AutoScientist, an AI tool that helps models train themselves, has a big focus

For years, AI researchers have been waiting for the moment when AI systems can outperform humans. With investors pouring money into a new generation of research-driven AI laboratories, there are many resources available to achieve that goal. Now, one of the neolabs has taken a big step to make it a reality.

Wednesday, Change introduced a new medicine called AutoScientist which enables models to learn a specific skill quickly using an automated system of optimization. These methods are used for different purposes, but the Adaptation team is focusing on what can be done to speed up and reduce the process of training and improving the AI ​​model of the frontier.

According to co-founder and CEO Sara Hooker, who previously worked as VP of AI research at Cohere, AutoScientist represents a new way to advance AI education. “What’s really interesting is that it organizes the data and the model, and learns the best way to learn each skill,” Hooker told TechCrunch. “It shows that we can really push the limits of AI outside the lab”

AutoScientist builds on the company’s offering, Adaptive Datawhich aims to make it easier to create high-quality datasets over time. AutoScientist, meanwhile, is designed to transform those who are constantly manipulating data to improve AI models. “Our philosophy at Adaption is that the whole stack should be flexible, and should be optimized on the fly for whatever application you have,” says Hooker.

Yes, that method will be as good as the result. For its base hardware, Adaption boasts that AutoScientist has double the number of championships in a variety of categories – impressive numbers, but hard to pin down. Since this machine is designed for specific applications, benchmarks like SWE-Bench or ARC-AGI are not used.

However, Adaption is confident that users will notice a difference when they test AutoScientist out – with full confidence that the lab will make the tool comfortable to use for the first 30 days after release.

“In the same way that coding opens up more jobs, this will open up more across the board,” says Hooker.