(Re)Share #58 - Microwave the white flag
AI Scientists | Drone defense | OSS Robots | DeSci | 3D-printed tissue | Batteries
Greetings from the beautiful inferno that is Arizona! Steph, Moose and I have officially commenced our nomadic summer where we mooch off friends and family for lodging while we completely gut our own abode. Despite literally everyone I know clearly telling us how painful home renos can be, we still decided to do the entire thing at once. Totally smart financial move btw. Anyway, this has been a busy few weeks with a huge range of fascinating stuff to cover so let’s get to it.
Shameless Plug
I recently kicked off a research project focused on the dreary state of affairs across UK and European universities and their spin-out/NewCo norms via TTOs. If you’ve spun out a company or invested in one, you already know about the rent-seeker equity grabs, hyper-opaque IP terms, and glacial negotiation timelines. Despite how common these issues are, there’s shockingly little shared data—and sunlight is the best disinfectant. I’m running an ongoing survey for founders to help bring transparency to the process and benchmark norms with real data. If you or anyone you know fits the bill, I’d truly appreciate you taking a few minutes to participate. This short survey will feed a 100% open-sourced database of NewCo history, and all responses are completely anonymous.
Stuff Worth Sharing
A bird in the hand - I’ve started developing a working investment thesis on Autonomous Science, and in that field, FutureHouse is leading the charge. Last month, the non-profit unveiled Robin—a multi-agent AI system capable of autonomously navigating the entire scientific research cycle. This includes literature review, hypothesis generation, experimental design, data analysis, and iterative refinement. The catalyzing event was Robin’s identification of ripasudil, a drug traditionally used to treat glaucoma, as a novel therapeutic candidate for dry age-related macular degeneration (dAMD), a leading cause of blindness worldwide. Robin integrates three specialized AI agents to form a cohesive system that mirrors the collaborative nature of human scientific research. The entire process was completed in just 2.5 months and the identification of a ROCK inhibitor was a sort of bio-move 37. This only further cements the idea that AI will radically accelerate our capacity to discover and design novel treatments. Full paper here, and special kudos to my buddy Ludo Mitchener for his contributions.
Autonomous Prime - Speaking of AI-driven Science, my good friends at Compound penned another incredible thought piece on the topic. Author Mackenzie Morehead offers a comprehensive exploration of how artificial intelligence (AI) is poised to revolutionize scientific discovery, evaluating both the current state and the envisioned future through the full end-to-end discovery process—data review, reasoning, experimentation, and beyond. There’s a lot covered, but the core argument is clear: the time is now to actively invest in order to unlock the full potential of autonomous science. An unsurprising take from some of my favorite deep thinkers.
Fuck that noise - DeepMind had another big release with Gemini Diffusion, an experimental text generation model that applies diffusion techniques to language tasks. Many are already familiar with this approach from image generation, and Gemini Diffusion works the same way: instead of generating text sequentially, it starts with random noise and iteratively refines it into coherent output. This enables the model to generate entire blocks of text simultaneously, leading to faster generation and improved coherence. One of my techbio founders was ecstatic about the speed gains and the potential to enhance on-device models—so definitely watch this space.
Microwave the white flag - Defense-tech startup Epirus has developed the Leonidas family of high-power microwave (HPM) systems designed to neutralize drone swarms. Unlike traditional kinetic defenses (blow shit up), Leonidas uses electromagnetic pulses to disrupt multiple targets simultaneously, offering a more cost-effective and efficient approach to modern warfare. Imagine the most terrifying microwave you can think of, mount it on a military convoy, and you’ve got a pretty good idea of the system. I’m uniquely interested in electronic warfare—it’s inherently more defensive in nature and obviously less lethal. Ukraine has become a test bed for this new era of conflict, and given some of the absurd developments in DC, Europe is in a uniquely motivated position to adopt new solutions.
I/O of the storm - Google held their annual flagship event, I/O, a couple of weeks ago—and even for the big boys, this one was jam-packed. The scope and scale of AI offerings were pretty remarkable. The immense focus on AI comes as no surprise, but for the first time, it feels like Google has finally caught up from their early stumbles. The FOMO is real, and they’re bringing the full force of their innovation engine to bear. I have a buddy who runs Flash pre-training, and I now see just how busy he’s been. The link above is a 32-minute highlight reel, but if that feels like too much, I’d recommend skipping to Project Mariner (02:35), Veo3 (18:00), and SynthID (19:30). Veo3 was the clear winner of the day—my X timeline absolutely blew up with some pretty remarkable demos.
