For years, AI researchers have anticipated the moment when AI systems will be able to improve themselves better than humans. With investors pouring money into a new generation of research-driven AI labs, there are more resources available than ever to achieve the goal. Now, one of them has taken a big step towards making it a reality.

On Wednesday, introduced a new product called Adaption Auto Scientist which helps models quickly learn specific capabilities by using an automated approach to traditional fine-tuning. The technique has applications in a wide range of fields, but the adaptation team is particularly focused on its ability to speed up and simplify the process of training frontier-level AI models.

According to co-founder and CEO Sarah Hooker, who previously served as VP of AI Research at Coher, AutoScientist represents a new way to approach the AI ​​training process. “What’s really interesting about it is that it optimizes both the data and the model together, and basically learns the best way to learn any capability,” Hooker told TechCrunch. “This suggests we may eventually allow successful frontier AI training outside of these labs.”

AutoScientist builds on the company’s existing data offering, Adaptive datawhich aims to facilitate the construction of high-quality datasets over time. AutoScientist, meanwhile, is designed to continuously improve AI models on datasets. “Our view of adaptation is that the entire stack should be fully adaptive, and basically optimized on the fly for whatever task you have,” Hooker says.

Of course, this approach will only be as good as the results. In its launch materials, Adaption boasts that AutoScientist has more than doubled the win rate across models — an impressive number, but difficult to put into context. Because the system is designed to adapt models to specific tasks, traditional benchmarks such as SWE-Bench or ARC-AGI are not applicable.

Still, Adaption is confident users will notice the difference once they try AutoScientist — so confident that the lab is making the tool free to use for the first 30 days of release.

“Just as code generation unlocked a lot of work, it’s driving a lot of innovation across the boundaries of different disciplines,” Hooker says.