Grounded (Hyper)Ambition/State of Deep Tech/ Neural Networks Need Data to Learn. Even if Fake/ Catalyzing Low-Carbon Flights/ The Sound of Wonders
Massimo Portincaso
Founder & CEO at Arsenale, Industrial Romantic and Antidisciplinarian Stoic
Grounded (Hyper)Ambition. This week I attended the Green Business Building Summit 2023 in Stockholm, organized by McKinsey. It was admittedly a strange feeling, given my history, but I am extremely thankful that I could attend, as I learned a lot, coming away with a very reinvigorated level of ambition when it comes to Arsenale BioYards, the company I am currently building.
The gathering was guided by the Chatham House Rule; thus, the details of the speakers remain confidential. Still, I am eager to share the valuable insights I gained, which have provoked deep reflection and contemplation.
First, the summit reiterated the enormity of the task before us – the transformation of the industrial and economic landscape for a greener future. This endeavor demands an investment of $9.2 trillion, approximately 7.5% of global GDP. This enormous commitment is a colossal undertaking and an exceptional business prospect. The obviousness of this point is often obscured by day-to-day complexities, and it's important to remind ourselves of it regularly.
Second, there are pioneers out there that are establishing the blueprint of industrial deep tech acceleration. The clear and best example is what Vargas is doing (and has done already) with Northvolt and H2GreenSteel, two deep tech unicorns poised to influence their industries massively, and, in the case of H2GreenSteel, defining the blueprint on how to transition a “hard to abate” industry like steel successfully.
Vargas??designed” the two ventures starting?from two "inevitables" – the demand for batteries and the necessity for steel production?decarbonization. By considering the?entirety of the?supply and value chain, and by prioritizing cost and scalability from the outset, these startups?managed to raise?billions at their Series B funding stage.?And?demonstrated?key principles any industrial deep tech startup?should?follow, like?securing orders prior to production,?building?close?relationships?with incumbents, focusing on scale and ambition from the beginning, and managing cost and scalability effectively.
Third, their truly inspiring journeys reinforced my belief that the current VC setup and approach are absolutely not equipped to drive this necessary green transition. I can very confidently state that what Northvolt and H2Green Steel did, would have not been possible with a classic VC, even one specialized in deep tech.
One of the consequences of the current VC setup is that it beats down the level of ambition of startups (particularly in Europe) and that, by relying on power law distributions, forces investors to place single-point solution bets, instead of going for end-to-end approaches, which is what is needed to achieve the cost and scale necessary to make the green transition happen.?
The above is negatively amplified, in a self-reinforcing loop, by the allergy to CAPEX and infrastructure building by VC investors. Luckily, Vargas has shown the way and made clear that it is possible to follow a different approach. One that is NOT following the classical pilot->scaleup->commercial plant template and instead is dramatically accelerating the deployment of capacity and considers “plants” as products, in a way similar to Tesla’s approach to building its factories worldwide.?
Fourth, as I was ruminating about this last point, after the conference, I started realizing that maybe the current “ambitions beating” situation is also a consequence of the end of zero-interest-rate policy (ZIRP) and the consequent bearish market conditions. In fact, in 2020-2021, fueled by ZIRP, we had many instances of too much ambition… but then quickly realized that the dynamics at play there were not those driven by the right type of ambition…?
So, one question spontaneously arose: How to conjugate an insane level of ambition with enough humbleness to be aware of the challenges ahead and avoid hubris and greed, as was often the case in the ZIRP years…?
And this is when I remembered the words of a former DARPA director, who said that she was always looking for people with their heads in the sky (i.e. visionary) but their feet on the ground. And I think this is the biggest challenge for those visionaries with the aspiration to redo the economic and industrial tissue of modern civilization and drive the green transition.?
They will have to be almost schizophrenic: on one side have insane levels of ambition, while grounding it in the humbling reality of engineering new supply and value chains. It is not a small feat, but Vargas, with Northvolt and H2GreenSteel, has shown that it is possible. Grounded (Hyper)Ambition. That’s what is needed.
Bessemer Venture Partners State of Deep Tech surveys “what innovators are building today” and announces the “XB100, the definitive ranking of 100 of the most promising deep tech startups.” Bessemer’s "favorite definition of deep tech is technology that was science fiction in the past but is reality today.” Historically, governments have played a crucial role in “building deep tech — for example, rockets and the internet.
