Two Startups Want to Replace How AI Learns: One Just Raised $180M, Another Is Seeking up to $1B

Why It Matters

Achieving orders‑of‑magnitude data efficiency could dramatically lower AI development costs and accelerate deployment across robotics, manufacturing, and scientific research, reshaping competitive dynamics in the AI industry.

Key Takeaways

• Flapping Airplanes raised $180M to develop data‑efficient AI.
• Startup aims for human‑level data efficiency, 100k‑1M× better.
• Core Automation seeks up to $1B for low‑data learning models.
• Goal: AI agents needing 100× less data, continuous learning.
• Research may replace gradient descent with novel loss functions.

Pulse Analysis

The past decade has been defined by ever larger language models that consume petabytes of text, images, and code to achieve incremental performance gains. While this brute‑force approach has produced impressive benchmarks, it also inflates compute costs, carbon footprints, and barriers to entry for smaller players. Researchers increasingly recognize that human cognition extracts rich concepts from remarkably sparse experiences, suggesting a more efficient learning paradigm. As the industry confronts diminishing returns on size alone, investors are turning to startups that promise a fundamental shift away from data‑intensive training.

Flapping Airplanes, backed by GV, Sequoia and Index with a $180 million Series B, is betting on that shift. The company’s mission is to engineer AI systems that match the human brain’s estimated 100,000‑ to 1,000,000‑fold data efficiency advantage. To achieve this, its research agenda explores novel loss functions and even alternatives to gradient descent, aiming to extract maximal information from minimal examples. If successful, such technology could unlock rapid prototyping in robotics, personalized retail recommendations, and scientific discovery without the need for massive pre‑training corpora.

Core Automation, founded by former OpenAI senior researcher Jerry Tworek, is pursuing an even larger capital raise—between $500 million and $1 billion—to build continuously learning agents that require 100× less data. Tworek envisions a future where AI not only designs industrial automation solutions but also constructs self‑replicating factories and, eventually, terraforms planetary environments. By prioritizing continual, experience‑driven learning, the startup aims to sidestep the static, one‑shot training cycles that dominate today’s models. Should these ambitions materialize, the competitive landscape could tilt toward firms that master low‑data, adaptive intelligence, reshaping supply chains and R&D pipelines worldwide.