For years, artificial intelligence researchers have anticipated the moment when AI systems would be able to improve themselves more effectively than humans could. With investors channeling substantial funding into a new wave of research-focused AI labs, there are now more resources available than ever to pursue this ambition. Now, one of these emerging labs has taken a significant step toward making that vision a reality. On Wednesday, Adaption launched a new product called AutoScientist, which enables models to rapidly acquire specific capabilities through an automated approach to traditional fine-tuning. While these techniques can be applied across various fields, the Adaption team is especially focused on their potential to accelerate and simplify the process of training and fine-tuning frontier-level AI models. According to co-founder and CEO Sara Hooker, who previously served as VP of AI research at Cohere, AutoScientist represents a novel way of approaching the AI training process. "It suggests we can finally allow for successful frontier AI trainings outside of these labs," she said.

AutoScientist builds on the company’s existing data platform, Adaptive Data, which is designed to simplify the creation of high-quality datasets over time. Meanwhile, AutoScientist aims to convert those continuously improving datasets into continuously improving AI models. "Our view at Adaption is that the whole stack should be completely adaptable, and should basically optimize on the fly to whatever task you have," Hooker explains. Of course, this approach will only be as effective as its results. In its launch materials, Adaption claims that AutoScientist has more than doubled win-rates across different models—impressive figures, though difficult to contextualize. Because the system is built to adapt models to specific tasks, conventional benchmarks like SWE-Bench or ARC-AGI are not applicable. Nevertheless, Adaption is confident that users will notice the difference once they try AutoScientist—so confident, in fact, that the lab is making the tool free to use for the first 30 days after its release. "The same way that code generation unlocked a lot of tasks, this is going to unlock a lot of innovation at the frontier of different fields," Hooker says.