However, in the early 1980s, R&D spending by private sector companies exceeded R&D spending by the US government. In 2019, private R&D spending exceeded government spending by almost 220%.
In compiling its?XB100 list ?of the “world’s leading private deep tech companies,” Bessemer assessed the businesses based on four criteria:?impact on humanity;?valuation;?scientific?difficulty; and?commercial?traction?and broke the companies down into 9 deep tech categories:
Forbes offers its own take on the?XB100 here .
News items:
Rep. Brandon Williams (R-N.Y.), recent Chair of a House Subcommittee?hearing on fusion , believes the “process… is finally within our grasp.” Now, “commercial fusion is more than just a science experiment — [it’s] an engineering challenge similar to putting a man on the moon [and] an economic opportunity even greater than the California gold rush.” Williams said, “My journey from a fusion-skeptic just a few months ago, to a fusion-optimist, surprised me.”
He is now convinced that the US “must dramatically increase our public investment… in commercial innovation… to ensure that we are the first to harness this history-changing energy source,” pledging to “work tirelessly with my colleagues on both sides of the aisle to bring US fusion forward from research to “fusion on the grid.”
Machine learning and AI models’ “insatiable appetite for training data” is making real-world data “expensive, private, or in short supply.” As covered in this section of the Antidisciplinarian last week, large companies, the military, and especially startups who lack the resources to buy from data brokers are increasingly turning to synthetic data to train their neural networks and AI/ML models. “ML has long been struggling with the data problem,” said?Sergey Nikolenko , Head of AI at Synthesis AI. “Synthetic data is one of the most promising ways to solve that problem.”
One area where real-world data has proven?highly problematic ?is?facial recognition , raising issues of privacy and bias. Microsoft’s?Mixed Reality & AI Lab ?has attempted to address such concerns by “releasing a collection ?of 100,000 synthetic faces” for AI training. By mixing and matching from the scans of 500 peoples’ faces who gave their “permission,” Microsoft generates “new and unique combinations” that researchers claim “spans a wide range of ethnicities, ages, and styles.”
Tadas Baltru?aitis , Principal Scientist at Microsoft Computer Vision, says, ”There’s always a long tail of human diversity. We think and hope we’re capturing a lot of it.”
News items:
领英推荐
At a?recent presentation ,?AMD ?CEO Lisa Su revealed new details about its forthcoming AI chip that AMD claims will challenge market leader Nvidia Corp. The MI300X will soft launch in Q3, with “mass production beginning in Q4. The MI300X will have 192GB of onboard memory.
Su claimed, “The more memory that you have, the larger the set of [AI] models” the chip can run. We've seen in customer workloads that it runs much faster. We really do think it's differentiating.” One thing that did differentiate the MI300X announcement from past AMD product launches was that Su “gave no details on how much the chip will cost” — or who will buy it. AMD shared closed down 3.6% after the AI presentation.?Kevin Krewell , Principal Analyst at TIRIAS Research , said ”I think the lack of a (large customer) saying they will use the MI300X may have disappointed the Street. They want AMD to say they have replaced Nvidia in some design.”
Aviation is already responsible for “2-3% of annual global CO2 emissions.” That number is expected triple by 2050 due to increased demand. “Electrifying ground transport is well underway,” but the same cannot be said for aviation. The most likely “lever” to reduce reliance on fossil fuels and emissions is sustainable aviation fuels (SAFs), typically produced from biomass feedstocks or renewable energy. Early biofuel experiments using vegetable oils “have largely been grounded” to environmental and food security concerns.
But promising new tech from?OXCCU , which recently?raised $23M , may be able “to create SAFs from CO2, electricity, and water” at significantly lower costs than existing methods. In this interview with OXCCU co-founder and CEO Andrew Symes, he discusses the company’s “business model and strategy for scaling SAFs.” By simplifying the “two-step process” of converting CO2 and Hydrogen into AF “into one step,” OXCCU believes it can substantially reduce capital and operating expenses. Symes says OXCCU’s team “sees ourselves as a CO2 catalysis company and reactor design company, rather than the product developer,” and its long-term business model — after building a few proofs of concept plants — relies on licensing its tech to other companies, rather than “selling the fuel.