Low-cost labor - Hugging Face threw its hat in the robotics ring with the debut of two open-source humanoid robots, HopeJR and Reachy Mini, aiming to make robotics more accessible and community-driven. HopeJR is a full-size humanoid with 66 degrees of freedom, capable of walking and manipulating objects. At roughly $3,000, it’s one of the most affordable full-body humanoids available—and it’s entirely open-source. Reachy Mini is a desktop robot priced under $300, equipped for head movement, speech, and listening tasks. It offers a physical sandbox for testing AI applications—and I do have a birthday coming up… Both models build on the earlier launch of LeRobot, Hugging Face’s platform for open AI robotics, and clearly signal their ambition to become the nexus of decentralized physical innovation.
Behind the moderate ball - This past weekend I watched Mountainhead, an absurdist satire on the detachment of super-rich tech leaders and the damage they inflict on the world. Sadly, art imitates life. Just last week, a deadly militant attack in Kashmir triggered a military confrontation between India and Pakistan, which in turn catalyzed a parallel information war across social platforms. False reports of military victories, AI-generated footage, and recycled videos flooded X and Facebook, misleading the public and escalating tensions. The platforms clearly failed to moderate the spread—only a fraction of the misleading content was flagged. But to be fair, with millions of views surging in the heat of conflict, I don’t think any manual review system could realistically hold up. For those interested in the topic, check out this Wired article on deepfake financial scams.
Recede investing - A pretty fun read and a new topic for the newsletter, DeSci. HairDAO, a decentralized autonomous organization, aggregates efforts of online biohackers to develop and test experimental hair loss treatments. DAO members contribute to research and share personal health data in exchange for crypto tokens, fostering a collaborative environment for scientific exploration. Ethical review boards be damned! That may sound crazy but I know the VC that backed their Seed round and it’s one of the most respected firms I know. So if you feel like it’s getting a bit breezy back there, check it out.
Take a brain check - A new study published in Nature introduces IMPACT (Integrative Multimodal Pipeline for Advanced Connectivity and Time-series), a machine learning framework designed to detect clinically significant changes in brain activity. This is particularly valuable for diagnosing and monitoring the progression of neurological disorders like Parkinson’s disease. In tests, IMPACT achieved high diagnostic accuracy for PD, with AUC values ranging from 0.97 to 0.98—outperforming conventional methods. The analysis revealed transient connectivity disruptions that align with dopamine deterioration. This reproducible pipeline is also adaptable to other conditions where short-lived brain changes serve as key diagnostic biomarkers. I happen to know a lot about this particular Oxford-based research group—for reasons that will become clear in the coming months…
Invisible ink - Researchers at Caltech have developed a technique called Deep Tissue In Vivo Sound Printing (DISP), which enables 3D printing of tissues directly inside the body without the need for invasive surgery. DISP uses an injectable bioink that remains liquid at body temperature but solidifies into precise structures when activated by focused ultrasound waves. A specialized monitoring molecule within the bioink allows real-time tracking of the printing process to ensure accuracy, while any excess material is safely metabolized by the body. The team has already demonstrated the technique by printing tissues inside the stomachs of rabbits and the bladders of mice. They also incorporated conductive nanoparticles to create soft biosensors and drug depots capable of releasing medications. This ultrasound-based approach marks a significant advancement over light-based bioprinting, which is limited by shallow tissue penetration. Ultrasound, by contrast, can reach depths of up to eight inches without causing harm—potentially opening up a wide range of non-invasive medical interventions. (Paper)
Salting the Earth - Researchers at MIT have developed a sodium-air fuel cell that uses liquid sodium metal as fuel and ambient air as the oxidizer, separated by a solid ceramic electrolyte. In lab tests, the prototype achieved energy densities exceeding 1,500 watt-hours per kilogram—surpassing the 1,000 Wh/kg threshold considered necessary for practical electric flight. That’s more than three times the performance of current lithium-ion batteries, which typically offer around 300 Wh/kg. The team now plans to develop a brick-sized, 1,000 Wh fuel cell to power large drones within the next year. (Paper)
Portfolio Flex
Orbital CEO, Jonny Godwin, presented the future of advanced materials at GTC.