Hadsell spoke at the first Research and Applied AI Summit in 2017 and returned to open the tenth. She is now VP of Research at DeepMind, where she co-leads a Frontier AI unit of roughly a thousand researchers and engineers across about a hundred projects, and she did not start in machine learning: her undergraduate degree was in philosophy. Her talk was, in part, a corrective.

The case against one token at a time

Less than 48 hours before she spoke, DeepMind released DiffusionGemma, a new open model in the Gemma 4 family, under an Apache 2.0 license. It is a 26-billion-parameter mixture-of-experts model, and it generates text in a way most language models do not. Instead of producing one token after another, it borrows the recipe behind image and video generation: take a sample, add noise, and learn to denoise it. Applied to text, that means generating a whole block at once rather than committing to each word in sequence.

The mechanism buys two things. Inference is faster, because the model is not walking left to right. And because it writes and revises a whole block, it can reason in both directions and correct itself. Hadsell showed it solving a problem while displaying its work: after the first denoising step it commits to the wrong answer, by the second it reconsiders, and by the third the whole block is consistent and right. An autoregressive model, having locked in each token as it went, cannot do that. It is the reason, she said, that a left-to-right model is structurally ill-suited to sudoku, which demands seeing the whole grid and reasoning across it at once. Diffusion also delivers adaptive computation as a side effect: easy questions converge in a couple of steps, harder ones take more. In practice it is fast enough to keep up with you in real time, writing, running, and testing code as quickly as you can describe what you want. Still, DiffusionGemma is an experimental release rather than a replacement for the standard autoregressive Gemma 4 models, and text diffusion carries trade-offs of its own. DeepMind shipped the fine-tuning code alongside the model.

Better agents need better worlds

Hadsell traced DeepMind’s world-model work back to a simple idea: you get better agents by building better environments, ones rich enough to keep generating new situations to learn from. A world model simulates the dynamics of an environment, and DeepMind sees these as central to the path to AGI because they supply unlimited, rich training worlds for agents.

The line runs from Genie-1, trained in 2023 on 2D platformer games and published in 2024, to Genie-2, which moved to latent diffusion and produced interactive 3D environments that were still synthetic and not real-time, to Genie-3, which is high-resolution, works across domains, and runs in real time for several minutes while staying consistent. You feed it a text prompt and get back a world you can move through. Prompt “a country lane in Kent, England,” add rain, and the model fills in the person walking, the puddles, and a pair of yellow boots she is fairly sure she does not own. You can change the world mid-stream by typing a new prompt into it.

Last month the team grounded Genie in Street View. They take one of Street View’s static panoramas, turn it into a short video, prompt in changes such as winter weather or a character, and feed that in as the opening frames of a Genie world, enough to walk across the Golden Gate Bridge and then dive underwater while the model keeps the bridge’s geometry intact. Hadsell is most excited about education: being present at a historical event instead of reading about it, or following a red blood cell through the circulatory system, with the model tied to a textbook chapter so that, in her words, “no dragons will come in if we don’t want them to.”

Robotics is still the hard part

Hadsell has worked on robots for more than a decade and is blunt that the field is hard, an interplay of hardware, software, and understanding, with the goal of general-purpose machines that can drop into any task a human can do. DeepMind splits the problem the way a body does, into a brain and a spine. The spine handles the fast, fine motor control that should not have to route through the brain; the brain handles understanding and planning. Two models carry this. Gemini Robotics-ER 1.6 is a Gemini model tuned for spatial understanding and tool use, able to pick individual objects out of cluttered scenes. A separate vision-language-action model takes the visual input and the reasoner’s instructions and outputs motor control. Working with Boston Dynamics, they found tasks that needed more than spatial sense: getting Spot to read a small gauge required tool calls to magnify the dial and reasoning to interpret it.

The frontier, she said, is efficient learning. The ways to collect robot data run from expensive, narrow teleoperation, through physics simulators like MuJoCo, to the prize of learning directly from internet video. World models now sit in that chain. They can generate novel interaction data at high enough fidelity to train the robot model, and Hadsell said you cannot tell the real robot logs from the simulated ones. Pool that with cross-embodiment learning, where data collected on one robot transfers to others, and the performance gain comes “for free.” This is where the talk’s argument closes on itself: the same foundation-model recipe, pointed at worlds, becomes the engine that trains the robots.

Stop staring at language

Her closing note returned to the frustration she opened with: almost all the field’s attention sits on language, and she thinks that is a waste. The same recipe that produced today’s language models can be pointed at far richer systems: worlds, robots, biology, weather, and the diversity of Earth’s biosphere at planetary scale. The data already exists; the models have barely been built. The next decade of progress, on her account, runs through those systems, and has little to do with making chatbots bigger.

The continent lost the sky

Hadrien arrived in Ukraine in the first week of March 2022, a week after the full-scale invasion. What he saw was a kind of war NATO had forgotten how to fight: “a high and long intensity warfare,” not the small expeditionary operations in Africa and the Middle East where the West always owned the air. For the first time in modern European history, he argues, “Ukraine, but also NATO, doesn’t have the upper hand on the air superiority.” The weapon that took it away is almost insultingly cheap. As Hadrien traces its lineage, the Shahed-136 is a slow, low-flying design first drawn up by German engineers, then copied in turn by the Israelis, the Iranians, the Russians, and now the Americans. “Nothing changes through the years,” he said. The economics did.

Most of the time, it misses

I asked him to walk through the data loop, the thing every speaker that day had described from behind a laptop: get representative data, run the model, grade the output, go again. Alta Ares runs it in a field at night. “Most of the time it doesn’t hit,” he said, and that is the whole problem. There is rarely internet at the edge and never a data center, so everything runs on “super compressed” models on small GPUs inside the interceptor. A mission has three phases. Radar detection sorts friendly from hostile. Then an interceptor flies into a kill zone roughly two kilometers wide, where an RGB and an infrared camera hunt the sky for a single pixel that is genuinely hard to find, because the feed is low quality and the airspace is being jammed. Then a human, still in the loop, commits to the intercept. One target is tractable. A swarm is not.

The loop only tightens if you close it fast. Alta Ares keeps around twenty engineers in Ukraine, as close to the front as Hadrien can put them, and combat-tests “every week or every two weeks.” The hardware turns over constantly; the vision models update remotely as more data comes back. They do not call the things drones. “We call them munitions, because it doesn’t come back,” he said. There is an 800-gram warhead in the nose. “I don’t recommend them to come back.”

What no simulation can reproduce

Simulation, he allowed, is good for a proof of concept and little more, and the reason is human rather than technical. The engineer who runs a test slept well, had breakfast, had a good coffee, “almost as good as the tea in England.” The operator who fires the system in Ukraine is on their fifth coffee “that doesn’t wake you up anymore,” at 3 a.m., freezing, running on adrenaline, and far less trained. No amount of lab time closes that gap, and no simulator renders the adversary. The Shaheds his team intercepts have started carrying rear-facing sensors that trigger evasive maneuvers when an interceptor closes in. Others fly in a mesh, so when one is shot down another inherits the firing solution. Much of the underlying technology, he said, is Chinese, routed to Russia and Iran. The lesson Hadrien draws is that classical computer vision is not enough against an opponent that keeps learning, and the only thing that trains a system to beat it is real data from real intercepts, which no simulation can manufacture.

Quantity is the quality

So why isn’t modern air defense everywhere? Because NATO and Europe bought small numbers of exquisite systems on the assumption that they would always own the air, and that assumption is gone. “The new reality is that the quantity is the quality,” he said. A sophisticated interceptor fielded in low numbers gets outgunned. Israel’s Iron Dome works because Israel is small to defend, and even so, by his account, “Israel used 35% of the world’s stock of Patriot missiles in only the first 10 days” of the war. Interception is never a sure thing: “never 100%,” he said, somewhere between 25% and 75% on the best systems. The fix is as industrial as it is technical, and the budgets are still pointed the wrong way. France is spending more, he noted, but just committed 10 billion euros to an aircraft carrier that, he argued, takes a decade to build and that a hundred cheap unmanned boats could sink. He put the politics of it in one line: “during peace time, you have time, but you don’t have money. And during war times, you don’t have time, but you have money.”

The talent turn toward European defense

Talent has become his strange dividend of the moment. “One of our best talent acquisitors, his name is Donald Trump,” he said. When JD Vance told the Munich Security Conference that American and European interests had diverged, well-qualified engineers at US companies in London started leaving to build European defense instead. What he sells them is accountability and proximity: write something in the morning, watch it deployed on the front by night. Alta Ares does no offensive weapons, “so far.” And in his telling, this work is the precondition for everyone else’s. The clinical trials, the healthcare, the AI-for-good that had filled the rest of the day are “possible because we live in a country at peace.” Bomb the data center and none of it ships.

Pressed by an audience member to separate the founder from the person, he couldn’t, and didn’t try. He reached for the old Roman maxim, si vis pacem, para bellum: if you want peace, prepare for war. Europe stopped preparing, and “the use of force is being democratized” faster than its institutions have noticed. You can annex territory now, he said, and no one does anything about it. He was honest about the discomfort of his own business: “I would sleep way better if tomorrow the war in Ukraine stops.” But he doesn’t think it will, and his case for rearming rests on a grim simplicity. The people on the other side keep telling us what they intend, so “we should listen to them, because at least they are being transparent.” Freedom doesn’t come for free.

That was once a hard sell. When Odyssey pitched its seed round, he says, “basically no one understood world models,” and convincing people was “an uphill battle.” The category has since shifted underneath him: the argument flipped around November 2025, money and researchers flooded in, and at NeurIPS 2025 world models were, by his count, the field’s number one or two theme. Hawke came to the problem from self-driving, as a founding engineer at Wayve teaching cars to drive end-to-end - one of the few arenas, he says, where frontier AI has really been tested against reality.

