12/ What is most exciting is that we have uncovered a signal deep in the brain that richly represents a stimulus you’re not aware of. This finding, in my opinion, makes it tractable to understand how a pattern of neural activity accesses consciousness. The previous dominant idea--that suppressed stimuli become progressively weaker as you go anterior (“fail to ignite”)--left the question of how a representation becomes consciously accessible extremely hard to pin down. Whereas now, with whoppingly clear signals corresponding to both what is and isn’t conscious, we can follow both sets of footprints, to track down the full machinery for generating conscious percepts.

Real life, baby

study material science; every serious technology bottleneck eventually collapses into materials. performance limits. cost curves. manufacturing yield. reliability over time. all material problems. better batteries aren’t software problems. lighter robots aren’t algorithm problems. faster chips aren’t “ai” problems. they’re crystal structures. grain boundaries. defects. phase transitions. stress, fatigue, heat. engineers who ignore materials design abstractions that fail in the real world. engineers who understand materials bend reality. steel built empires. silicon built the internet. lithium reshaped energy. composites reshaped flight. the next leap won’t come from dashboards. it’ll come from atoms arranged correctly. learn how matter behaves. everything else stacks on top.

Of the past 312 orders we could fulfill in 48 hours, we won 79.5% of them. Crazy. I am sure this is true for parts, valves, components, etc. Go build!

why dont any websites feel like this

DC Deep Tech Week has officially wrapped. Across six days, the program drew 3,691 unique registered attendees—including 695 founders—to 26 public events throughout Washington, DC and Northern Virginia. Twenty-eight startups and 47 organizations presented across a week of founder talks, technical presentations, panels, demonstrations, screenings, roundtables, showcases, startup crawls and community gatherings. Thank you to the hosts, organizing partners, venues and sponsors who made the week possible: 8090 Industries; Air Force Research Laboratory (@AFResearchLab); AKULAR (@AKULAR_AR); Arlington Economic Development (@AEDBizInvest); Atypical Ventures (@atypicalvc); Auron Solutions; Brave Capital (@brave_cap); Cornell University (@Cornell); Fulcrum Venture Group; HAUS (@deeptechagency); and Hyperstition Incorporated; JBG SMITH (@jbgsmith); Johns Hopkins University Applied Physics Laboratory (@JHUAPL); KdT Ventures (@KdT_Ventures); Level Five VC (@LevelFiveVC); National Innovation Quarter (@NationalIQ); New Vista Capital (@NewVistaCap); Point72 Ventures (@p72vc); Ramification Laboratories; Red Cell Partners (@RedCellPartners); Squadra Ventures (@squadra_vc); and SRI International (@SRI_Intl); Station Alpha; The Capital Factory (@CapitalFactory); The Nuclear Company (@TheNuclearCo); VC in DC; Veteran Ventures (@VenturesVeteran); Virtus Innovation Center; WIseCost; Armored Fresh; Statecraft (@statecraft_labs); and Anomalous Heat LLC. The energy in DC was incredible - there are people shoulder to the wheel at the highest leverage positions in the world accelerating a science fiction future into reality. See you next year at DC Deep Tech Week 2027, mid September. deep-tech-week.com/washingto…

New role. Same mission. I’ve joined @LevelFiveVC to invest in deep tech deployment that moves civilization forward. Build. Ship. Make the future better. Let’s go 🚀

Titanium … Does. Not. Corrode. Obviously we continue to build and fabricate the vast majority of our buildings, equipment and working assets from steel. Steel is an incredibly versatile material, we are very good at using steel. Steel is great. Steel corrodes Things that are made of steel are temporary, the Forth Bridge (my pfp) is a cantilever rail bridge built in 1883. It was constructed from 51,000 tonnes of steel. The Forth Bridge is painted from end to end every two years, and it takes two years to paint the bridge. The bridge is perpetually being painted to protect it from corrosion and that’s why it is still standing after 143 years. Titanium does not corrode, in fact it is so inert that titanium is used for medical implants. There are people walking around with titanium plates in their skull. Quite a staggering feat of corrosion resistance. If a society really wanted to build lasting monuments we would use stone and titanium. 1,000 years from now the only surviving buildings of this era will be the ones that did not use steel. In many countries the suburbs are more durably constructed than the metropolitan centres. We don’t use titanium because it is expensive, and it is expensive because we still rely on the Kroll process to process titanium ore. 200 years ago aluminium used to be very expensive, kings and queens would use aluminium cutlery as a display of opulence. Today we figured out how to process aluminium very cheaply and now we make 1 billion aluminium cans every day. Titanium is still stuck in the expensive-to-process phase, even though it is the 9th most abundant element on earth. Soon someone is going to solve titanium processing and the aesthetic of every city will change dramatically and things will start feeling very futuristic.

