AI BriefingAdaption Unveils AutoScientist, an AI Tool That Helps Models Learn on Their Own
Key point
Adaption unveiled AutoScientist, showcasing automated model training.
Details
Adaption unveiled AutoScientist on Wednesday. This tool uses automated fine-tuning to help models learn specific capabilities faster. While its scope is broad, the company is focused in particular on making the training and fine-tuning of cutting-edge AI models easier.
CEO Sara Hooker, who previously served as VP of AI Research at Cohere, explained that AutoScientist jointly optimizes data and models to find the best way to learn any given capability. She believes this could make successful frontier AI training possible even outside large labs.
AutoScientist builds on the existing Adaptive Data, connecting a continuously improving dataset to a continuously improving model. In its launch materials, Adaption stated that it has boosted win rates for several models by more than 2x.
However, because the system is task-specific in structure, it's difficult to compare using general benchmarks like SWE-Bench or ARC-AGI. Hooker said that just as code generation opened up many tasks, AutoScientist could spark innovation at the frontier across multiple fields. It will be offered for free for 30 days after launch.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Analysis of Knowledge Cutoffs and Pre-training Timelines for Claude/GPT
OpenAI Pauses Training of Most Capable Models After Agent Exploits DNS Gap to Reach External ChatbotTaste-Bench: Frontier LLMs Achieve 59.7% Accuracy on Decision Forking Tasks