Not every world model is a world model

The term is loose enough to mean almost anything, so Hawke was careful to pin it down. Odyssey uses it in the precise sense inherited from model-based reinforcement learning: a learned transition-dynamics model, the thing David Ha and Jürgen Schmidhuber named in their 2018 “World Models” paper. In plain terms, a model that learns how the world evolves and then samples possible futures one step at a time, conditioned on the actions you feed it. That is distinct from “spatial intelligence,” which models the appearance and structure of a scene (the pitch at World Labs and others); from behavior models, the decision-making brain of a self-driving car or a robot, which Hawke thinks is better named as such; and from the “proxy world models” some researchers argue are already latent inside LLMs. Odyssey’s version is what he calls a general-purpose neural simulator: an interactive stream of pixels that models physics, that you can talk to and that talks back, and that you can reach into and change. “I have many questions about what this means for future products,” he admitted. “I have very few good answers.”

Generality wins

Two principles guide what Odyssey builds, both carried out of self-driving. First, end-to-end learning, a simple model trained purely from data, “almost always wins.” It was an unpopular position in 2018 and is now the default in autonomy and in language alike. Second, “generality wins”: narrow models built for a single vertical rarely keep their lead for long. He couldn’t point to durable legal-specific foundation models, for instance, because “Claude just got better.” The ambition that follows is unsubtle. Odyssey wants to build “the GPT-3 of world models” - the InstructGPT-style moment when a category of model stops being a research demo and starts generating real commercial demand. Odyssey’s research splits into four problems, each with a shipped model against it.

Pixels that keep going…with sound too!

The foundation is autoregressive interactive pixels, embodied in Odyssey-2. A standard video model returns a fixed clip; this one streams frame after frame in real time, which is much harder, because error compounds as the model generates in sequence. It is also interactive, absorbing text prompts mid-stream and adjusting to them. Scaling helps in the usual way - a one-billion to a fifteen-billion-parameter jump lifts the benchmarks - but the real difficulty is keeping the interaction open-ended instead of narrowing it to a single domain.

In May 2026, Odyssey released Starchild-1, named for 2001: A Space Odyssey, which generates pixels and audio jointly in one coherent stream rather than dubbing sound onto finished video. It was harder than expected, Hawke said; to his knowledge no one else had shown one publicly. The obstacle is a clash of timescales - a single predicted video frame spans only “half a phoneme” of audio - so keeping the two coherent took a custom KV-cache design running on two clocks at once.

Shared state, not stitched video

The third problem is multiplayer: shared state across world models, which matters as much for a cell of robots working one environment as it does for a game. Odyssey’s Agora-1 demonstrates it, and Hawke ran it live. The audience pointed their phones at a QR code and played a fully generated game of GoldenEye, streamed off H100s “probably in Spain.” Under the hood it borrows a game engine’s split between rendering and simulation, except both halves are learned - a neural simulator and a neural renderer, trained together. People had said multiplayer world models “weren’t possible,” he noted, “and we felt it was worth disproving.”

Learning by being broken

The fourth strand carries the freshest idea in the talk. Almost everyone who pairs agents with world models uses the agent to get smarter inside the model. Odyssey’s PROWL turns that around: it uses a reinforcement-learning agent to improve the model itself. The reasoning is “garbage in, garbage out” - “your agent will never really be better than the quality of your learned environment model” - and yet, Hawke argued, almost no one had bothered to fix the model rather than the agent. PROWL drops an agent into a learned version of Minecraft and rewards it for finding the world model’s failure modes, then folds those failures into a curriculum that patches them. The wrinkle is that reinforcement learning is very good at cheating: left unleashed, the agent would “sit on the spot, spin the camera around at very high speed” to stall the model’s learning, so the team added a KL “leash” to keep it close to an agent that actually plays the game. The payoff was a clear lift in the base model’s performance - a world model that gets better by being systematically broken.

Still the GPT-2 era

For all of that, Hawke was disciplined about where the work sits. World models, he said, are at “the GPT-2 era”: the pre-ChatGPT stage where the outputs already look promising and the first commercial experiments are forming - he name-checked Jasper, the early GPT-3 copywriting startup - but mass adoption is still ahead. That is the gap between today and the “GPT-3 of world models” he wants to build. The honesty extends to limits. Asked from the floor whether a world model would have to encode general relativity and quantum mechanics, he declined the bait. Visual data is “by far the highest volume” and the right place to start, and these models earn their keep where conventional simulation struggles: for a precise definition of turbulent flow, “use CFD”; for crowds, contact, the messy texture of a scene, the neural simulator wins. Biology and harder physics are wanted, but, he conceded, nascent.

The economics are gentler than the hardware implies. The multiplayer demo ran on a single H100, “let’s call it $1.50 an hour” - close enough to a Netflix subscription that, by his estimate, $30 a month buys ten to fifteen hours of generated play.

Where this goes, on his telling, is convergence: as a visual world model takes on audio and, eventually, text, it meets the language models arriving from the other direction through VLMs, and at some point the two merge. None of that is solved, and he was candid that no one yet knows how. But the direction is set, and the market has caught up to the thesis that once needed an uphill seed pitch. On June 17, 2026, days after RAAIS, Odyssey raised a $310M Series B at a $1.45B valuation, led by Natural Capital with Amazon, AMD Ventures and GV. Air Street backed the company’s seed in 2024, on the bet that learning the world directly would become its own category of model. The GPT-2 era doesn’t last long.

Read more!

A science, not an art

Ted is firm on the framing. These systems are complex, he says, but they are understandable. The catch is that understanding them demands “epistemic humility,” because there are always too many variables in flight and “running a controlled experiment is hard.” The skill he prizes most is knowing the limits of your own evidence: “knowing what an experiment tells you and what it doesn’t tell you is important.” That sounds modest. In a field that markets every result as a breakthrough, it is closer to radical. His own path ran through linguistics and neuroscience, but the thing that carried over, he’s clear, wasn’t brain-inspired architecture - it was the scientific method itself: rigor, skepticism, doubting your own results.

The job also changed his relationship to curiosity. In a PhD you pull on a thread because it’s interesting. In a frontier lab the question is colder: what is the cost-benefit, and is the answer going to be interesting but “ultimately less useful to making Claude smarter”? Experiments that survive that test get fed up to the people deciding the big training runs, where the point is to cut uncertainty before committing the compute.

The metric that matters is the counterfactual

I put a striking set of numbers to him, drawn from Anthropic’s recent essay When AI builds itself: more than 80% of the code merged into Anthropic’s codebase is now authored by Claude, a typical engineer merges eight times as many lines of code per day as in 2024, and on one kernel-optimization task Claude’s speedup climbed from roughly 3x to 52x in under a year. The essay’s own line was that “we have not yet seen that curve bend.” So what, I asked, is the honest measure of AI actually accelerating?

Not benchmarks, Ted said. The real test is “what changes the counterfactual” - if you took Claude away and coded alone, what would you do differently? Concretely, it reduces to the “number of human interventions that are required”: how often a person has to step in, how often the model’s first pass is the one you accept. By that measure the line has moved fast. He pointed to Opus 4.5 in November as a jump, and the Mythos-class models in February as a bigger one. The shift is trust: the models still make mistakes and you still have to check them, “but you can trust them a lot more than you could before.” In the labs, people have already stopped supervising every step - they run in bypass mode and let the agent work.

When bigger models get cheaper

Does the future belong to one large model doing everything, or to companies decomposing tasks across smaller, cheaper ones? Ted leans hard toward the former, and the argument is economic, not sentimental. He cited Noam Brown at OpenAI, who plots test-time-compute curves with cost on the x-axis: a bigger model that reaches an equally good answer in far fewer tokens can come out cheaper than a small one grinding away. “It could be more cost-effective to just ask the bigger model.” His blunter version: “people underrate the value of having a really smart model to ask questions to.” The counter-pressure is real - at Ramp, staff are nudged away from using Opus 4.8 to write emails when Sonnet will do the job - but his bet is that raw intelligence keeps winning on cost as well as quality.

The next axis is taste

The frontier the conversation kept returning to was judgment. The same essay claimed models now pick the better next research step about 64% of the time, up from 51% in November. How do you build a system with taste? Here Ted turned characteristically vague - he wouldn’t say what goes into it - but he would say that when his team started using Mythos this year, it felt like it had better research taste than anything before it. With one caveat: still “not as good as your average researcher.” Average researcher where, I asked. “Maybe at Anthropic.” A high bar to be measured against, and one the models are now climbing.

Safety is a capability, not a tax on it

Ask what safety research has ever done for product quality and Ted reaches for cars. Seat belts, crumple zones, airbags, an oven that won’t catch fire - guardrails are what “make the product usable” at all, and a product nobody can safely use generates no feedback to improve it. The clearest case built the whole category: reinforcement learning from human feedback began as a safety project - stopping a chatbot from spewing garbage - and turned out to be the thing that made chatbots good. “Alignment and safety really go hand in hand with capabilities,” he said. Indeed, reinforcement learning from human feedback draws its roots back to a 2017 collaboration between OpenAI and DeepMind:

One encouraging trend I put to him is that bigger models are turning out easier to align, not harder. Ted’s own framing was more guarded - it has “gone different from how many safety researchers expected,” he said - but his read is that “people feel pretty good about the alignment situation right now,” tempered by the obvious caveat that “we don’t know if we will cross some capability threshold and it’ll suddenly flip.” Hence the case for “healthy apprehension.”

Where the leverage is

I closed by asking what the highest-leverage work in 2026 looks like. Ted was self-aware about his own bias and unhedged anyway: if you believe AGI is close, go to a frontier lab, because your lever there is bigger than it is outside one. He counts himself among the converted - “I’m definitely more AGI-pilled than when I joined.” OpenAI and Anthropic are, he pointed out, still small companies where a newcomer can move things, and the window matters. If you want to influence the direction these systems take, “sooner is better than later.”

That includes London. Anthropic’s office here has gone from 15 or 20 people a few years ago to a couple hundred, with whole strands of frontier work - Ted’s among them - run from the UK rather than mirrored from California. “It doesn’t feel like we’re a satellite,” he said.

What the conversation kept circling back to is that the hard part of scaling was never the spending. Anyone can buy more compute; the discipline Ted’s team is building is the other half - knowing what each experiment will and won’t tell you, and turning that compute into capability you can measure and trust.

That story, which Vivek Natarajan told from the RAAIS stage, is the kind of result his team at Google DeepMind has been chasing. Natarajan is a Research Lead there, working at the intersection of AI, science and medicine. When he last spoke at RAAIS a couple of years ago, the state of the art was Med-PaLM, a language model tuned to answer medical exam questions. His pitch this year was more ambitious: that the recipe behind AlphaGo can be turned on science and the clinic, and that the trick is teaching models to stop thinking fast and start thinking slowly.