This week, @ActinideInc became the first startup in history to enrich uranium, producing HALEU, the fuel for next-gen and small modular reactors. We did it using Endurance, a particle accelerator that we designed and built in Texas. Endurance usually produces enriched Yb-176, a precursor for cancer therapies worth roughly $30,000 per gram. After a bulk Yb-176 delivery on July 30, we reconfigured Endurance for uranium.

A list of things that I think will happen soon in the deep tech space / in general after drinking too much coffee today: The time of non-silicon / conventional computing is upon us. There will be a massive influx of capital into computing platforms that are either radically different, or else, use existing mature platforms in novel ways. I would say adiabatic / reversible computing like @VaireHQ @rodolfor, probabilistic / thermodynamic asics like @extropic @beffjezos, silicon photonics like @LightmatterCo and others. Why? Moore's law is petering out in various ways, but more simply the clock speed of compute flat-lined a long long time ago. A silicon transistor is like a tiny current amplifier, and there is a maximum refresh rate beyond which the amplifer gain is below unity. That, and the chemistry of the charge depletion zone means the electrons are always bouncing around, causing heat, and indeed thermal transport and clock speeds are two limiting factors on compute density. More generally, however, is that probabilistic-based AI models running on deterministic finite gate logic is a fundamental mismatch. If you wanted to e.g. implement energy based models, there is a far far more elegant and intuitive way of implementing this by leveraging the physics of a system to represent the energy landscape directly. Civilization will do everything it can to maximize compute density across all metrics - volumetric, weight, energy, etc. The fastest implementations likely leverage existing CMOS / silicon fab tech but use the same things in completely new ways. Biotech is not even 1% started Biological systems are by far the most advanced technological platform in the known universe, something that has been optimized over billions of years in parallelized evolutionary algorithms. Biology is the strong-form of nanotech - human beings unpack via self-assembling molecular turing-complete computers that can take something smaller than a grain of sand and grow it into a physical form factory that has 100x the performance capabilities of any known robot, or any known AI model, in versatility and adaptability. Even a trivial example like insects represents a kind of robot that humanity is an unimaginably far distance from being able to design from scratch, but this distance is massively compressed in the era of AI. The limit on the speed of development in this area is fundamentally data acquisition through more and more modalities across the biological spectrum. Transcriptomics and sequencing do not give a complete picture - things like light sheet microscopy (@loicaroyer ), raman scattering (@Precigenetics), and other optical methods that can provide volumetric time-resolved 4D datasets on living biological systems are the lynchpin in reverse-engineering this branch of nanotech, and large-scale automated biology factories, like Gingko Bioworks (@jrkelly), need to produce several internets-worth of training data to decode biology. Things like therapeutics, peptides, cancer vaccines, and so on are the highest value but only first wave of applications of biotech. Biotech is a general technology - it is the most dense and versatile form of robotics imaginable, the most energy and resource efficient means of manufacturing, and there is really nothing in physical law that says you cannot grow a car, a building, a factory via biological nanotech. AI Labs will become Everything Companies Anthropic and OpenAI will massively expand into all areas in which AI has high leverage, into things like self-driving robotic laboratories for material sciences, biological discovery, but even further - into manufacturing, metallurgy, design and simulations. One month's revenue can king-make them as the best-funded player in any one of those categories. We think it is wild that Jeff Bezos could raise $12B to build Prometheus, the vertically integrate design and fabrication engine. This is probably four weeks of revenue for Anthropic in 2027. By 2028 I would not be surprised if frontier labs are planning to vertically integrating the entire techno industrial stack, from raw minerals extraction, refining, chip production, to space programs and rocket designs. It will not be humans designing and implementing these systems, all the capital pressure in the universe will be on removing humans from the loop in any of these domains to accelerate the volume of compute being brought online. Robotics and Manufacturing will produce a generalized API to the physical world We think it is amazing that an ai agent can tool-call notion or slack, or now click around on a screen and debug a browser. Consider now if that same agent could order 100 tons of raw materials and transform this into a house, an apartment block, a functioning one-off multi-ton casting press, and so on. The strong form of robotics as applied to manufacturing means that humanity and machine intelligence both will have a software-defined API to the physical world that will unleash a tidal wave of abundance that is beyond imaginable. It is convenient to have something summarize your emails, it is something else when the cost of every physical good falls to the marginal cost of energy required to produce it, and in general, everything will decline to the cost floor of energy.

A quick process reference for everyone looking to sell to the DoW

It’s rude to ask - a woman her age - a man his salary - a robot deployer how many robots they’ve deployed for non-research use cases