Adaption Takes a Leap Forward with AutoScientist: An AI Tool That Helps Models Train Themselves Automatically
For years, researchers have been eagerly anticipating the day when artificial intelligence (AI) systems could improve themselves faster than humans. With an influx of investment into research-focused AI labs, there are now more resources available to pursue this goal. One such neolab has recently taken a significant step towards realizing it.
This week, Adaption unveiled a new product called AutoScientist, which aids models in acquiring specific capabilities swiftly by employing an automated method for conventional fine-tuning. This technique is applicable across various fields but particularly focuses on speeding up and simplifying the process of training and refining frontier-level AI models.
Co-founder and CEO Sara Hooker, who previously served as VP of AI research at Cohere, highlighted AutoScientist’s innovative approach to the AI training process. “What’s incredibly exciting about this is that it optimizes both data and model simultaneously, suggesting we can finally allow for successful frontier-level AI training outside these labs,” she told TechCrunch.
AutoScientist builds upon Adaption’s existing data offering, Adaptive Data, which aims to facilitate the creation of high-quality datasets over time. AutoScientist is designed to transform those continuously improving datasets into increasingly sophisticated AI models. “Our view at Adaption is that every part of this stack should be adaptable and able to optimize on-the-fly for any task,” Hooker stated.
While impressive, the company’s claims about doubling win-rates across different models remain challenging to contextualize because AutoScientist is built around adapting models specifically for particular tasks. Traditional benchmarks such as SWE-Bench or ARC-AGI don’t apply here. Nonetheless, Adaption is confident that users will recognize the difference once they start using AutoScientist.
As a result of this confidence, Adaption has made the tool freely available to users for the first 30 days after its release.
“Just like how code generation opened up many tasks, this will unlock a lot of innovation at the frontier of different fields,” Hooker concluded.
Key Takeaways
- Adaption has launched AutoScientist, an AI tool designed to help models learn specific capabilities more efficiently through automated fine-tuning.
- The tool aims to facilitate the development of high-quality datasets and adaptable AI models across various fields.
- AutoScientist is expected to unlock new areas of innovation by allowing for successful frontier-level training outside specialized labs.
Originally published at techcrunch.com. Curated by AI Maestro.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.