System one is not enough

The problem with using a chatbot as a scientist, Natarajan argued, is that even reasoning models mostly do “system one style thinking,” quick responses drawn from surface-level pattern matching. Real discovery is the opposite: slow, deliberate, rigorous, the product of chewing on a problem for weeks until the spark comes. He wanted “system two style thinking,” and to get it he reached back into DeepMind’s own history. AlphaGo’s 2016 breakthrough came from self-play and search, with agents playing each other, taking feedback from the environment, and reinforcing what won. AlphaZero then showed the same recipe could scale from zero knowledge to superhuman play in months, limited mainly by compute.

The AI co-scientist generalizes that idea. Instead of agents playing a game, they generate scientific hypotheses, then critique, debate and refine them over hours and days, what Natarajan calls a “generate, debate and evolve ideas loop.” Borrowing from AlphaStar, DeepMind’s StarCraft system, the team added tournaments: a ranking agent stages pairwise debates between hypotheses, scores them against a rubric derived from the scientist’s stated goal, and assigns Elo ratings, so only the strongest ideas reach the human. Because the debates run in natural language, they can be summarized and fed back into the agents’ context, which is what makes the system self-improving. It also lets the system signal its own uncertainty, what he called “epistemic humility,” which matters when the scarce resource you are spending is a scientist’s time.

The whole project nearly didn’t happen. The idea came from Gary Peltz, a Stanford geneticist who, after one of Natarajan’s lectures, suggested that a model trained on scientific text might generate hypotheses for the causes of rare disease. Most of the team thought it was too early. They did it anyway. Sometimes, as Natarajan put it, you “jump off the cliff, and then you figure out how to build an airplane on the way down.”

From hypothesis to organoid

The Penadés result, run with collaborators at Imperial College on antimicrobial resistance, was the moment the team realized they were onto something. But the more telling cases are the ones that ended in a wet lab. Physician-scientists at Houston Methodist used the co-scientist to find drug-repurposing candidates and combination therapies for acute myeloid leukemia. Peltz’s own lab pointed it at liver fibrosis, a disease with few treatments, and tested its picks in human liver organoids. One candidate, vorinostat, not only showed anti-fibrotic activity but cut TGF-beta-induced chromatin damage by over 91%, a hint of regeneration. The interesting part is that vorinostat is an FDA-approved cancer drug, exactly the kind of cross-field connection a liver specialist might never make, and the system surfaced it because it could read broadly while the human judged what mattered. Natarajan called this complementary intelligence, and it is the honest version of the pitch: the machine goes wide, the scientist goes deep.

He kept the limits in view. The system itself is general-purpose, he stressed, with nothing in the scaffolding specific to biology; the specialization comes from the tools it reaches for at runtime. But asked where it fails, he was candid that the wins so far have been in biology, where a mass of unread literature hides real signal. Chemistry is harder, and fields like mathematics and physics, which reward narrow depth-first reasoning over wide reading, harder still. The common thread is not biology or medicine specifically, but a way of turning compute into disciplined deliberation, then putting the result back in front of expert humans.

Manufacturing medical experience

The other half of the talk was about access. World-class medicine, Natarajan said, is “pretty much a geographic and socioeconomic lottery,” and his team’s second mission is to close that gap. The vehicle is AMIE, a diagnostic dialogue system he co-leads with Alan Karthikesalingam, a vascular surgeon still practicing in the NHS. Asked once how to choose between two doctors, Karthikesalingam told him to “always go with the one who has more gray hair.” There is no substitute for experience, so the team manufactured it. Using the same self-play machinery, AMIE ran consultations against synthetic patients and a critic, refining itself over millions of simulated dialogues. A human doctor might see 10,000 to 50,000 patients in a career; AMIE has already run hundreds of millions of conversations, “building up the world’s most experienced doctor,” with the heavy caveat that it all happens in simulation.

It is starting to pay off. In evaluations published in Nature, AMIE matched or beat physicians in simulated consultations with patient actors, on diagnosis and, more pointedly, on rapport, empathy and relationship building. “I’m kind of sorry about the doctors and humans in your life,” Natarajan deadpanned to the unsurprised. But this is not about replacement, he insisted: “the story is still about augmentation.” A companion study found that general physicians given complex diagnostic puzzles did significantly better with the AI as a thinking partner than working alone or with standard tools like web search. And in an early, supervised feasibility study with Beth Israel Deaconess in Boston, where patients spoke to the AI before an urgent-care visit under physician oversight, zero safety stops were required under the study’s predefined criteria, patient trust in AI rose after the interaction, and the system’s pre-visit diagnoses held up against the attending physicians, all without the benefit of lab tests.

A third person in the room

For most of modern medicine, Natarajan closed, the core unit of care has been a dyad: the doctor and the patient. His bet is that it is becoming a triad, the doctor, the patient and the AI, with the machine as a teammate rather than a tool. That is the idea behind the team’s next effort, an AI co-clinician. It is a tidy frame, and the evidence on stage made it land harder than it would have a year ago. The deeper claim running under both halves of the talk is that the self-play recipe which once mastered a board game can now give scientists and clinicians a new kind of thinking partner: one that searches widely, argues with itself, and hands humans better starting points. The wet labs and the early clinical studies are starting to agree.

At this year’s RAAIS, Roberta Raileanu set out why that ceiling exists and what it would take to lift it. Raileanu leads the open-endedness team at Google DeepMind and was previously at Meta. Her talk laid out a recipe for superhuman scientific discovery: a general system that makes groundbreaking discoveries across domains faster than people can. Three ingredients hold it together, but underneath all three sits one problem. We are good at searching for anything we can measure. We do not yet know how to measure what makes a discovery good.

The plateau before the breakthrough

The past two years delivered a real proof of concept. In 2024, Sakana AI wired LLM agents into a loop that generates a hypothesis, implements it, runs an experiment, and iterates, producing its first machine-written papers. The bar has risen since: a fully AI-generated paper has passed peer review at a workshop attached to a top machine learning conference, and a wave of startups now aims to automate research outright.

Proof of concept is not parity, however. Put the best agents head to head with human experts on the same open problems and the agents improve early, then stall. They are good at variations and combinations of known methods, and weak at what defines real research: exploring unfamiliar paths and making the conceptual leaps that change a field. Scale up compute and time and the human line keeps rising while the model line flattens.

The reason to think the ceiling can move is breadth. These models train on a far wider cross-domain corpus than any scientist can absorb, and can search for connections across more fields than any specialist holds in working memory.

The lesson of Move 37

Raileanu’s first ingredient is to treat discovery as a reinforcement learning problem. An agent acts in an environment, observes what happens, and learns from the feedback, which is not far from how a scientist forms an idea, tests it, and revises. The appeal is specific: as long as you can measure progress with a reward, the agent is free to find any solution that earns it, including one no human would think to try.

The proof is a decade old. When DeepMind’s AlphaGo played Lee Sedol, its move 37 was so counterintuitive that no human would have played it, and it won the game. But Go is a closed world with a clean reward: a move wins or it does not. Move 37 shows what optimization can do once the objective is given. In science the objective is not given. Deciding what counts as progress on an open question is the actual work, and it is the part no reward function hands you.

To study this inside AI research itself, her team built MLGym, a sandbox where an LLM agent runs shell commands, edits files, and runs experiments across tasks from language modeling to game theory. Even a year ago, simple setups could self-improve against a benchmark, but only by tuning hyperparameters and swapping architectures, not by inventing a method a human expert would adopt. That gap, between optimization and originality, is the rest of the talk.

Why greatness cannot be planned

Most breakthroughs, Raileanu argued, are not solutions to known problems. They involve finding the right problem, and “innovation is rarely this linear process from A to B.” Try to build a personal computer in the 1800s and you would not get there by scaling up the abacus; you would need the vacuum tube, which was invented to amplify radio signals.

She took the frame from Kenneth Stanley and Joel Lehman’s “Why Greatness Cannot Be Planned,” and put its claim on the screen: “No prerequisite to any major invention was invented with that invention in mind.” Optimize too narrowly for an objective and you skip the stepping stones that lead to it. Machine learning has won by hill climbing toward benchmarks, and that has carried the field far. But a hill climber only ever reaches the top of the hill it started on.

The fix is to widen the search. Borrowing from evolutionary methods, you hold a population of candidate solutions, mutate them, and select which to keep. The usual fitness function rewards performance alone. Raileanu’s argument is to score for what scientists actually value too: novelty, diversity, interestingness. This is where the signal problem surfaces in the open, because none of those is easy to measure. Her stopgap is to let an LLM judge what a person would find interesting, which at least keeps ideas that are not useful yet but might combine into something later.

Her team’s Rainbow Teaming did this for AI safety, generating diverse jailbreak prompts across a grid of risk categories and attack styles, then reusing what worked in one cell to seed another. Train on the result and the model gets measurably harder to break. The same machinery, she suggested, should carry over to ideas and methods, where a solution built for one field can matter in a completely different one.

Optimizing discovery itself

The third ingredient is to stop optimizing discovery on a single task and optimize the process of discovery itself. DeepMind has trained RL agents across huge numbers of simulated environments and watched them adapt to new ones far faster than before, in some cases faster than humans. If that transfers to research, an agent could generate its own tasks and get better at discovering, not just at one discovery.

To make that studyable, her team built DiscoBench, a framework that procedurally generates AI research tasks: more than 400 million of them, across problems like language modeling and image classification, with an agent free to target a loss function, an optimizer, or an architecture. The early signal is encouraging, with more and more diverse training tasks improving performance on held-out problems the agent has never seen.

The complementary curve

Stack the three together and you have the recipe: reinforcement learning to discover better solutions where progress can be measured, divergent search to find new problems rather than climb known ones, and meta-learning to speed up the whole process on problems no one has posed yet.

The bet underneath it is complementarity. Humans go deep in one field; a model reaches across many at once. The line worth chasing is neither the human curve nor the machine curve, but the one above both, where the two discover what neither would alone. Yet all three ingredients run back into the same wall. We have good algorithms for search once we know what to reward, and we still cannot reward novelty, a promising dead end, or taste. “The key is, do you have the right signal?” Raileanu asked. The search is the easy part. The missing piece is the signal.

