first."

SITUATION DETECTED: Noetra, a new Japanese government-backed venture, will buy 27,500 Nvidia Rubin chips to build a homegrown foundational AI model for robots, per Bloomberg. Allocated $2.4 billion in public funds, it will build a 140 MW data center going online in June 2028.

Infinite Machine CEO Joseph Cohen on why the $50,000 car has become completely overkill for the way most people actually move around: "A non-car is anything that isn't a car that is a vehicle that helps you move around the world. We think about non-cars as the white space around cars." "It's really hard to tell cars apart if you took the logos off of them. They do the same job more or less. And they're getting more and more expensive. The average car is $50,000 in America." "52% of trips in the United States in cars are under three miles. A car is a really overkill solution for moving less than three miles." "What we're seeing is really a Cambrian explosion of new form factors, and Infinite Machine exists to make what we think of as the best of these kinds of vehicles." @josephcohen @infinitemachine

SITUATION EXPLAINED: What happens when everyone's AI agent starts talking to everyone else's AI agent? We asked @timourxyz, co-founder and CEO of @JoinEdgeCity "The bulk of the conversation around agents and AI is very much in single-player mode. It's like, how can I use this to build my second brain, be better at my work, organize my family logistics?" "But I think it's fundamentally more interesting to think about what happens in multiplayer modes. What happens when my agent starts to interact with your agent and everyone around us, and then eventually everyone in the city, the nation, the world?" "It feels inevitable that probably in the next 12 to 24 months, everyone will have some kind of agentic entity that they're collaborating with. Probably multiple." "What happens when there are millions of agents out there linked to their human principals, but having their own agency and doing things in the world? We need to be thinking about that right now." @sodofi_

SITUATION EXPLAINED: Jeff Bezos just announced Prometheus, a new frontier AI lab focused on the physical world: • $12 billion raised, $41 billion valuation, Bezos is the primary backer • Founded alongside Vik Bajaj... the former director of Google X • Mission: build an artificial general engineer that ingests data from the physical world to accelerate engineering and manufacturing • If you're starting a frontier lab right now, you need a story that distinguishes you from LLMs, Prometheus has one @theojaffee: "Bezos has money, and he's got that dog in him."

We looked at the jobs pages of 910 early-stage startups from the top accelerator programs, analyzing who they want to hire and for how much. Explore at our latest drop:

We built a live job board of all the frontier labs hiring right now. Plus a way to see news, funding rounds, compare model releases and more. Monitor the frontier at: