In a field dominated by ever-larger models and voracious data appetites, research lab Flapping Airplanes is betting that the future of artificial intelligence lies in doing more with far less. Armed with $180 million in seed funding, the young company is trying to rethink how machines learn, drawing inspiration from the human brain without trying to copy it.

Co-founders Ben and Asher Spector and Aidan Smith argue that today’s frontier systems are trained on “the sum totality of human knowledge,” yet still struggle to learn new skills quickly. Humans, by contrast, extract rich understanding from comparatively tiny amounts of experience. That gap in data efficiency is the core problem Flapping Airplanes wants to attack.

The team sees itself not as a rival to scale-obsessed giants like OpenAI or DeepMind, but as an explorer of a different part of the landscape. Large language models excel at memorization and breadth, they say, but remain sluggish at adaptation. The brain, with its slow, noisy hardware and radically different learning rules, is an “existence proof” that other algorithms are possible. Those algorithms might be unlike both the brain and transformers, but the brain shows the space of options is far larger than current orthodoxy.

The company’s name captures this philosophy. If today’s systems are Boeing 787s, Flapping Airplanes wants to build “flapping airplanes” rather than birds: machines that borrow ideas from biology while embracing the very different constraints of silicon. The goal is not to be better in every dimension, but to be different enough to unlock new trade-offs and new applications.

That ambition shapes both their research agenda and their culture. They are hunting for 1000x gains in data efficiency, not 20 percent improvements. They recruit unusually young researchers, including students, prioritizing raw creativity over long publication lists. The key signal, Ben says, is whether a candidate can teach him something genuinely new in conversation.

If they succeed, the implications could be profound. More data-efficient models could make robotics and scientific discovery commercially viable, enable rapid post-training on just a handful of examples, and push AI away from mere cost-cutting toward generating genuinely new science and technology. For Flapping Airplanes, the most exciting future is not one where AI simply automates existing work, but one where it helps humanity do things it was never smart enough to do alone.