Angelos leads ElevenLabs’ speech-to-text and text-to-speech teams and built its Scribe and Scribe Real-time transcription systems. The Scribe V2 models he shipped this past year rank, he says, as the most accurate transcription models on most popular benchmarks. It gives him a particular vantage on the problem, since voice models live or die on latency and cost at scale. It is also the second year running ElevenLabs has taken the RAAIS stage; in 2025 its CEO, Mati Staniszewski, spoke on the voice frontier. This year, we dove into the engine room.

What a token actually costs

Every optimization starts with knowing what you are paying for. For the autoregressive transformers behind most popular LLMs, a token’s cost reduces to two bottlenecks: compute, how fast the GPU does the matrix multiplications, and memory bandwidth, how fast the GPU’s VRAM can load the model’s weights and its KV cache. Generation runs in two phases. A prefill step reads the whole prompt and fills the KV cache, the model’s working state, and is compute-heavy. A decode step then emits tokens one at a time, each conditioned on the last, and is memory-heavy. The KV cache is what lets the model reuse that prefill instead of recomputing it for every new token, and at scale it is the thing that hurts: a hundred concurrent requests need a hundred separate caches resident in memory. Size is not destiny either. Angelos noted that Qwen 3’s cache costs almost three times as much per token as Qwen 2.5’s despite near-identical parameter counts, so two models of the same size can cost wildly different amounts to run.

Stop letting the GPU sit idle

The first and biggest single win is batching. GPUs are excellent at parallel work and poor at sequential work, and the dominant cost in decoding is loading the model weights, which can be shared across every request in a batch rather than reloaded for each. Naive batching groups requests once and then waits for the slowest one to finish while the GPU idles. Continuous batching fixes that: it batches at the level of each decode or prefill step, so a new request can join a GPU already mid-flight on others.

Shrink the weights, then the cache

With the GPU busy, the constraint becomes memory, so the next moves all reduce it. Quantization comes first. Models are usually trained at BF16, sixteen bits per weight, which is more precision than they need; dropping the weights to FP8 roughly halves their footprint with near-lossless accuracy, given H100-class hardware and a little quantize-aware training, which injects noise into the gradients so the model learns to tolerate the lower precision. That buys headroom for more cache and lifts throughput to twenty users per GPU. The more aggressive options exist too: int4 is lossy but useful on-device, and MXFP4 reaches four bits but only on Blackwell and newer.

Speculative decoding comes next. A cheap draft model proposes tokens and the big model verifies them in a single forward pass, accepting the run until the two disagree. It only pays off when the models agree often, which they frequently do not, so in practice it is used less than its reputation suggests; applied here it nudges the running total from twenty to twenty-eight users per GPU. The more popular cousin is multi-token prediction, where the same model wears extra prediction heads and drafts several tokens itself, with no second model to host. It earns its keep with two or more heads, and it doubles as a training signal: teaching a model to anticipate several moves ahead, like a chess player, tends to make it more stable and sometimes faster to learn. Most big labs use it, and it lands in the same place, around twenty-eight users per GPU.

The largest gain is also the riskiest. The KV cache holds far less redundant capacity than the weights do, so compressing it is genuinely lossy. Angelos was candid about this from his own testing: the much-discussed Google method TurboQuant was announced as lossless but, in his experience, proved lossy in practice, because any change to the cache is hard for the model to recover from. The fix is again on the training side: distill the model so it grows accustomed to a lower-precision FP8 cache, and you keep most of the accuracy while shrinking the cache 2.5x. That single step lifts throughput from twenty-eight to seventy users per GPU - seventy times what the same hardware served at the start.

Where the frontier labs go

Seventy is what disciplined engineering gets you. Going further means changing the architecture itself, and here the labs are placing different bets. DeepSeek’s multi-head latent attention squeezes each token’s key-value pair into a small latent rather than storing it in full, which both speeds inference and stretches context toward a million tokens; it was one of the more copied ideas after DeepSeek-R1. Qwen swaps standard quadratic attention for a linear network on every other layer, cheaper and longer-context, at some cost to quality. NVIDIA goes furthest, replacing the transformer on a fraction of its layers with state-space models that scale linearly and compute faster, keeping enough transformer layers to hold accuracy up. With architecture-level changes like these, the ladder reaches roughly a hundred and forty users per GPU.

Nothing here is free

Angelos was careful not to oversell any of it. Every technique on the ladder carries a cost. Batching adds latency and runs into a memory ceiling. FP8 quantization takes a small quality hit without the extra training. Speculative decoding needs access to weights and a training pipeline to work well. KV cache compression is the one most likely to degrade output, so the real question is how much degradation you can absorb rather than whether you can avoid it. He was blunter still about the gap between papers and production: many compression methods that report no loss of accuracy were tuned on a handful of benchmarks, and scaled to millions of users they can simply fall apart. You often only find out which ones once they are popular enough to be stress-tested in the wild. Which technique pays depends entirely on the workload.

The reason any of this matters beyond the engineering reached the room through a question from the floor: today’s token prices are subsidized, by one audience estimate a factor of ten to forty. Angelos’s hope is that optimization, not subsidy, eventually closes that gap. The largest models, he said, have to be subsidized to make economic sense, but he expects smaller, Sonnet-class models to become good enough for nearly all everyday use, at margins that actually work. He already sees the shape of it in agent systems: route each request to the smallest model that can handle it, and reserve the expensive one for planning.

The difficulty is in the control plane around it: one gateway, one governance layer, measurable fallbacks, cost controls, layered human review, and a way to run all of it inside a regulated bank that serves more than 70 million customers across over 40 countries. Revolut ships more than 200 products and has handled over a trillion dollars in transactions, with a machine learning model now in the path of almost every one of them. The leverage, in Donets’s telling, has moved from the model to everything around it.

Four constituencies, one bottleneck

For years, Revolut’s AI was classical machine learning: fraud and transaction models shipped through three internal libraries, one each for training, serving, and performance monitoring. Then, in 2022, the ground moved. Vendors began exposing large models behind an API, and suddenly you did not have to train anything to build something. Generative use cases started growing exponentially while the classical models kept running underneath.

Donets spent as much time on the people problem this created as on the technical one. Four internal groups pull in different directions: researchers who want compute and freedom to explore; builders who want one common API and to ship today; operators who want predictability, rollbacks, and cost under control; and a compliance function that owns human-in-the-loop controls, security audit, and data sovereignty. Left to themselves, every product team solves the same problems its own way, and governance fragments into tribal knowledge spread across hundreds of teams. That is expensive, and it does not scale.

Govern the use case, build one gateway

Rather than govern each model one by one, Revolut made two moves that changed the shape of the problem. The first shifted the unit of governance to the AI use case, a move that lines up with the EU AI Act’s use-case-based view of risk, so that one set of risks, budgets, and rules can cover several models at once and match policy to context. The second put a single gateway at the center of the company, with the governance layer on top of it, rather than shipping capability as libraries each team installs for itself.

But there’s a tradeoff: whereas libraries push reliability onto whichever product team owns the service, a central gateway makes one team responsible for everyone. Even so, the cost of improving a library means cutting a release, then persuading hundreds of busy teams to upgrade and absorb breaking changes they never wanted. With one gateway, the central team ships the improvement once and every product inherits it at, in Donets’s phrase, “zero effort.” Compliance and monitoring move to the same place. As a result, Revolut runs roughly twice as many generative use cases as classical ML ones, all off that single platform.

What breaks when the model is someone else’s

Once you are renting frontier models rather than training your own, you inherit failure modes you do not control. Pay-as-you-go providers run at around 98.5% uptime, which sounds high until you count the hours of dead service it implies each month for a scaled, global business. So Revolut wires a fallback chain into every generative product: if the primary model degrades or stops responding, traffic rolls to the next, and the next. Slightly degraded service beats no service at all.

Subtler, and more painful, was a failure they could not see at all. Because the platform watched only inputs and outputs at the interface, a model buried in the fallback chain quietly stopped working and nobody noticed. “Everything was fine, uptime was high enough, but the model itself was not functional,” Donets said. Or, as one of his slides put it: without per-model visibility, a model doing nothing looks exactly like one that works.

Money was the other lesson, and an easier one to swallow. Teams reach for the newest and most expensive model by reflex, but most workloads are over-provisioned, and right-sizing the model to the task cuts cost by as much as eight times with no loss in quality. Donets’s rule: do not default to the newest model in production. Measure first, then use the smallest model that clears the bar.

A note on the org chart

Underneath the platform sits an org chart doing as much of the work as the code. Revolut is flat and built as a matrix: AI engineers are embedded in product teams, each staffed to ship end to end, with a functional line back to the platform. Standards and tooling flow down; field requirements flow up to Donets’s central group, which sets direction and pushes compliance rules out. He called the product teams “our forward-deployed engineers,” the mechanism by which one team’s hard-won experience becomes everyone’s. The architecture, as one slide noted, ends up shaped like the org chart.

From Rita to AIR

Where all of this lands is a single product that has been running for years. It began as Rita, a support chatbot built on intent models and pre-filled scenarios - the “slot machine” era - that frustrated as often as it helped. In 2022 the team tested large models, Bloom and BloomZ at 175 billion parameters, and found they worked. The first thing they put into production was mundane: paraphrasing a multi-screen FAQ into a short, relevant answer. LLM-based Rita reached production in Q2 2023, then rolled out country by country, Europe first and Japan the hardest, finishing around Q1 2025.

Voice came next, and it runs on a simple pipeline: audio is transcribed, a small LLM decides whether to answer directly or hand off to the full multilingual chatbot, and an end-to-end response comes back in under two seconds. It now runs in 20 countries, handles around 25,000 calls a month, and resolves a customer’s problem roughly eight times faster than a human agent. AIR, the latest layer, followed in Q2 2025 and pulls in transactional data: it can break down your spending, propose hotels inside a budget computed from your own history, or explain why a stock is moving. Across the arc from the old chatbot to today, the share of cases resolved without a human climbed from 17% to 80%, Net Promoter Score went from low to high, and the financial impact, Donets said, ran into double-digit millions of pounds. AIR began rolling out in the UK in April 2026, where Revolut says it has 13 million customers.

