For years, AI researchers have anticipated the moment when systems will improve themselves more effectively than humans can. With investors pouring resources into a new generation of research-driven labs, that goal is moving closer to reality. On Wednesday, Adaption introduced AutoScientist, a product designed to help AI models learn specific capabilities faster through an automated approach to fine-tuning.
What AutoScientist does differently
AutoScientist builds on Adaption’s existing data platform, Adaptive Data, which focuses on creating high-quality datasets over time. The new tool takes that a step further by turning continuously improving datasets into continuously improving 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 the model itself.
Also read: Amazon replaces Rufus with Alexa for Shopping, a new AI assistant powered by Alexa+
“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.”
Technical approach and results
AutoScientist uses an automated, iterative approach to conventional fine-tuning. Instead of relying on manually curated datasets and human-guided training loops, the system dynamically adjusts its training strategy based on real-time performance feedback. Adaption claims the tool has more than doubled win-rates across different models in internal tests, though the company acknowledges these results are difficult to benchmark against standard industry metrics like SWE-Bench or ARC-AGI, since AutoScientist adapts models to specific tasks rather than general performance.
Also read: Anthropic overtakes OpenAI in business customer count, Ramp data shows
To encourage adoption, Adaption is making AutoScientist free for the first 30 days after release.
Why this matters for the AI industry
The announcement comes at a time when the cost and complexity of training frontier AI models have become a major barrier to entry. Most leading models are developed by a handful of well-funded labs. If AutoScientist delivers on its promise, it could lower the barrier for smaller teams and specialized applications, enabling more organizations to fine-tune models for niche use cases without requiring deep in-house expertise.
“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.
Conclusion
AutoScientist represents a significant step toward self-improving AI systems, though independent validation of its claims is still needed. By automating the fine-tuning process, Adaption is positioning itself at the intersection of data quality and model optimization, a space that is likely to become increasingly competitive as more players enter the market. The 30-day free trial will give early adopters a chance to evaluate the tool’s real-world impact.
FAQs
Q1: What is AutoScientist?
AutoScientist is an AI tool from Adaption that automates the fine-tuning process, helping models learn specific capabilities by co-optimizing both the training data and the model itself.
Q2: How is AutoScientist different from standard fine-tuning?
Traditional fine-tuning relies on manually curated datasets and fixed training loops. AutoScientist dynamically adjusts its training strategy based on performance feedback, aiming to improve efficiency and outcomes.
Q3: Is AutoScientist available to the public?
Yes. Adaption is offering AutoScientist free for the first 30 days after its release. After that, pricing details have not yet been announced.
Be the first to comment