Adaption launches AutoScientist, an AI tool that lets models train themselves

For years, AI researchers have anticipated the moment when systems could improve themselves more efficiently than humans. On Wednesday, Adaption took a significant step toward that goal with the launch of AutoScientist, a product that automates the fine-tuning process, allowing models to learn specific capabilities quickly and with minimal human intervention.

What AutoScientist does

AutoScientist builds on Adaption’s existing data platform, Adaptive Data, which focuses on creating high-quality datasets over time. The new tool is designed to turn those continuously improving datasets into continuously improving AI models. According to co-founder and CEO Sara Hooker, who previously served as VP of AI research at Cohere, the system co-optimizes both data and model architecture simultaneously.

“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 Bitcoin World. “It suggests we can finally allow for successful frontier AI trainings outside of these labs.”

Automated fine-tuning in practice

Traditional fine-tuning requires researchers to manually curate datasets, adjust hyperparameters, and run multiple training cycles. AutoScientist automates these steps, using an iterative loop that tests different data compositions and training strategies to find the most effective approach for a given task. The system is designed to be model-agnostic, meaning it can work with various open-source and proprietary architectures.

Adaption claims that in internal tests, AutoScientist more than doubled win-rates across different models. However, because the system adapts models to specific tasks rather than general benchmarks, conventional metrics like SWE-Bench or ARC-AGI do not directly apply. The company is confident that users will see the difference through hands-on use.

Why this matters for the AI industry

The ability to automate fine-tuning could lower barriers for smaller teams and organizations that lack the resources of major AI labs. If AutoScientist delivers on its promise, it may accelerate innovation in specialized fields such as medical imaging, legal document analysis, and scientific research. Hooker compared the potential impact to that of code generation tools, which have dramatically expanded what developers can achieve.

“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,” she said.

Pricing and availability

To encourage adoption, Adaption is making AutoScientist free for the first 30 days after release. This trial period is intended to let researchers and developers evaluate the tool’s performance on their own tasks before committing to a paid plan. The company has not yet disclosed pricing beyond the initial free window.

Conclusion

AutoScientist represents a notable step toward autonomous AI improvement, a goal that has long been a focus of research labs and investors alike. By automating the fine-tuning process, Adaption aims to democratize access to frontier-level model training. Whether the tool lives up to its ambitious claims will depend on real-world results, but the company’s decision to offer a free trial suggests confidence in its technology.

FAQs

Q1: What is AutoScientist?
AutoScientist is an AI tool developed by Adaption that automates the fine-tuning process, allowing models to train themselves on specific tasks more efficiently.

Q2: How does AutoScientist differ from traditional fine-tuning?
Traditional fine-tuning requires manual data curation and parameter adjustment. AutoScientist automates these steps by co-optimizing data and model architecture through an iterative learning process.

Q3: Is AutoScientist free to use?
Adaption is offering AutoScientist free for the first 30 days after launch. Pricing beyond that period has not yet been announced.