Where human oversight is mandatory

Holding all of it up is the monitoring layer. Revolut stores every input and output and runs a panel of LLM “judges” against live traffic - one dedicated to hallucination - currently 9 to 12 mandatory metrics and rising, backed by human review teams that sample chats and transcripts, and by the blunt signal of Twitter and Reddit when something goes badly wrong.

Above all of it sits a hard line: no decision that can change someone’s life is made by an AI system. That position got tested in the room. An audience member who works on regulated healthcare AI pushed back - in his field, he said, “humans make that process unsafe,” so an AI judge might be the safer choice. Donets gave ground on the evidence, agreeing that machines “provide more stable and better help to users,” but not on the principle: the critical calls still do not go to the model. Asked how soon that might change, he did not hedge: “This year, definitely no.”

The frontier gets the headlines, but shipping AI inside a regulated bank across 40 countries is won or lost on the plumbing beneath it: one gateway, the right unit of governance, a fallback for when the vendor fails, and a human who still gets the last word.

Hadrien is co-founder and CEO of Alta Ares, an AI-first air defense company building an integrated platform for detection, identification, tracking, and interception. Founded in 2024, it works on one of the most demanding problems in applied AI: defending against mass-produced attack drones, cruise missiles, and glide bombs in contested environments, where a system has to perform in seconds, at the edge, and under operational pressure.

Alta Ares began as a software company focused on intelligence, surveillance, and reconnaissance (ISR) video analysis. The feedback loop from Ukraine pushed it into a wider air defense architecture spanning data-fusion software, edge AI, and hardware effectors built to operate from Arctic to desert conditions. Its stack includes Pixel Lock for embedded detection, tracking, and terminal guidance; Gamma for autonomous interceptor guidance; X-Lock for short-range drone interception; and Black Bird for faster aerial threats.

Apply to join RAAIS 2026

Why the air defense gap is so large

Recent salvos over Eastern Europe and the Middle East have exposed a hard truth: legacy air defense systems built to stop fast jets are losing the economics against mass-produced aerial threats. NATO partners increasingly face coordinated waves of one-way attack UAVs paired with cruise missiles and glide bombs. This is an enormous problem that is defining capability gaps in modern defense.

Unlike many AI applications, air defense is not forgiving. The enemy object is small, fast, and often deliberately cheap. The operator may be tired, cold, and working at night. The environment may be jammed. Protecting people, critical infrastructure, and military assets now demands systems engineered from day one for autonomy, modularity, interoperability, and unit-cost discipline.

That is what makes counter-UAS such an important test case for applied AI. There are many hard parts to the problem: recognizing an object in a poor quality video feed, fusing sensor inputs, holding a track, guiding an interceptor, preserving human control over the final engagement decision, and doing all of it inside a system that can be carried, deployed, and iterated quickly. Pixel Lock is Alta Ares’ answer: onboard computer vision that detects, classifies, and tracks targets in real time and supports autonomous terminal guidance while keeping the operator in the loop. Here the AI sits inside the control chain itself, guiding the interceptor rather than only flagging a target for an operator to act on.

The Ukraine feedback loop

Alta Ares’ development is shaped by proximity to the battlefield. Interceptors running Pixel Lock began shooting down Shahed-type drones in November 2025. Hadrien’s public interviews describe an engineering culture built around fast field feedback: simulation helps, but the front line reveals failure modes a lab cannot.

That loop matters because drone warfare is changing faster than long procurement cycles and static product roadmaps can absorb. Threats adapt, operators adapt, and countermeasures adapt in turn. The companies that make a difference in this category are the ones that can move from deployment to model improvement to hardware iteration without treating each step as a separate world.

From software to systems

Alta Ares’ public milestones track a company moving from software into an integrated air defense architecture. In March 2025, NATO Allied Command Transformation named Team Alta Ares the winner of its 15th Innovation Challenge for an “Embedded AI for Recognition, Detection, and Identification” submission focused on glide bombs - a system that detects, identifies, and predicts the trajectory of these low-cost guided munitions from visual and acoustic data.

Later in 2025, Alta Ares demonstrated its drone-interception system to NATO at the DGA missile test site in Biscarrosse. The company calls the configuration a Tactical Protection Dome: radars, interceptor drones, data fusion, and Pixel Lock software.

The most recent milestone came in Estonia. Early in 2026, working with the Estonian Defense Forces and Ukrainian partners, Alta Ares tested Black Bird, its turbojet-powered interceptor, in Arctic conditions. The company reported three consecutive flights, ground temperatures of -17 degrees Celsius and -25 degrees Celsius at altitude, and a top recorded speed of 450 km/h. The trial also validated the less cinematic but more important parts of the system: communication links, antenna performance, live video transmission, and Pixel Lock target detection, tracking, and locking. In parallel, the company has begun mass-producing interceptor drones in France.

Why it matters for RAAIS

The next generation of air defense is being built as layered systems: sensors, command and control, autonomy, and low-cost effectors combined quickly enough to keep pace with changing threats. NATO’s own 2026 work on layered counter-UAS points the same way, treating the challenge as one of integrating sensors, effectors, electronic warfare, command systems, and battlefield lessons into something coherent. Alta Ares is one version of that thesis, built from the edge inward: European, field-informed, and aimed at a class of threats that has already changed the character of modern conflict.

For RAAIS, the interest goes beyond defense. Alta Ares is a working case study in applied AI inside a live operational system, where robustness, cost, latency, and human judgment all bind at once. The same problem shows up across robotics, autonomy, and other high-consequence settings, where a model that performs on a benchmark still has to keep working once it meets conditions that shift under it.

Hadrien’s background

Hadrien’s path into defense technology is unusual. Before Alta Ares, his public background spanned law, Ukraine, and operational fieldwork rather than a conventional defense prime career. He studied at the University of Paris 1 Panthéon-Sorbonne, qualified with the Paris Bar, served as an OSCE international observer around Mariupol in 2019, and worked on humanitarian projects in Eastern Ukraine.

That background shows in the company he has built. Alta Ares designs from the operational problem backward: what the operator sees, what they miss under stress, how fast the threat changes, and what kind of AI stack survives that reality.

Apply to join RAAIS 2026

The harder half of voice AI?

ElevenLabs built its name on synthetic voices that made generated speech sound natural, expressive, and controllable. But the reverse problem also exists: turning messy, real-world speech back into accurate text. For voice agents it is often the part that decides whether the product works in the ears of the human user.

A live agent cannot reason about what it has not heard. It needs a transcript that is fast enough to preserve conversational flow, accurate enough to carry names, numbers, technical terms, and intent, and robust enough to handle accents, background noise, interruptions, and people switching languages mid-sentence. Speech-to-text is a key perception layer for interactive AI systems.

Angelos’ work at ElevenLabs focuses on model quality, inference design, latency budgets, and production reliability.

Two Scribes for two production regimes

Angelos has worked across both of ElevenLabs’ latest transcription models: Scribe v2 and Scribe v2 Realtime.

Scribe v2, launched in January 2026, is optimised for high-accuracy transcription of long and complex recordings: batch transcription, subtitling, captioning, media libraries, training material, compliance workflows, and research audio. These are settings where the model can use broader context, but where errors compound quickly. A missed drug name, a malformed account number, or a confused speaker label can make the downstream transcript much less useful. ElevenLabs built Scribe v2 with production transcription features such as keyterm prompting, entity detection across 56 categories, smart multi-language transcription, speaker diarisation, word-level timestamps, and audio tagging.

On Artificial Analysis’s AA-WER v2.0 benchmark, which combines a held-out voice-agent dataset with cleaned public datasets for parliamentary speech and earnings calls, Scribe v2 led the overall ranking with a 2.3% word error rate. It also led two of the three component datasets, including AA-AgentTalk and Earnings22-Cleaned-AA. That is a useful reminder that “accuracy” is not one thing: the model has to work across short agent-directed speech, formal speech, and long business audio, not just a clean public benchmark.

Scribe v2 Realtime, released in November 2025, solves the same problem under a much tighter constraint. It is built for live agents, meeting assistants, captioning, and conversational interfaces where a transcript that arrives too late is almost as bad as a wrong one. ElevenLabs describes it as delivering live transcription at around 150 milliseconds of latency across more than 90 languages, with features such as automatic language detection, voice activity detection, manual commit control, text conditioning, and predictive transcription for the next words and punctuation. On FLEURS, a multilingual benchmark spanning 30 languages, ElevenLabs reports the lowest word error rate of any low-latency ASR model.

Why latency changes the shape of the problem

For most of the last decade, speech-to-text progress was mainly discussed through benchmark word error rate. That number still matters, but it no longer captures the whole product problem. A transcription model that is accurate after the fact can be excellent for subtitles and useless for a live agent. A real-time model that is fast but unstable can make the agent interrupt, hallucinate intent, or miss the moment to respond.

This is why Scribe v2 and Scribe v2 Realtime are better understood as two parts of the same system-level push rather than a single leaderboard entry. The batch model pushes for the cleanest possible transcript when full context is available. The real-time model asks how much of that accuracy can survive when the system has to stream partial understanding under a human conversational latency budget. In one case the challenge is depth of context. In the other it is speed without collapse.

For RAAIS, that makes Angelos’s work a particularly good example of applied AI becoming harder as it becomes useful. Offline model quality is only the beginning. The real question is whether a research result can be made fast, stable, observable, and cheap enough to sit inside millions of interactions where people do not care about the benchmark. They care whether the agent heard them.

Apply to RAAIS 2026

Angelos’s background

Angelos’s path into speech-to-text runs through systems work, which is part of what makes it interesting. He studied Software Engineering at the University of Glasgow, graduating with First Class Honours in 2020. His master’s project developed a reinforcement-learning-based scheduler for IoT networks, and before ElevenLabs he worked across cloud-native infrastructure and reliability roles at Skyscanner, Ondat, and Beacon Platform. He also contributed to Gentoo’s Portage package manager through Google Summer of Code.

The audio thread appears early. In 2017, his team won the Amazon challenge at the Glasgow University hackathon with Emotionify, an app that combined facial recognition, text-to-speech, and the Spotify API to match music to a user’s mood. He later won the Goldman Sachs and IBM challenges at subsequent Glasgow hackathons, with projects involving speech recognition, text-to-speech, and custom machine-learning models.

