Lux Capital

We so enjoyed sponsoring and participating in Lux Capital Second Annual AI Summit today. It was a tremendous gathering of incredible leaders and talented entrepreneurs driving innovation from “Sci-Fi to Sci-Fact.” Thank you Lux team for inviting Banc of California to be part of this engaging event! #LuxAISummit #bancofcal

Small molecule optimization is complex, difficult, and expensive. At the Lux Capital AI Summit, Josh Haimson shared how Inductive Bio is transforming this process by combining unique data with cutting-edge ML models to enhance the quality and effectiveness of small molecule drug discovery. Thank you to Lux Capital and all the attendees today!

Our final conversation today at the #LuxAISummit was on AI and the future of modern war, with former SOCOM commander Tony Thomas and West Point Dean Shane Reeves. Gen. Tony Thomas: “The conversation invariably tilts to Skynet right away. I would like to disabuse us of that — that's a long way off, but it is comprehensive. In fact, as the focus has been on lethal effects and maneuver, it's absolutely about lethal maneuver, but it's [also] about maintenance, it's about logistics, it's about medical, it's about, more importantly, information ops. … AI will both be a positive player and a negative player for that in the future.” Brig. Gen. Shane Reeves: “Pick your area. If you want to talk about trying to solve the contested logistics problem in the Pacific; if you want to talk about targeting in a really complicated urban environment in Gaza or southern Lebanon; if you want to talk about collecting all the ubiquitous data that's all over the battlefield in Ukraine and trying to process it to the into the drone revolution that's taking place; or you talk about what probably Israel is trying to do right now with the Iron Dome, with missiles coming into Tel Aviv. At this moment, it's all going to be reliant on artificial intelligence … and what we have known — and we know — is that technology always wins. Technology always finds its way into the into the battle space — that's never not happened. There's a long history of this. The one I always go to is famously Pope Urban II in the end of the 11th century, realized crossbows were very effective, and they were good at piercing armor, and it was upsetting the societal norms of the time. So what did he do? He's like, ‘We're going to ban crossbows,’ right? And crossbows, anyone who used them, are excommunicated. So what does everybody do? Well, let's get a bunch of crossbows, right? And this has happened repeatedly, aircraft, submarines, balloons — you pick it — the technology finds itself into into the battle space.” Brig. Gen. Shane Reeves: “We are also increasingly recognizing that [West Point cadets] have to be AI-enabled officers. If they aren't AI-enabled, then we're going to lose. And that's just not a viable solution for us. So what did we have done? We've really aggressively stepped into this AI revolution at the military academy in lots of different ways. As mentioned, we have an academic year theme called The Human and the Machine: Leadership on the Emerging Battlefield.” … “They can take these skills with them, and they have to be AI-enabled, not just AI-comfortable, not just know about it, but they have to be able to embrace it and employ it, though, without having critical thinking skills atrophy. We still need them to be able to think through problems when all the technology perhaps goes away.”

Our last panel at the #LuxAISummit focused on AI x Bio, and included Maruan Al-Shedivat, PhD of Genesis Therapeutics, Mohammed AlQuraishi of Columbia, Alex Rives of EvolutionaryScale and Rahul Satija of NYU. Some top quotes: Mohammed: “Let me start with the long-term vision, which is probably a 20-year horizon. And to me, what that would constitute is basically a system, a computational artifact, that essentially reproduces anything you could do experimentally, right? So there's an experimental question that you need to do in the real cell. You could pose that question to the system, and essentially it gives you the same answer.” … “What I suspect is going to be true in the next few years are essentially, kind of digital twins of cellular behavior in more constrained settings…” Rahul: “I like that the theme of this, which is on the back wall, is ‘Impossible to Inevitable,’ which I think really highlights the idea of this virtual AI cell.” … “What an incredible opportunity that would be: half of my lab at NYU in the New York Genome Center is focused on doing experiments. We take cells, we perturb them in very specific ways, and then we run these very intricate measurement technologies to figure out what's happening. This costs hundreds of thousands to millions of dollars. We have PhD students, postdocs feeding the cells, making these measurements, and we're doing individual experiments at a time. The idea that we could do this at scale on a computer is absolutely transformative and incredibly exciting at the same time.” Alex:  “Nature wasn't created to be comprehensible to the unaided human intellect. What we want to do [at Evolutionary Scale] is really build tools that can … make biology programmable from first principles.” “We started with proteins, because there's an incredible wealth of data there, there's incredible impact. If you can solve that problem, [you can span] the hierarchy of biology from molecule all the way to cell to organism.”

