Adaption launches AutoScientist to automate AI model fine-tuning

In short: Adaption has launched AutoScientist, a new tool that automates the process of training an AI model to do a specific task.

What happened

Adaption introduced a product called AutoScientist on May 13. The company says the tool helps AI models pick up specific capabilities quickly.

To understand what it does, it helps to know what “fine-tuning” means. Fine-tuning is when you take an existing AI model and retrain it on new examples so it gets better at a particular job, like answering questions in a company’s style or handling a certain type of document (like giving a chef a base recipe and then practicing it until it matches a restaurant’s menu).

Adaption says AutoScientist automates this fine-tuning work. CEO and co-founder Sara Hooker told TechCrunch the tool tries to improve both the training data and the model at the same time, so the model learns in the most effective way for the task.

AutoScientist builds on Adaption’s earlier product, Adaptive Data, which is meant to help teams create and improve datasets over time. A dataset is a collection of examples used to train an AI system (like a workbook full of practice questions and answers). Adaption’s pitch is that continuously improving data can lead to continuously improving models.

Adaption also says AutoScientist more than doubled “win-rates” across different models in its own testing. The company notes these results are hard to compare to common public tests because AutoScientist is aimed at custom tasks.

AutoScientist will be free to use for the first 30 days after release.

Why it matters

If AutoScientist works as promised, it could make it cheaper and faster for more organizations to adapt AI models to their own needs, without needing as much hands-on trial and error.

Source: TechCrunch AI