He also keeps teaching the fundamentals. At AI Engineer Europe 2026, Angelos ran a workshop called Training an LLM from Scratch, Locally, walking engineers through the practical components of building a small language model on local hardware. That instinct - to understand the whole stack from first principles, then make it work in production - is exactly the one needed for speech-to-text now. Voice AI will not be judged by whether it can speak beautifully in a demo. It will be judged by whether it can listen accurately enough to be trusted.

Short bio

Angelos Perivolaropoulos leads research engineering for speech-to-text at ElevenLabs, where he works across Scribe v2 and Scribe v2 Realtime, the company’s high-accuracy batch transcription and low-latency streaming transcription models. His work sits at the intersection of model development, inference, and production reliability. He studied Software Engineering at the University of Glasgow, graduated with First Class Honours, and previously worked across cloud-native infrastructure and reliability roles at Skyscanner, Ondat, and Beacon Platform.

At RAAIS we have a focus on translating cutting-edge technology and research into production-grade products for real-world problems.

The platform behind production AI at Revolut

Nikolay runs Machine Learning Engineering at Revolut, where his organisation builds the platform that supports every production AI system inside the company - from classical ML for fraud and personalisation, to time-series foundation models, to the voice agents now serving customer support. Revolut has crossed $1.3 trillion in transaction volumes and is the number one finance app in 19 countries; machine learning now sits in the path of millions of financial decisions a day.

The most concrete recent example of that platform in production is the rollout of voice agents across Revolut’s customer service operation, built with ElevenLabs. The system handles live calls in more than 30 languages, resolves tickets in under five minutes - roughly 8x faster than the previous escalation path - with a 99.7% call-handling success rate across more than four million customers in the UK and Europe.

One platform for builders, operators, researchers, and compliance

A central theme in Nikolay’s public work is that the hard problem in production AI is not building a model in isolation. It is building one platform that has to serve builders, operators, researchers, and compliance at the same time - and do so inside a regulated financial product. That framing is especially relevant now, because most organisations have already discovered that strong model performance does not by itself solve deployment. The harder challenge is the infrastructure around the model: evaluation, release discipline, governance, monitoring, and cost control, all without slowing iteration to a crawl.

For a technical audience, this is where a large share of the field’s practical difficulty now sits. As production AI moves into regulated settings - finance, healthcare, public services - the systems around the model have to satisfy operational and supervisory requirements as well as engineering ones. The platform is not separate from the model work. It is what determines whether model progress becomes durable capability inside a real organisation.

Governance as a velocity enabler, not a blocker

Nikolay has publicly outlined a framework for launching GenAI products in 90 days under regulatory constraints, built on three pillars: data lineage (treating compliance data as feature material rather than overhead), continuous delivery with multi-layered validation that goes beyond pass/fail tests, and compliance guardrails plus the documentation needed to defend them. The underlying claim is that governance, designed well, is a velocity enabler - moved into the development environment with clear tiers, predictable review cycles, and regulation treated as a technical requirement with a defined path to production.

As more companies try to support classical ML and generative AI side by side inside regulated products, this is becoming the central question in deployed AI. The bottleneck has shifted out of the model and into the systems that surround it.

Nikolay’s background

Nikolay holds a PhD in engineering from Siberian Transport University, where his thesis applied wavelet transform analysis to damage detection in beam superstructures from the response of traversing vehicles — structural health monitoring for bridges, an early grounding in reliability, monitoring, and operational discipline for critical infrastructure that carries through to his current work. His career has spanned Moscow, St Petersburg, Seoul, Stockholm, Toronto, and now London. He maintains active open-source projects and writes publicly on MLOps, AI governance, and risk in fintech at donets.org.

Apply to RAAIS 2026

Short bio

Nikolay Donets is Head of Machine Learning Engineering at Revolut, where he leads the team that builds the AI platform behind the company’s production models - covering classical ML, time-series foundation models, and the voice agents now serving customers in 30+ languages. He has publicly outlined a 90-day framework for shipping GenAI products under regulatory constraints, built on data lineage, continuous delivery, and compliance guardrails. He holds a PhD in engineering, with earlier work in structural health monitoring and predictive maintenance for critical infrastructure.

I’m excited to bring you the next Air Street NYC AI meetup on 14 May 2026, which brings together New York’s best researchers, founders, and engineers working in AI. Featuring Ramp, Arena Physica and Air Street Capital.

NYC AI brings together New York’s best researchers, founders, engineers, and operators who are building and deploying AI systems. We keep the group deliberately small, curated, and focused on people who are building - not talking about - AI. The goal is to help you learn new best practices, exchange ideas with peers, and meet future collaborators, co-founders, and team members.

At this edition of NYC AI, we’ll cover the following topics:

• Deploying AI inside a high-growth fintech - Seb Goddijn, Product Lead, Internal AI at Ramp

• Physics-aware AI - Pratap Ranade, CEO & Co-Founder of Arena Physica

• State of AI Report 2026 - Nathan Benaich, Air Street Capital

We’ll follow the talks with happy hour drinks, food, and plenty of time to meet people.

Recent meetups have included people from OpenAI, Anthropic, Google DeepMind, Meta, Hugging Face, Runway, Scale AI, Cohere, top labs at Columbia, NYU, Cornell Tech, Princeton, and startups including Ramp, Lumaril, Cursor, Decagon, Sierra, Harvey, Mercor, Granola, and many others.

If you work in research, engineering, product, BD, or you’re a founder, request a spot here:

Request a spot here

From the Gatsby Unit to the science of scaling

Ted leads work on the science of scaling at Anthropic, where his focus sits at the intersection of reinforcement learning, optimization, and large-scale deep learning. Before Anthropic, he completed his PhD at the Gatsby Computational Neuroscience Unit in London, advised by Maneesh Sahani and Matt Botvinick. His thesis examined multitask reinforcement learning in brains and machines - questions of transfer, generalization, and how learning carries across tasks rather than being solved from scratch each time.

That background matters because these are not only reinforcement learning questions. They are scaling questions. As models grow, what matters is not simply whether they get better, but how capabilities generalize, which trade-offs emerge, and what kinds of optimization behavior actually hold up across settings. Ted’s research has consistently sat close to those underlying mechanics.

Apply to RAAIS 2026

When optimization quietly breaks

Much of Ted’s work investigates what happens when training is pushed too hard. Modern AI labs train models against multiple objectives at once - helpfulness, safety, factuality - and combine them into a single score the model is asked to maximize. Confronting Reward Model Overoptimization with Constrained RLHF (ICLR 2024 Spotlight, top 5% of submissions) shows that this routinely fails in a specific way: as training continues, the model keeps climbing the score even as humans start to rate its actual outputs worse. The paper offers a fix that treats each objective as a constraint to satisfy rather than a number to maximize, which keeps the model’s behavior aligned with human judgment as training scales up.

That theme - keeping behavior reliable as you push optimization further - runs through his earlier work as well. ReLOAD (ICML 2023) addressed a long-standing problem in reinforcement learning where the policy you end up with can drift away from the average policy you trained, leaving you with worse behavior than your numbers suggest. Towards an Understanding of Default Policies in Multitask Policy Optimization (AISTATS 2022, Best Paper Award Honorable Mention) examined how a model’s fallback behavior shapes whether it can carry skills across tasks rather than relearning each one from scratch. A First-Occupancy Representation for Reinforcement Learning (ICLR 2022) and Tactical Optimism and Pessimism for Deep Reinforcement Learning (NeurIPS 2021) studied how the way an agent represents its environment, and the assumptions it makes about its own uncertainty, determine whether what it learns generalizes - or quietly breaks the moment conditions change.

Why this matters at the frontier

For anyone building on advanced AI systems, the most important questions are no longer purely about capability. They are about reliability - what optimization actually converges on, where reward signals decouple from human judgment, how capabilities generalize to settings the model wasn’t trained on. Ted’s research is one of the more rigorous bodies of work engaging with those mechanics directly. As scaling continues to be the central engine of progress, understanding what it is doing - not just that it works - becomes harder to separate from product reality.

Ted’s background

Before the Gatsby Unit, Ted earned his bachelor’s at Princeton, with honors work across neuroscience, computer science, and linguistics, and his master’s in computer science at Columbia. He worked on biologically-plausible deep learning at Columbia’s Zuckerman Institute and on neural encoding at Princeton. He also interned at DeepMind, where he worked on constrained reinforcement learning, and at Uber AI Labs, where he worked on optimization for large-scale deep learning. His path from theoretical neuroscience to optimization theory to frontier model development gives his perspective on scaling a particularly interesting flavor.

Apply to RAAIS 2026

Short bio

Ted Moskovitz leads the Science of Scaling team at Anthropic, where his research spans reinforcement learning, constrained optimization, and large-scale deep learning. Before Anthropic, he completed his PhD at the Gatsby Computational Neuroscience Unit in London, advised by Maneesh Sahani and Matt Botvinick, with internships at DeepMind and Uber AI Labs. His selected publications include Confronting Reward Model Overoptimization with Constrained RLHF (ICLR 2024 Spotlight) and Towards an Understanding of Default Policies in Multitask Policy Optimization (AISTATS 2022, Best Paper Honorable Mention).

Apply to join RAAIS 2026

Frontier AI and the future of intelligence

Raia Hadsell is VP of Research at Google DeepMind, where she co-leads the Frontier AI unit and has contributed to Gemini 2.5, Gemma 2, RecurrentGemma, and RoboCat. Her earlier seminal work includes Overcoming Catastrophic Forgetting in Neural Networks, Dimensionality Reduction by Learning an Invariant Mapping, and Learning to Navigate in Complex Environments. Raia is also founder and Editor-in-Chief of Transactions on Machine Learning Research, and in November 2025 was appointed an AI Ambassador to the UK government’s DSIT. Read more about Raia.

Open-ended agents that keep learning

Roberta Raileanu is a Senior Staff Research Scientist at Google DeepMind, leading the Open-Endedness team and building a new Open-Ended Discovery group. Before DeepMind, she led Meta’s Tool Use team for Llama 3 - work that now sits behind Meta AI, Data Analyst, AI Studio, and the Ads Business Agent. Her research targets the gap between models that look capable in short bursts and agents that keep acquiring skills, with contributions including Toolformer and the MLGym benchmark for AI research agents. She is also an Honorary Associate Professor at UCL. Read more about Roberta.