Infrastructure is critical for running AI models effectively and efficiently, and we had an excellent pair talking about the subject during the #LuxAISummit: Vipul Ved Prakash of Together AI and Clem Delangue 🤗 of Hugging Face in conversation with Brandon Duderstadt of Nomic AI. Here are some highlighted quotes: Vipul at Together: “We can create synthetic data with more entropy, with all these different synthetic data methods. So I do see that being another exciting area, which is you don't have yield problems there. You can run experiments really quickly, and we'll see more and more of that.” Vipul at Together: “I think we need more power and more data centers. That's pretty clear. Like right now, it's way [too] difficult to find anything above 15 megawatts in North America. All these data centers have already been reserved and will not provide enough capacity for building and serving models. So I think we need more power. I also think the GPU power envelope is quite off.” Hugging Face’s Clem: “I also think that we can do some things to make AI more energy efficient today. I think this movement of only using and focusing a lot of our efforts on large generalist models is a mistake in many aspects. You don't need to take a private jet to go to work. In a similar way when you're doing like a specialized, customized use case, as I mentioned, you don't need a model that is going to tell you about the meaning of life. You can actually use a smaller model that is going to take less energy to train, take less energy to run. The world is a bit like biased right now, and a lot of the investment goes towards large, very energy-intensive models and directions, I think as a field, we can take a different direction and focus on specialized, customized, smaller models that give us a more credible path to continuing to build AI capabilities without ruining the planet.” Hugging Face’s Clem: “Usually what we see is that companies start with like using a large, generalist model behind an API, because it's easier, it's in a way safer, because that's what others are using. And after a few months, especially when it's production with users, and you start to see more users and the cost is starting to pile up, they think, ‘Okay, can we build different systems where we have more control, where we can optimize the models to run cheaper, faster, more focused on our specific use case, specific constraints?’ And that's usually when they when they start experimenting with other approaches, with taking open-source-based models and fine-tuning them, optimizing them, training them, and I think ultimately it's going to pay off for them, because it's a learning curve, right? It takes more time, it takes more investment. But at the end of the day, if you want to be an AI company, you have to be able to build AI, right?”

Grateful to Senator Mark Warner for joining the #LuxAIsummit. He has been a steady voice and has shown leadership on national security/defense, new energy sources and climate tech, and maintaining US leadership in AI. I also like his views on AI regulations.

One of the most important companies in the world for machine learning is Hugging Face, and we got a full update on the company and its activites from co-founder and CEO Clem Delangue 🤗. Here are a few highlights from the conversation at #LuxAISummit with Brandon Reeves: “We always felt like AI is a new paradigm to build software. So in a way, you have to do things differently if you want to be successful. And frankly, our goals as founders wasn't really to build a successful company.” … “If you end up building a big company that is similar to all the big companies that are out there today, I think it doesn't really move the needle. So we try to be very intentional about doing things quite differently. You [Brandon Reeves] mentioned how decentralized we are, both in terms of location, but in terms of responsibilities ownership. All team members have the freedom to develop their own products, their own projects, their own features. We have this framework where everyone can do that and progressively grow that into impactful projects. We didn't have for a very long time, product managers. We never had Community Manager, which is surprising for community platform.” “We envision a world and a field where ultimately all companies or organizations are going to do their own specialized, customized, optimized models. Right now, we're not exactly there, but my understanding is that we are pretty early in the technology cycle, and usually early in technology cycles, decision-making might not be the most rational one. It's more based on hype, or a lot of companies are basically doing what others are doing because it's safer. There's been a lot of rush in companies to adopt AI. And so it's going to be interesting to see if one thread that we're seeing that companies start by using API and then building their own models. Does this trend continue, and how it accelerates in the future?” On what’s outside text, audio and video? “One area that I'm also extremely excited about are time series. So for example, last week, NASA and IBM released on Hugging Face their climate prediction foundation model. And it's interesting because nobody talked about it [yet it’s] quite, quite a big deal. But the media is so focused and obsessed on NLP that sometimes they're missing some interesting things like that. It's a foundation model to be able to predict climate. So it's a very interesting example of AI for good, because on the topic, if you can actually improve your predictions, or make your predictions a few hours earlier, you actually have the ability to save thousands of life by, for example, telling people to evacuate regions or stuff like that.” “We're trying to advocate for regulation that fights concentration of power, because in AI, there are very strong natural tendencies for concentration of power in the hands of a few. And we're trying to do that all over the world.”

During the panel "The Empire (State) Strikes Back" at #LuxAISummit, we cover AI in the NYC ecosystem with some of the city's top AI leaders, including He He at NYU's CILVR Lab, Danqi Chen of Princeton NLP Group, Laurens van der Maaten at Meta's open-source ML team, and Sasha Rush at Cornell and Hugging Face. A few highlighted quotes: He He of NYU: “In our work, we find two problems with the proxy human feedback. So first, they're very noisy, either because of general human cognitive bias or because they just lack the context and incentive of real users that aim to compute certain tasks. … Another problem we saw is that these proxy human annotators are not necessarily representative of your user population. So as a result, the preference or supervision signal you get from these humans would be biased, and from a research perspective, I feel there should be more work trying to extract some implicit supervision signal from user feedback, because it's really expensive to ask users to provide annotations. Their interest is really in completing the task.” Meta’s Laurens van der Maaten: “In Llama 3, we spend a lot of time thinking through, ‘What is the tone of the model?’ right? This is pretty vague, right? And sort of hard to capture. We don't want it to be too pedantic, but it also shouldn't be too casual, right? And so those kinds of things are the things you try to navigate, and through human feedback, try to steer, and this is very much an iterative process … But part of it is also bringing in some product vision on what do you want the product to be? How do you capture that in sort of guidelines that you use in order to get our feedback?” Princeton’s Danqi Chen: “So this year, we had our newest paper, actually, first time we figured out [that] we can actually find pretty sparse circuits in large models of 13 billion parameters. That's actually a very interesting result to me. But I think the current findings from these circuits, or how far we can push in this direction, it's still pretty far away to really make them really useful. So we get these sparse circuits, but how sparse can we push for and how can we even analyze these circuits?”

