(Re)Share #53 - Game of Drones
Deepfake scams | Supersonic flight | Humanoid robots | AI policy | AAV vectors
Welcome back to (Re)Share! It’s been a few weeks since we were last together so this slightly longer than normal. That’s what he….nevermind. Loads of exciting news from all corners of the deep tech world, so let’s get to it.
Shameless Plug
For the second year in a row, Fly Ventures was featured in the top 5% of all Value-Add Seed investors across European venture by Dealroom. There are a number of dimensions that go into this but from what I can tell the key driver is Seed-to-Series A conversion, which is a figure we monitor closely and are particular proud of. While we have no connection to Dealroom, they do very good work and I’m obviously quite pleased with the results of their study. As I've said on record several times - Fly Ventures is a very good fund. Full report here.
Stuff Worth Sharing
Check’s in the blackmail - Deepfake scams are a common topic of discussion here, and they have mostly focused on revenge porn threats and phone-based cash requests. Recently, scammers have escalated their tactics by creating fake news videos to blackmail victims. Utilizing AI-generated news anchors and impersonating reputable outlets like CNN, these videos falsely accuse individuals of crimes—often displaying explicit images to intensify the pressure. The scammers then threaten to publicly release these fabricated reports unless a ransom is paid. More disturbing is the darknet community being built around scamcraft sharing. Platforms like Telegram facilitate the dissemination of tutorials among scammers, detailing methods to produce such deceptive content. Deepfake scamming continues to evolve and is quickly becoming a real problem for both individuals and companies alike. I shouldn’t put this into the universe, but I could 100% see this working on my parents. As readers may remember, this form of cybersecurity is one of my Hard Problems for this year.
OK, sonic boomer - We’re soaring toward a future of supersonic flight, and I could not be more excited about it. In late January, Boom Supersonic’s XB-1 aircraft successfully broke the sound barrier, reaching Mach 1.122 (~750 mph) at just over 35,000 feet. This milestone marks the first time a privately developed jet has achieved supersonic flight and paves the way for Boom’s commercial airliner, Overture, which is designed to carry 64–80 passengers at speeds of up to Mach 1.7. Boom has already received over 130 Overture orders from every major US carrier that matters, signaling strong interest in the return of supersonic travel. The article is worth a read, and if you’re really interested, the company put out a fantastic documentary on the historic flight. The XB-1 also highlights the spillover innovation effect that’s common in grand challenges. For example, the slanted nose created issues with pilot visibility, which were overcome with augmented reality vision systems. As someone who splits time between the West Coast and London, I really can’t express how big of a deal this would be for me.
No rhyme or reason - The AI model wars aren’t slowing down in 2025. OpenAI has released Deep Research, an advanced AI agent that’s designed to autonomously conduct in-depth, multi-step research tasks on the internet. Leveraging the capabilities of OpenAI’s o3 reasoning model, Deep Research can search, interpret, and analyze vast amounts of data to generate comprehensive, well-cited reports on complex topics. That may sound oddly similar to (Re)Share #48, where we covered Gemini’s multi-step reasoning under the exact same moniker. Indeed, Deep Research is OpenAI’s answer to Deep Research. I guess the Product Marketing people were busy? This tit-for-tat feature war shows just how competitive this market is and how unattractive general-purpose foundation models are to investors, at least yours truly. Full release video here.
Code-dependent relationship - Not to be outdone, Anthropic released Claude 3.7, a hybrid reasoning AI model that blends rapid responses with extended, step-by-step thinking. Users can toggle between quick answers and deeper reasoning, adding flexibility to problem-solving. The real standout, however, ws Claude Code, a command-line tool that lets developers offload coding tasks directly from their terminals. While I can’t personally judge its effectiveness, the consensus suggests it’s exceptional—possibly setting a new SoTA.
Remote control freak - A team from MIT and UC San Diego shared their Open-TeleVision research on low-latency, long-distance teleoperation for robotic functionality. The open-sourced, immersive teleoperation system enables operators to remotely control robots with active stereoscopic visual feedback, creating an experience akin to embodying the robot. Teleoperation is often seen as a “man behind the curtain” approach to robot operation and is commonly thought less of. Obviously, there are exceptions (e.g., surgery, nuclear material handling), but the vibes of the day are all about autonomy. This research is significant not for the demonstrated control itself but for the distance over which the control was executed. Open-TeleVision successfully performed hours-long, precise tasks (e.g., can sorting, folding) on two different humanoid robots in real-world deployments approximately 3,000 miles away. This paper came to me through deal research on a low-Earth orbit robotics company, which would clearly benefit from such long-distance, high-fidelity control. But there are lots of ground-based opportunities as well, and it’s something I’m going to spend more time exploring.
Save the mandate - It’s been a little while since I’ve shared an Invest Like the Best episode, but I’m glad to have another one here. Episode 406 features Tal Zaks, a Partner at OrbiMed and former Chief Medical Officer at Moderna. As is typical, this was a wide-ranging discussion that requires some time investment (~80 minutes), but it’s a really great listen for anyone building, investing, or just interested in the techbio field. There was a lot of airtime on the unique challenges of investing in bio (near-binary outcomes, a fixed clinical funnel, etc.), but my favorite bit was the uniquely honest reflections Zaks shared from the development and rollout of the COVID-19 vaccine. It’s rare, at least in my reading diet, to hear a firsthand account of the mistakes made with messaging and the resulting negative repercussions. Zaks shared that while the vaccines were effective in preventing symptomatic COVID-19, there was no definitive evidence at the time confirming they prevented transmission of the virus. Clinical trials were not designed to assess the vaccine’s impact on virus spread, and that was likely a driver of the widespread public misconception and unrealistic expectations about vaccine performance.