News items:
There are currently about 15OK public EV charging stations across the US. By 2030, 2.1M+ charging stations will be required to keep “an expected 28 million electric vehicles on America’s roads” — an increase of 1300%. Recent alliances between Ford, GM, and Tesla will make charging cross-compatible across EVs from each manufacturer. In the future, Ford and GM will “build their vehicles with charging ports based on Tesla’s design standards,” offering seamless charging interoperability.
Tesla’s chargers “are widely regarded” as being the most reliable. According to former Telsa exec?Jon McNeill , now a GM board member and founder at DVx Ventures: “Ninety-nine times out of 100 times you go to a {Tesla) charger, it works.” McNeill estimates chargers from other manufacturers fail about 25% of the time. Amazon-backed “electric truck and SUV company” Rivian also?just announced ?it too will adopt Tesla’s North American Charging Standard(NACS) .
As Pixar recently it’s much easier to start a?real?fire than to simulate “the ethereal nature of fire onscreen.” Since one of the stars of its recently released film, Elemental, was “actually supposed to be made of the stuff,” capturing fire convincingly “was a real conundrum.” Pixar teamed up with?Disney Research Studios . Paul Kanyuk, Technical Director at Pixar Animation Studios, said, “Many of us didn't do machine learning until it started becoming prevalent recently, so we've kind of learned on the job, whereas the research coming out of the Disney lab—they live and breathe this stuff.”
Even with the DRS experts on board, the compute “requirements went up and up” until becoming ”an absolutely monumental task, especially considering that the process required a whole lot of graphical processors.” Rather than compromising on Ember (the flaming lead), the team came up with a novel solution. “We ended up getting a 20X speed-up from [where] we started… by tapping into the GPUs everyone at Pixar has on their computers. We figured out a way to virtualize the GPU and take half of it to use overnight, making the time to render a frame go from about five minutes to one second.” And “it worked.”
News items:
You won’t see GenAI collecting a Grammy award anytime soon — at least not on its own. The Grammy’s?updated rulebook ?states: “A work that contains no human authorship is not eligible in any categories.” But not all AI use in music is banned. “A work that features elements of [GenAI] material… is eligible in applicable categories. However, “the human authorship component of the work submitted must be meaningful and more than de minimis.”
Ever wonder what a black hole sounds like? Or a light curve? Thanks to a process known as?sonification , we are beginning to understand?what the universe sounds like . According to?Matt Russo , assistant professor at U of T: “Sonification is the use of non-speech audio to convey information. It involves converting data into sound so that the relationships within the data can be perceived, explored, and enjoyed with our exquisite auditory system.” Though sonification was largely?developed ?“to make [astronomic] data accessible that wouldn’t otherwise be” to the “blind and visually impaired (BVI) community… similar experiments are underway in chemistry, geology, and climate science.”
Not only can sonification make data more accessible to the visually impaired, but “some astronomers realized that sound is a powerful way to communicate the wonder of the cosmos to a wide audience.” The?sounds of space ?have collectively been heard on YouTube and Soundcloud “hundreds of millions of times,” and a vinyl LP of the?Chandra X-ray Observatory ’s “greatest hits” has seen?release on vinyl LP .
Non-astronomy-related use cases for sonification include everything from?SoundPAD Lab ?at UF’s work with the Gainesville Fire Dept to develop a system that uses sound to help firefighters navigate through smoke-filled buildings. In that situation, everyone is visually impaired.”
News items:
Recently, Amazon launched an AI charm offensive at Dearborn STEM Academy where SVP and Head Scientist?Rohit Prasad ?“assured students there would soon be millions of new jobs in AI.” Prasad then presented Boston Public Schools with a $25K grant. Across town MIT President Sally Kornbluth delivered “a more sobering message to students from other local schools, saying, “Because A.I. is such a powerful new technology, in order for it to work well in society, it really needs some rules. We have to make sure that what it doesn’t do is cause harm.” Both workshops were put together by an MIT?“responsible AI” initiative .”
Massimo Portincaso Thanks for Sharing! ?
Knowledge Worker
1 年Thanks for this interesting article! Data Sonification is very promising: not only the Universe and the position of the planets can be turned into musical notes but also our DNA, the brain waves of humans and animals, the EEG of plants, the colours. The whole Universe is musically shaped.