What's the future of media and how will AI transform filmmaking? Lux Capital's Grace Isford talks with Cristóbal Valenzuela of Runway on what's coming next. Here are some highlights from Cris: “I always think of development in AI tends to behave much more like a like a staircase of sorts, where you have big updates, major models, new generations of things that you can do. You stay there for a couple of months. You stay there for a couple of months. You build a bunch of optimizations around that, and then you jump again.” “I think I'm less of a fan of this idea that you're going to prompt something and then you're going to get exactly what you needed.” … “ideally for us, you want to have a system, a creative system that works and aids you…” “When people speak about AI more broadly, I think there's still this misunderstanding of AI as a language, as a chatbot. And so those two things in the broader world have become synonyms. Someone might be speaking about AI when they're actually speaking about like ChatGPT or cloud and this other way around. And I think what happens with that is that it simplifies … the technology to this very simplistic output, right? And then what you realize is that, well, it's not that simple, right?” “Many people in Hollywood thinks that's the case that you go into Runway, you type ‘movie,’ enter, and you get a movie, and that's it, and the movie is incredibly well-done and well-produced. And it's like, we're out of a job, you know? And so a lot of the literacy and education we're doing with studios and production teams and creatives is like, educate them in showing them that this is not a binary tool that creates copyrighted content with no control, you actually have a lot of control, and if you don't, then you're actually probably not going to make anything good.” “This technology for us is a lot about making things that used to be very hard and expensive and time consuming more broadly accessible to more people, which means that smaller studios, independent filmmakers, and many people who never thought of themselves of filmmakers or artists, can now do the things that a handful of people could do before. And so we're trying to work with both the traditional studios and the new and upcoming smaller studios.” “I'm a big believer in New York. I mean, I love the city. I think this is the greatest city in the world. And I think specifically, depending on where you're building, it's a great way to pull in talent from different disciplines and different kind of backgrounds. For us, we daily get great people from a bunch of creative industries, from music to advertising to film in our offices, to work with us. And that's just very hard anywhere else in the world.”

Great conversation between Senate Intel Chairman Mark Warner and Lux Capital managing partner Josh Wolfe at the #LuxAISummit today in New York City. Here are some highlights and notes from the conversation: “At least for a politician, I think I've probably spent more time on AI than most others…” yet “…there's no linear relationship between time spent and actually thinking, ‘I’ve mastered the subject.’” “If I have a goal as chairman of the Intelligence Committee, it is to redefine national security. Not just in terms of tanks and guns and ship and planes, but really in terms of technology competition.” “I still would take our hand over China, but it is a it is the competition of our time.” Telco is even, AI ahead, bio is “nip and tuck”, quantum and energy are critical. “What you're also talking about is, in a sense, an AI-driven weapon system without a human-in-the-loop. So we're kind of creeping up on all of these issues that, to a degree, are ultimately existential about warfare, and the fact that we have no kind of international rules of the road means it could be a wild patch in front of us.” On AI and electoral interference: “To a degree, at least in terms of, say, deepfakes, AI has been the dog that didn't bark. We didn't see it.” “I believe with almost certainty, particularly Russia, as we get closer the use of AI disinformation, misinformation will ramp closer as we get to the election.” “My particular fear is in the 72 hours, 96 hours after the election, if you suddenly see a figure that may not be a presidential candidate, but somebody is represented as a election official, appearing to tear up ballots or stuff ballots, you could see violence in the streets.” “I’m not sure we've got evidence yet, but I would bet you almost that there is AI manipulation of the public markets going on right now. I wouldn't focus on a Fortune 100 company, but a Fortune 100 to 500 public company, all the AI tools you could use to try to manipulate that stock, way beyond deepfakes, I think, is out there.” “I do think we need legislation that makes sure that America engages with the rest of the world on AI standards. We've kind of retreated from our prior 70-year history, post-Sputnik of even if we didn't invent the next innovation here, we always kind of got to set the rules, the standards, the protocols. That was kind of a secret sauce for our country.” Message to AI developers: “We need your help and input, and we also need it at the policymaker level. Because, these are, unlike kind of my background in telecom or when we did internet stuff, this has consequences exponentially greater. I am not part of the crowd that thinks this is an existential threat, the world is ending, but I do think it'll be transformative. And I'm spending a lot of time as well thinking about when we think about job loss, how do we rethink about retraining? Can we capture some of those savings and actually reinvest in some of the human capital that needs to be retrained?”