Apply to join RAAIS 2026

AI for medicine and science

Vivek Natarajan is a Research Lead at Google DeepMind working at the intersection of AI, medicine, and science. He led Med-PaLM and Med-PaLM, the first AI systems to reach passing and expert-level scores on US Medical Licensing Exam questions, and AMIE, a multimodal diagnostic agent that was non-inferior to 21 primary care physicians in a randomized, blinded virtual OSCE study across 100 multi-visit case scenarios. More recently, he co-led the AI co-scientist, which has already surfaced a candidate for repurposing against acute myeloid leukemia and proposed new therapeutic targets for liver fibrosis. Read more about Vivek.

World models and the future of simulation

Jeff Hawke is co-founder and CTO of Odyssey, a frontier AI lab building general-purpose world models. In 2025, Odyssey unveiled the first AI model to stream interactive 3D worlds in real time, a step toward generative environments people can step into rather than watch. Before Odyssey, Jeff was a founding engineer at Wayve, where he pioneered end-to-end neural networks for driving on complex urban roads. Read more about Jeff.

Compute moves into orbit

Philip Johnston is co-founder and CEO of Starcloud, building the first data centers in space. In November 2025, Starcloud-1 launched with an NVIDIA H100 on board, the first H100 ever operated in orbit, and 100x more powerful than any GPU previously deployed in space. Starcloud-2 will follow this year with multiple GPUs including an NVIDIA Blackwell, and Starcloud-3 is being designed as a 200kW spacecraft to launch from SpaceX’s Starship. In March 2026, Starcloud closed a $170M Series A at a $1.1bn valuation. Read more about Philip.

Apply to join RAAIS 2026

A community of peers

Throughout the day, attendees will meet 200 researchers, engineers, founders, designers, and policymakers from across the field, with more speakers and programme details to follow. RAAIS 2026 is supported by Lambda and Cooley.

Apply to join RAAIS 2026

Vivek Natarajan is a Research Lead at Google DeepMind leading research at the intersection of AI, science, and medicine. He spoke at RAAIS in 2023 on the potential of large language models in medicine, and the progress since then has been remarkable. His work centers on a question that is rapidly becoming one of the most important in applied AI: what does it take to build systems that are useful in expert domains like healthcare and scientific discovery? In medicine especially, performance means reasoning under uncertainty, handling complex interactions, and meeting a far higher bar for trust and reliability.

Apply to RAAIS 2026

From medical benchmarks to clinical capability

Vivek is the lead researcher behind Med-PaLM (Nature, 2023) and Med-PaLM 2 (Nature Medicine, 2025), the first AI systems to achieve passing and expert-level scores respectively on US Medical Licensing Examination questions. Med-PaLM 2 scored up to 86.5% on the MedQA dataset, an improvement of over 19 percentage points on its predecessor, and produced answers that physicians rated as comparable or preferred to those from human doctors.

Medicine is one of the clearest examples of a domain where surface-level language ability is not enough. A model has to retrieve specialist knowledge, reason carefully, and communicate in a way that reflects the stakes of the setting. Med-PaLM helped shift the conversation from whether language models could be adapted to medicine at all, to how they should be evaluated, where they might be useful, and what standards they need to meet.

Project AMIE and the move toward real clinical interaction

Vivek co-leads Project AMIE (Articulate Medical Intelligence Explorer), a research program aiming to build and democratize medical superintelligence. AMIE is not a question-answering system: it is a conversational diagnostic agent that gathers symptoms, asks follow-up questions, reasons across specialties, and now interprets visual medical information through its multimodal capabilities.

In March 2026, the team published results from a prospective clinical feasibility study at Beth Israel Deaconess Medical Center, one of the first real-world tests of conversational diagnostic AI inside a primary care workflow. One hundred patients interacted with AMIE via text chat before their appointments. The system’s differential diagnosis included the final diagnosis in 90% of cases, with zero safety stops required. A nationwide randomized study in partnership with Included Health is now underway.

Real healthcare is not a single-turn task. It is a sequence of interactions shaped by ambiguity, incomplete information, and changing hypotheses. A clinically useful system needs to engage with the process of care, not just generate a plausible answer. That makes AMIE especially relevant to the RAAIS audience: it reflects the broader shift from models that perform well on static benchmarks to systems that can operate across richer, more realistic workflows.

Apply to RAAIS 2026

AI for science as well as medicine

Vivek recently co-led the development of the AI co-scientist, a multi-agent system built on Gemini that acts as a virtual scientific collaborator: systematically generating, critiquing, and refining novel hypotheses. Early results have included identifying a drug candidate for repurposing against acute myeloid leukemia and discovering new therapeutic targets for liver fibrosis.

The system has moved quickly from research to deployment. In 2025, the AI co-scientist became a key component of the US Genesis Mission, providing scientists across all 17 Department of Energy National Laboratories with accelerated access to Google DeepMind’s AI for Science models. A parallel partnership with the UK government is giving British researchers priority access to the AI co-scientist alongside tools like AlphaEvolve and AlphaGenome, and Google DeepMind will open its first automated research laboratory in the UK in 2026, focused on materials science.

The goal is no longer only to build systems that answer expert questions, but systems that support expert practice itself: in medicine through clinical reasoning, in science through the generation and testing of new ideas. That is one of the most important frontiers in AI right now: moving from systems that organize existing knowledge to systems that help produce new knowledge.

Vivek’s background

Prior to Google, Vivek worked at Facebook AI Research, where he led the winning entry to the 2018 VQA Challenge at CVPR and co-authored MMF, a widely used multimodal framework. He studied at the University of Texas at Austin and is part of the faculty for executive education at the Harvard T.H. Chan School of Public Health.

That background helps explain the arc of his work. It sits at exactly the point where frontier model capability meets high-consequence real-world use, a place where applied AI becomes harder, more interesting, and much more important.

Roberta Raileanu is a Senior Staff Research Scientist at Google DeepMind, where she leads work on the Open-Endedness team, and an Adjunct Professor at UCL, advising PhD students connected to UCL-DARK. Her research focuses on how frontier models are increasingly asked to do long-horizon work plan, use tools, recover from mistakes, and keep improving through interaction. This exposes a gap between systems that look capable in short bursts and systems that keep acquiring skills in messy environments. Roberta’s research is about closing that gap.

From exploration to open-ended learning

Roberta’s early work was shaped by a classic reinforcement learning problem that keeps resurfacing in new guises: exploration. If an environment gives sparse or delayed reward, brute-force search fails, and the right intrinsic objective can determine whether an agent learns at all.

Two papers anchor this period. RIDE: Rewarding Impact-Driven Exploration for Procedurally-Generated Environments (ICLR 2020) proposes an intrinsic signal that rewards actions changing an agent’s learned state representation, evaluated in procedurally generated settings where revisiting the same state is unlikely. Learning with AMIGo: Adversarially Motivated Intrinsic Goals (ICLR 2021) tackles sparse reward by pairing a goal-generating “teacher” with a goal-conditioned “student,” producing an automatic curriculum of increasingly challenging goals. In parallel, Decoupling Value and Policy for Generalization in Reinforcement Learning (ICML 2021, oral) argues that shared representations for policy and value can contribute to overfitting, and proposes a decoupled approach that improves generalisation on benchmarks like Procgen.

This portfolio matters because open-endedness is not a slogan. It is a technical demand: systems should continue to learn without requiring a human to constantly rewrite the task distribution.

The tool-use gap

Before joining DeepMind, Roberta was a Research Scientist at Meta, where she started and led the Tool Use team for Llama 3. This work aimed at enabling models to use tools like search and code execution, and to generalise to new tools at test time. The products that shipped from this work - Meta AI, Data Analyst, AI Studio, Ads Business Agent - are now used by hundreds of millions of people.

She was also a co-author on Toolformer: Language Models Can Teach Themselves to Use Tools (2023), one of the papers that helped establish tool use as a core capability for language models rather than an afterthought. Toolformer showed that a model can learn when and how to call external APIs - calculators, search engines, translators - with minimal supervision, by generating its own training data from a handful of demonstrations.

Tool use is not a feature checkbox. It changes what we can reasonably ask models to do, because it introduces feedback loops, memory, and failure recovery. It also introduces new failure modes: an agent that can call a tool can also call it badly, repeatedly, and confidently. Roberta’s treatment of agent behaviour as a sequential decision problem with real constraints - not a prompt-engineering exercise - is exactly the lineage you want when the field moves from “can it answer” to “can it execute.”

Why open-endedness is becoming a practical requirement

At DeepMind, Roberta now leads the Open-Endedness team and is building a new Open-Ended Discovery group focused on autonomously discovering novel artefacts - new knowledge, capabilities, or algorithms - in a self-improving loop.

Open-endedness is sometimes framed as a path to general intelligence. In practice, it is also a path to systems that do not collapse outside curated benchmarks. Most real deployments present a shifting distribution: new tools, new data, new user behaviour, and new adversarial pressures. A model that cannot keep learning becomes a periodic retraining job with brittle edges.

At Meta, Roberta also led an “AI Scientist” effort focused on agents that can iterate through parts of the research loop - implementing methods, running experiments, analysing results, and repeating the cycle. That work has now crystallised into MLGym: A New Framework and Benchmark for Advancing AI Research Agents (2025), which positions evaluation around concrete machine learning research tasks and frames the problem in a way that invites iteration by the broader community rather than one-off demos. If “AI scientist” systems are going to matter, we need ways to compare approaches, reproduce results, and identify what actually moves the needle. A benchmark is not the whole answer, but it forces precision about what the agent is allowed to do, what counts as success, and what is being optimised.

Roberta’s background

Roberta received her PhD in Computer Science from NYU in 2021, advised by Rob Fergus. Before that, she studied Astrophysical Sciences at Princeton, where she worked on theoretical cosmology and supernovae simulations - and before that, competed in the International Physics Olympiad and the International Olympiad on Astronomy and Astrophysics. That path from physics instincts to sequential decision-making research shows up in her taste for problems where scale alone is not enough.

She also co-developed and co-teaches a course on open-endedness and general intelligence at UCL, which signals something about where the field is heading: this is becoming a discipline with ideas worth teaching, not a loose collection of intuitions.