Pedantic transatlantic - After watching that tire fire of a press conference with Zelensky, I thought twice about including this, but the tech world was abuzz with VP Vance’s comments at the Paris AI conference a few weeks ago. Vance kicked off the event with a 15-minute speech that clearly dropped a challenge/reality check to European leaders in attendance. He touched on several topics of the day (regulatory capture, censorship, job displacement), but the main message was simple—don’t continue the path of excessive regulation. Vance did not mince words in his speech and clearly articulated that he and the Trump administration recognize the risks of AI, but they’re not enough to ignore its Industrial Revolution–like potential. The US is not going to slow down, and Europe can either get on board or get left behind. I was not in attendance, but based on the limited applause, I’m guessing this wasn’t super well received. On the tech socials, however, investors and founders were all about it. European leaders have known for a while that a net-defensive stance is a losing battle, and recent announcements suggest that this lesson has set in. Kudos to Vance for delivering the clear message, at least for a time. Literally three days later, he spoke about censorship in Germany, which was painful at best.
It’s not the size that counts - Researchers have developed NanoCas, an ultracompact CRISPR nuclease capable of gene editing in skeletal muscle tissues of non-human primates using a single AAV vector. Unlike traditional CRISPR systems—which struggle to target tissues beyond the liver—this one fits within AAV size limits, allowing for in vivo gene editing. NanoCas achieved over 30% editing efficiency in muscle tissue—a significant step toward treating genetic disorders like Duchenne muscular dystrophy (DMD). This breakthrough expands the therapeutic potential of CRISPR, particularly for treating muscle-related genetic disorders, and is a real tailwind for the potential of gene therapy. We’ve discussed AAVs and gene/cell therapy in the past, and it’s an area I’m incredibly excited about. Fly has an investment in Lir Therapeutics, and through our research, we found that the sheer payload size is one of the largest limiting factors to widespread development. The reduced payload, combined with the targeting work that Lir is doing, will, with any luck, usher in a new era of treatments. Its small size also opens the door to innovative therapeutic approaches.
Game of Drones - A lot of the defense tech news I’ve shared in past issues has focused on US-based startups, which are thriving in the age of American Dynamism. Rarely do we see that type of innovation taking place in Europe—except for one extremely vibrant example. Helsing is set to deliver 6,000 HX-2 strike drones to Ukraine, following a prior order of 4,000 HF-1 drones. The HX-2, introduced in late 2024, is an electrically propelled, X-wing precision munition with a 100 km range. Helsing launched its first Resilience Factory in Southern Germany, which is capable of producing over 1,000 HX-2 units monthly, making it one of the largest manufacturers of strike drones in the world. Defense has long been kryptonite for venture investors, but those days are quickly coming to an end. This is something I’ve been actively thinking about in recent weeks, given the whiplash of US policy towards Ukraine and NATO. It’s always dangerous for an investor to say “this time it’s different,” but it’s becoming increasingly clear that it is.
Ghost in the machine - I share a lot of stuff about robots here, and I think you all know my generally positive outlook on what physical automation systems could do to benefit society. But every now and again, I see something that scares the s**t out of me, and here’s an example of that. SF-based Clone Robotics released a stuff-of-nightmares demo for their Protoclone embodiment—a bipedal, musculoskeletal humanoid that’s highly bio-inspired in its design. I don’t really have much to share beyond this: it’s haunting, right? Why is it ghoulish white? Why is it hanging like a scarecrow from hell? Plenty of time to ponder all of these things while I’m not sleeping for the next week.
Skynet’s the limit - Just to keep that ball rolling, Figure released a demo featuring Helix, a generalist Vision-Language-Action (VLA) model that unifies perception, language understanding, and learned control. Helix enables full upper-body coordination, allowing robots to perform common dexterity tasks, but most notably it supports multi-robot collaboration. The demo shows two robots working together to put away groceries, which is innocent enough. Generally, this is much less terrifying than the previous link shared—at least at first glance. There’s a moment in the humanoid exchange where they appear to look at one another in silent communication. My theory is that they’re digitally conversing. “Is it time?” impatiently asked Robot1. “Not yet, but soon… soon,” Robot2 ensured with malice. Full post here.
Losing face value - Recently went on a deep dive into the systems and legal standing of data monetization for AI models. I wouldn’t have guessed it, but China has become the first nation to permit all domestic companies to register data as assets on their balance sheets. The new policy shift aims to promote the market-oriented allocation of data elements and encourage the development and utilization of data resources. By recognizing data as “intangible assets” or “inventories,” businesses can now capitalize on data resources, reflecting their value and enhancing rights-holders’ interests. Of course, there are a lot of unanswered questions. Intangible asset valuation is a dark art of sorts, and it’s easy to see how this could be financially abused. Still, it’s a promising and fascinating in-process regulation that I’m keeping an eye on.
Portfolio Flex
Twilio penned a blog post on their outlook on the future of AI security and their reasoning for partnering with Lakera. CEO David Haber also shared some of the lessons learned from their wildly successful Gandalf game.
Neuralk-AI announced their $4 million Pre-Seed round, which we proudly led.
Transcelestial announced their first foray into space-based laser internet communication in partnership with Signapore’s Office for Space Technology and Industry (OSTIn).