The era of manual prompt engineering and armies of low-paid data labelers has met a formidable challenger: startup Recursive has emerged from stealth with a massive $650 million capital injection. Backed by industry titans including GV, Greycroft, Nvidia, and AMD Ventures, the round values the company at $4.65 billion. The intrigue lies in the lack of public performance benchmarks; for now, the valuation rests entirely on ambition and a steadfast investor belief that "recursive self-improvement" is the shortest path to Artificial Superintelligence (ASI).
Led by Richard Socher (formerly of Salesforce) and Tim Rocktäschel (ex-Google DeepMind), the team is deliberately distancing itself from the traditional LLM arms race. Their goal is to build open algorithms capable of generating innovation without human intervention. This appears to be a direct assault on the operational bottlenecks currently slowing tech scaling: human-in-the-loop dependencies are becoming prohibitively expensive and sluggish.
Rocktäschel’s strategy draws on Stanisław Lem’s concept of the "information barrier." As the co-founder explains, humanity has reached a point where the volume of knowledge grows faster than the biological brain can process it. Recursive intends to leapfrog this biological limit by fully automating the scientific method. The process begins with applied tasks—optimizing code and neural network architectures using AI itself—but the ultimate goal is more grand: turning intelligence production into an autonomous industrial cycle capable of breakthroughs in fundamental science.
For the labor market, the signal is clear and unsettling. While the industry remains focused on fine-tuning models via human feedback, Recursive is building a system designed to make that labor-intensive process a relic of the past. If this team of OpenAI, Meta, and Uber AI veterans succeeds, demand for traditional data labeling and prompt engineering will collapse. However, the absence of hard data invites skepticism: can an algorithm generate scientific breakthroughs in a vacuum, or is it destined to degrade without fresh human insight? For now, the project looks like a high-stakes bet that a massive bank balance can substitute for empirical experience.