Adaption launches AutoScientist for self-training AI models
On Wednesday, Adaption introduced AutoScientist, an AI tool that automates fine-tuning to help models learn specific capabilities. Co-founder and CEO Sara Hooker, formerly VP of AI research at Cohere, said the system co-optimizes data and model to learn any capability. AutoScientist builds on Adaption's existing Adaptive Data offering, turning continuously improving datasets into continuously improving models. Adaption claims AutoScientist has more than doubled win-rates across different models, though conventional benchmarks like SWE-Bench or ARC-AGI are not applicable. The tool is free for the first 30 days after release. Hooker compared its potential impact to that of code generation.
Key facts
- Adaption launched AutoScientist on Wednesday.
- AutoScientist automates fine-tuning for AI models.
- Sara Hooker is co-founder and CEO of Adaption.
- Hooker previously worked as VP of AI research at Cohere.
- AutoScientist co-optimizes data and model.
- It builds on Adaption's Adaptive Data offering.
- Adaption claims AutoScientist more than doubled win-rates.
- AutoScientist is free for the first 30 days.
Entities
Institutions
- Adaption
- Cohere
- TechCrunch