Adaption, a rising AI research company, is pushing the boundaries of self-improving AI systems. They recently launched AutoScientist, a new tool designed to help AI models learn and improve specific skills more quickly and efficiently. This innovation aims to reduce the need for extensive manual fine-tuning, potentially accelerating progress in frontier AI development.

What AutoScientist Does

AutoScientist builds on Adaption’s existing data platform, called Adaptive Data, which helps create better datasets over time. The new tool combines this data with an automated approach to fine-tuning, allowing models to learn new capabilities faster. Essentially, it co-optimizes both the data and the model, learning the best way to adapt to different tasks without human intervention.

According to Sara Hooker, co-founder and CEO of Adaption, this method allows for more flexible and efficient AI training. She explains that the system can suggest the best way to teach models new skills by continuously improving datasets and training strategies. This could make frontier-level AI models more accessible outside of traditional research labs.

The Potential Impact

AutoScientist is designed to speed up the process of developing advanced AI models. The company claims that it has already more than doubled success rates across various models, though these results are hard to compare directly with standard benchmarks. Because the system is tailored to specific tasks, typical tests like SWE-Bench or ARC-AGI don’t fully capture its capabilities.

Adaption is confident in AutoScientist’s potential, especially since they are offering a free trial for the first 30 days. They believe this tool could unlock new levels of innovation in AI, similar to how code generation transformed many tasks in the industry. By making AI training more automated and adaptable, it could significantly reduce costs and time in developing cutting-edge AI systems.

Overall, AutoScientist represents a step toward more autonomous AI systems that can improve themselves with minimal human input. If successful, it might lead to broader access to frontier AI capabilities, opening new opportunities across industries and research fields.

Inspired by

• https://techcrunch.com/2026/05/13/adaption-aims-big-with-autoscientist-an-ai-tool-that-helps-models-train-themselves/

Sources

• adaptionlabs.ai
• adaptionlabs.ai