Adaption launches AutoScientist AI that trains models on their own
▼ Summary
– Adaption launched AutoScientist, a product that automates fine-tuning to help AI models quickly learn specific capabilities.
– AutoScientist co-optimizes both data and the model to learn the best way to acquire any capability, potentially enabling frontier AI training outside major labs.
– The system builds on Adaption’s Adaptive Data platform, turning continuously improving datasets into continuously improving AI models.
– Adaption claims AutoScientist has more than doubled win-rates across different models, though these results are not measured by conventional benchmarks.
– The tool is being offered free for the first 30 days after release to encourage user adoption.
For years, the AI industry has been building toward a long-anticipated milestone: systems that can refine and improve themselves more effectively than human engineers. With substantial venture capital now flowing into next-generation research labs, the resources to chase that vision have never been greater. This week, one of those emerging labs delivered a significant breakthrough.
On Wednesday, Adaption unveiled AutoScientist, a new product designed to help models acquire specialized capabilities rapidly through an automated fine-tuning process. While the underlying techniques can be applied across many domains, Adaption is especially focused on how AutoScientist can streamline and accelerate the training of frontier-level AI models.
Sara Hooker, co-founder and CEO of Adaption and a former VP of AI research at Cohere, described the innovation as a fundamentally different approach to model training. “What’s super exciting about it is that it co-optimizes both the data and the model, and learns the best way to basically learn any capability,” Hooker told TechCrunch. “It suggests we can finally allow for successful frontier AI trainings outside of these labs.”
AutoScientist builds directly on Adaption’s existing product, Adaptive Data, which simplifies the process of creating and maintaining high-quality datasets over time. While Adaptive Data focuses on data quality, AutoScientist is engineered to turn those continuously improving datasets into continuously improving 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 explained.
Of course, any new approach to AI training must prove itself through results. In its launch materials, Adaption claims that AutoScientist has more than doubled win-rates across different models , impressive metrics, though difficult to evaluate without standard benchmarks. Because the system is designed to adapt models to highly specific tasks, conventional evaluation frameworks like SWE-Bench or ARC-AGI don’t apply.
Nevertheless, Adaption is betting that users will recognize the value once they see it in action. The lab is so confident in AutoScientist’s capabilities that it is offering the tool free for the first 30 days after launch.
“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 said.
(Source: TechCrunch)
