Poolside AI Chief Predicts Human-Level AI Within 3 Years

Human-level artificial intelligence across knowledge work domains could arrive within just 18 to 36 months, according to poolside AI's chief technology officer Eiso Kant, a dramatically shorter timeline than many industry predictions.

"I personally think now is eighteen to thirty-six months away, where human level intelligence across the vast majority of knowledge work is achieved," Kant declared during an extensive interview with Machine Learning Street Talk this week. End of Miles reports this represents one of the most aggressive timelines publicly stated by a frontier AI builder with the capability to develop foundation models.

Building from the ground up

Kant emphasized that achieving this milestone requires a fundamentally different approach than most current AI development. "You don't get to do that unless you build from the ground up. You don't fine-tune your way to AGI," Kant stated, challenging conventional wisdom that iterative improvements to existing models will lead to human-level capabilities.

Poolside AI, founded in April 2023 by Kant and former GitHub CTO Jason Warner, is among a small group of companies building foundation models from scratch. The company has raised significant funding, including a reported $500 million Series B round in October 2024.

"The reason we built this company is because we saw a future that I personally think now is eighteen to thirty-six months away, where human level intelligence across the vast majority of knowledge work is achieved." Eiso Kant, CTO of poolside AI

The missing scaling axis

Central to poolside's strategy is what Kant describes as "the missing axis of scaling" in AI development: reinforcement learning from code execution feedback. While much of the industry focuses on increasing model size and training data, Kant believes reinforcement learning represents a critical third dimension for advancing capabilities.

"There was a missing axis of scaling that wasn't being discussed, and it's frankly why we started this company," Kant explained. "It was the axis of scaling through the use of reinforcement learning."

The company has built infrastructure containing nearly a million containerized code repositories with their test suites, enabling models to learn through trial-and-error by executing code and receiving feedback on the results.

Industry's competing timelines

Kant's prediction stands in stark contrast to more conservative timelines from other AI leaders. Many prominent researchers have suggested human-level artificial general intelligence (AGI) might be a decade or more away, with significant technical and safety hurdles remaining.

The poolside CTO's confidence stems from the company's experimental approach to AI development. "Our team in January ran over four thousand experimental runs," he noted, describing their systematic exploration of architecture variations, data combinations, and reinforcement learning techniques.

These experiments help the company identify the optimal balance between different scaling dimensions, with reinforcement learning emerging as a particularly efficient path to advancing capabilities.

Beyond software development

While poolside currently focuses on AI for software development, the company's vision extends much further. Kant outlined a three-step progression: first assisting developers in building software, then enabling anyone to build software, and ultimately generalizing these capabilities across all fields and domains.

"The main objective is to achieve human level capabilities and go beyond," Kant stated, describing software as "a lever that we have on the world to be able to bring productivity" and "a lever to abundance."

If realized, Kant's timeline would mark a watershed moment for artificial intelligence, potentially transforming knowledge work across industries far sooner than many have prepared for.