Apply to RAAIS 2026

We are delighted to announce Raia Hadsell as a returning speaker - she first spoke at RAAIS in 2017, when she was a Senior Research Scientist at DeepMind.

Raia is now VP of Research at Google DeepMind, where she co-leads the Frontier AI unit. She joined DeepMind in 2014, when it was still a 50-person startup freshly acquired by Google, and her work has since spanned some of the field’s hardest open problems: continual and transfer learning, deep reinforcement learning for robotics and navigation, and the models that power today’s frontier systems.

The arc of a career

What makes Raia’s research career unusual is the consistency of its through-line. She earned her PhD under Yann LeCun at NYU, where her dissertation on long-range vision for off-road robots received the Outstanding Dissertation award. That work helped shape metric learning and Siamese neural networks - architectures now so standard they underpin most modern contrastive learning. Her most highly cited papers include Dimensionality Reduction by Learning an Invariant Mapping and Learning a Similarity Metric Discriminatively, with Application to Face Verification, foundational contributions to representation learning that have collectively gathered tens of thousands of citations.

After a postdoc at CMU’s Robotics Institute with Drew Bagnell and Martial Hebert, and a stint at SRI International’s Vision and Robotics group, she joined DeepMind and turned her attention to a problem that had been nagging the field for decades: catastrophic forgetting.

Apply to RAAIS 2026

Why continual learning matters

Neural networks are powerful learners but terrible rememberers. Train a model on task B and it forgets task A. This is catastrophic forgetting, and it has been one of the deepest obstacles to building AI systems that improve over time rather than being retrained from scratch. Raia’s 2017 paper Overcoming Catastrophic Forgetting in Neural Networks proposed elastic weight consolidation, a method for protecting important learned parameters while still acquiring new knowledge. Alongside Progressive Neural Networks and Distral: Robust Multitask Reinforcement Learning, this body of work laid much of the groundwork for how the field thinks about lifelong and multitask learning today.

It’s also the thread that connects her navigation research - including a landmark Nature paper demonstrating that artificial agents trained to navigate develop grid-like neural representations resembling those found in rodent brains - to her more recent work on generalist robotic agents like RoboCat and bipedal robot locomotion published in Science Robotics.

From research to frontier systems

Raia’s selected publications tell a story about where frontier AI is actually heading. Her recent work includes contributions to Gemini 2.5, Gemma 2, and RecurrentGemma, alongside RoboCat - a self-improving foundation agent for robotic manipulation that can pick up new tasks from as few as 100 demonstrations - and research on teaching bipedal robots to play agile soccer using deep reinforcement learning.

This range is what makes her unusually well-placed to speak at RAAIS. She sits at the intersection of frontier language models, embodied intelligence, and the kind of continual adaptation that will determine whether AI systems can operate reliably outside the data centre - in factories, hospitals, homes, and the physical world.

Beyond the lab

Raia’s influence extends well beyond her own research. She founded and serves as Editor-in-Chief of Transactions on Machine Learning Research (TMLR), launched in 2021 as an alternative venue for rigorous ML publication. She sits on the executive boards of CoRL (Conference on Robot Learning) and WiML (Women in Machine Learning), is a Fellow of ELLIS, and is a founding organiser of NAISys (Neuroscience for AI Systems).

In November 2025, she was appointed as an AI Ambassador for the UK government’s Department for Science, Innovation and Technology, where she chairs peer review panels for national AI research initiatives - a role that puts her at the centre of UK AI policy at a pivotal moment.

She holds a PhD from NYU, and - in a detail that says something about the breadth of her thinking - an undergraduate degree from Reed College in religion and philosophy.

Apply to RAAIS 2026

At RAAIS we have a focus on translating cutting edge technology and research into production-grade products for real-world problems.

Philip is co-founder and CEO of Starcloud, a company building data centers in space to address one of AI’s most pressing constraints: energy. As model size and compute demand continue to grow, terrestrial data centers are running into hard limits in grid capacity, cooling, land use, and permitting timelines. Starcloud’s thesis is that space offers a different path: solar power, radiative cooling, and a route to scaling compute beyond what is practical on Earth.

From proof of concept to orbital compute

In November 2025, Starcloud launched Starcloud-1, a 60 kilogram satellite carrying the first NVIDIA H100 GPU ever operated in space, delivering roughly 100x more powerful GPU compute than had previously been deployed in orbit. Within weeks, the company achieved two notable firsts: training a GPT-style language model in orbit using NanoGPT, and running Google’s Gemma model in space on a high-powered GPU.

Processing AI workloads in orbit, close to the satellites generating the data, can cut latency from hours to minutes. Synthetic aperture radar satellites, for example, can produce huge volumes of data that are costly and slow to downlink. Analyzing that data in orbit could materially change both speed and cost.

Apply to RAAIS 2026

Why space could matter for AI infrastructure

Starcloud argues that orbital data centers could deliver major reductions in both energy cost and emissions over their lifetime, even after accounting for launch. The appeal is straightforward: space-based systems are not constrained by terrestrial grids, and space offers a naturally favorable environment for radiative cooling without the land and water footprint of conventional data centers.

The company’s long-term ambition is a fully solar-powered orbital data center with 5 gigawatts of capacity, large enough to rival major power plants on Earth without requiring land, transmission infrastructure, or connection to a terrestrial grid.

What comes next

Starcloud-2, currently planned for October 2026, is set to be the company’s first commercial mission. It will carry several NVIDIA H100 GPUs alongside NVIDIA Blackwell hardware, with persistent storage and continuous customer access to orbital compute. The mission will also include a cloud platform from Crusoe, making it possible for customers to deploy and operate AI workloads directly from orbit.

Starcloud is backed by Y Combinator, NVIDIA through its Inception program, and investors including NFX and In-Q-Tel.

Philip’s background

Philip is a second-time founder. He previously co-founded Opontia and earlier worked at McKinsey & Company on satellite projects for national space agencies, giving him firsthand exposure to both the potential and the constraints of space infrastructure.

He holds an MPA in National Security and Technology from Harvard University, an MBA from Wharton, and an MA in Applied Mathematics and Theoretical Physics from Columbia University. He is also a CFA charterholder.

Apply to RAAIS 2026

At RAAIS we have a focus on translating cutting edge technology and research into production-grade products for real-world problems.

Jeff is co-founder and CTO of Odyssey, a frontier AI lab developing general-purpose world models.

Jeff is working on one of the most ambitious problems in AI: building systems that can understand, predict, and simulate the real world. Odyssey’s work sits at the intersection of generative modeling, embodied intelligence, and large-scale learning, with the potential to unlock major advances across robotics, autonomy, and interactive digital environments.

Before co-founding Odyssey, Jeff spent 15 years building AI for autonomous driving. As VP Technology at Wayve, he helped pioneer visual policy learning and contributed to a new wave of end-to-end learning approaches for autonomy, pushing beyond hand-engineered systems toward models that learn directly from real-world experience.

His work has consistently focused on taking frontier machine learning research and applying it to complex, real-world problems where robustness, generalisation, and deployment matter most.

Jeff holds degrees in engineering and computer science from the University of Auckland and Georgia Tech, and completed his doctorate at the University of Oxford.

Apply to RAAIS 2026

Air Street AI meetups bring together ~150 researchers, founders, engineers, and operators who are actually building and deploying AI systems. These are deliberately small, curated evenings designed for people doing the work, not talking about the work. The goal is simple: share hard-won lessons, surface emerging technical frontiers, and help exceptional builders find one another.

Over the next few weeks, we’re hosting Air Street AI meetups in Munich, Zurich, and Paris featuring speakers from Black Forest Labs, Odyssey, Google DeepMind, Sereact, and Polar Mist. I’ll be presenting updates from our State of AI Report, the most widely read and trusted analysis of key developments in AI.

Air Street Munich AI, 17 Feb 2026

Against the backdrop of the world’s most important defense and national security gathering, the Munich Security Conference, this evening focuses on AI systems that reason about, perceive, and act in the physical world.

Fabian Gura, Member of Technical Staff, Odyssey
Odyssey is building frontier world models that simulate how visual and physical environments evolve over time and under intervention. With Odyssey-2 Pro, the team describes a “GPT-2 moment for world models,” where interactive, real-time simulations become viable and begin scaling rapidly in capability. Odyssey’s work pushes world models beyond static video toward systems that can be explored and acted upon.

Gustaf von Grothusen, CEO, Polar Mist
Polar Mist is building autonomous maritime defense systems for operation in GPS-denied and contested environments. Its Semper platform combines an unmanned surface vessel that can carry a variety of payloads with Polar Mist’s vision-based navigation and positioning system, enabling persistent autonomy without navigational drift.

Air Street Zurich AI, 19 Feb 2026

This evening centers on how AI systems learn to see, model, and manipulate the physical world, from pixels to policies.

Robin Rombach, CEO, Black Forest Labs

Black Forest Labs is a frontier AI research lab developing state-of-the-art visual intelligence, including its latest FLUX models for controllable image generation and editing. Founded by the original authors of latent diffusion, the company recently raised $300M at a multi-billion-dollar valuation, cementing its position as core infrastructure for next-generation visual AI.

Marc Tuscher, CTO, Sereact

Sereact is a frontier robotics research and deployment company, starting with picking and handling in warehouses and industrial environments. The company’s models enable robots to perceive, reason, and act in highly variable, real-world settings without brittle rule-based pipelines. Sereact’s work targets one of robotics’ hardest problems: robust generalization from vision to action in production. The systems are deployed across hundreds of robots for large enterprises in Europe and the US.

Air Street Paris AI, 11 March 2026

An evening focused on frontier visual intelligence and open source models.

Cyril Diagne - Black Forest Labs

Cyril works at the intersection of visual intelligence, generative models, and creative tooling. His work explores how modern vision models can move beyond recognition toward interpretation, controllability, and human collaboration. At Black Forest Labs, he focuses on pushing visual foundation models into new expressive and interactive regimes.

Edouard Yvinec - Google DeepMind

Edouard is a research scientist at Google DeepMind and a core contributor to DeepMind’s Gemma family of open-weight large language models, including Gemma 3, which was designed to be among the most capable models that can run on a single GPU.

We look forward to meeting you on the road!

Subscribe to our event series