Revolutionizing AI Training: Adaption’s AutoScientist Promises Smarter Self-Improvement

For years, researchers have dreamed of a day when artificial intelligence (AI) systems can improve themselves better than any human could. Now, Adaption, one of the leading research-driven AI labs, is taking a significant step toward making that dream a reality with their new product, AutoScientist.

AutoScientist helps models learn specific capabilities quickly using an automated approach to fine-tuning. According to Adaption's co-founder and CEO Sara Hooker, who previously worked at Cohere as the VP of AI research, this tool represents a new way to approach AI 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 explained.

Adaption's goal with AutoScientist is to make it easier for users to build high-quality datasets over time. These continuously improving datasets then turn into continuously improving AI models through the use of AutoScientist. '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 said.

While impressive results are claimed in Adaption's launch materials, such as more than doubling win rates across different models, it remains difficult to put these numbers into context. Since AutoScientist is built to adapt models to specific tasks, conventional benchmarks like SWE-Bench or ARC-AGI aren’t applicable.

Despite this, Adaption is confident that users will see the difference once they try AutoScientist — so confident that they are offering it free 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 said.