Investing In Flapping Airplanes
Building the “young person’s AGI lab” to unlock data efficient models, which we believe is the bottleneck to laddering up the next rung of AI intelligence
Sometimes you meet a founder who is so good that you want to back them no matter what they are working on. Just as rarely, you meet a company whose vision is so singular, it feels both inevitable and differentiated at the same time. When I first sat down with Ben and Asher Spector to discuss Flapping Airplanes, I felt both of these to be true at once.
Ben is a founder of Prod, which has quickly become one of the most powerful talent incubators in Silicon Valley. Asher, one year Ben’s senior, is a former debate champion turned Stanford statistics Ph.D. The brothers have an eye for raw talent, and aged 25 and 26 respectively, they are building the “young person’s AGI lab.” Their co-founder Aidan Smith is a former founder and Thiel fellow who worked at Neuralink for three years while in college at Georgia Tech. Ben and Asher believe that raw talent — the type of people who would previously have gone on to do Ph.D.s or join quant firms — are the key to unlocking the next phase of AI development. This stands in sharp contrast to the approach of focusing on “brand name” AI talent at any price, which seems so in vogue right now.
There is something else that is countercultural about this lab too: they recognize that scale alone is not enough. Indeed, the lab is predicated on this notion. There is only one Internet, and we’ve run out of it. Data is the bottleneck today to further AI scaling, not compute. Today’s AI models are wildly inefficient, something that Dwarkesh Patel has recently hammered home in his interviews with Andrej Karpathy, Richard Sutton and others. Flapping Airplanes is dedicated to searching for a more data-efficient AI. The company’s name is a nod to a more biological notion of machine intelligence.
Research vs. Scaling
One of the fundamental debates right now in AI is around scaling vs. research. The scaling paradigm argues for dedicating a huge amount of society’s resources, as much as the economy can muster, toward scaling up today’s LLMs, in the hopes that this will lead to AGI. The research paradigm argues that we are 2-3 research breakthroughs away from an “AGI” intelligence, and as a result, we should dedicate resources to long-running research, especially projects that may take 5-10 years to come to fruition.
While the divide between these two schools of thought is far from absolute — research breakthroughs make compute investments more efficient, and GPUs are required for net new research — they each lead to different tradeoffs. A compute-first approach would prioritize cluster scale above all else, and would heavily favor short-term wins (on the order of 1-2 years) over long-term bets (on the order of 5-10 years). A research-first approach would spread bets temporally, and should be willing to make lots of bets that have a low absolute probability of working, but that collectively expand the search space for what is possible.
One concern I have had is that by starving Ph.D. programs of talent, and by reallocating a lot of this talent to scale-oriented work (as opposed to fundamental research), we could actually push out the AGI timeline. As I noted in a previous blog post, “corporate politics tends to favor in vogue, consensus ideas over more radical, unpopular ones — the very kinds of ideas that can take years to prove worthwhile, and which scientific breakthroughs often depend on.” Flapping Airplanes is explicitly intended to provide a Ph.D.-like experience — true research independence, long-time horizons — while bridging the pay gap between academia and Big Tech.
One of the biggest learnings of the last decade in AI is the dramatic economic value of fundamental scientific progress. Since academia seems ill-suited to compete with Big Tech on this dimension right now, especially in AI, Flapping Airplanes is taking this into their own hands. And they are focused on precisely the problem that is likely to define the next wave of AI: building models inspired by the data efficiency of biology. With that as a north star, the company will make bets in a variety of directions to get there.
Profit Motive vs. AGI Motive
There are two coherent motives right now in AI. One is the profit motive, and this should inspire all application builders, given the tremendous opportunity ahead. As I have noted on many occasions, AI is so underpenetrated relative to the capability set available, it is likely to create a golden age for builders for many years to come. There is a large profit pool available, and we are in the very earliest stages of generating it.
The second coherent motive in AI is AGI, predicated on the belief that now that we’ve unlocked a machine intelligence, we should push toward upleveling that intelligence. Economically, progress in this direction is likely to be extremely valuable. The question is how to take on this objective in an efficient way, with a reasonable probability that resources invested will translate into long-term positive results. Flapping Airplanes is a pure play in that direction.
The New Guard
One of the most attractive aspects for me in partnering with Flapping Airplanes was the team. I have met nearly every candidate that has interviewed at Flapping Airplanes, and certainly every person the company has hired. This is one of the best teams I’ve ever seen assembled at the beginning of such an ambitious journey.
There are two core beliefs that make Flapping Airplanes different for top talent:
A belief that young people have always unlocked fundamental science in the past (Einstein was 26 during his “Miracle Year”)
A conviction that AI research is not a mystical domain for experts, but a domain learnable by anyone that is smart and ambitious
With just these two premises, Flapping Airplanes is at odds with a prestige-obsessed culture in the Valley right now. It is exactly this obsession with substance and with raw individual talent that made Ben so successful at Prod, and it is this “flame spotting” ability that gives me confidence in the courageous effort that is now Flapping Airplanes.
If you are interested to join, please reach out to recruiting@flappingairplanes.com.