A research-focused AI lab called Adaption has introduced a new tool that tackles one of the most ambitious goals in artificial intelligence: enabling AI systems to improve themselves more efficiently. The product, called AutoScientist, is designed to help AI models learn new skills faster by automating parts of the training process that are traditionally handled by human researchers.
Training advanced AI models usually involves a process known as fine-tuning, adjusting an existing model using carefully prepared data so it performs better at a specific task. This process is time-consuming, expensive, and often limited to a small number of well-funded labs. AutoScientist aims to simplify this by automatically adjusting both the data used for training and the model itself, based on what the system is trying to learn. This means the model can figure out how to learn more effectively with less manual intervention.
According to Sara Hooker, co-founder and CEO of Adaption, the key innovation lies in treating data and models as parts of a single, adaptive system. Instead of fixing the dataset first and then training the model, AutoScientist continuously improves both together. This approach could make it easier for organisations outside elite AI labs to train powerful models for specialised use cases.
AutoScientist builds on Adaption’s earlier tools that help create and improve training data over time. While those tools focused on making better datasets, AutoScientist goes a step further by using that data to automatically improve AI models themselves, allowing them to adapt to new tasks instead of being trained just once.
The company says early results from AutoScientist have been encouraging, with notable performance gains observed across multiple models. Because the tool is designed to adapt models to highly specific tasks, standard benchmarks are not always the best way to capture its impact. Instead, Adaption says the strongest results emerge when the system is applied to real-world problems, where task-specific improvements and faster learning become more apparent.
To encourage experimentation, the lab is offering AutoScientist free for an initial trial period. According to Sara Hooker, tools like this could play a role similar to early automated coding tools, which made it easier for more developers to build and experiment. If it works as intended, AutoScientist could help make advanced AI training less about sheer scale and more about smarter, more efficient learning.
Nominate now for ET AI Awards 2026
Training advanced AI models usually involves a process known as fine-tuning, adjusting an existing model using carefully prepared data so it performs better at a specific task. This process is time-consuming, expensive, and often limited to a small number of well-funded labs. AutoScientist aims to simplify this by automatically adjusting both the data used for training and the model itself, based on what the system is trying to learn. This means the model can figure out how to learn more effectively with less manual intervention.
According to Sara Hooker, co-founder and CEO of Adaption, the key innovation lies in treating data and models as parts of a single, adaptive system. Instead of fixing the dataset first and then training the model, AutoScientist continuously improves both together. This approach could make it easier for organisations outside elite AI labs to train powerful models for specialised use cases.
AutoScientist builds on Adaption’s earlier tools that help create and improve training data over time. While those tools focused on making better datasets, AutoScientist goes a step further by using that data to automatically improve AI models themselves, allowing them to adapt to new tasks instead of being trained just once.
The company says early results from AutoScientist have been encouraging, with notable performance gains observed across multiple models. Because the tool is designed to adapt models to highly specific tasks, standard benchmarks are not always the best way to capture its impact. Instead, Adaption says the strongest results emerge when the system is applied to real-world problems, where task-specific improvements and faster learning become more apparent.
To encourage experimentation, the lab is offering AutoScientist free for an initial trial period. According to Sara Hooker, tools like this could play a role similar to early automated coding tools, which made it easier for more developers to build and experiment. If it works as intended, AutoScientist could help make advanced AI training less about sheer scale and more about smarter, more efficient learning.
Nominate now for ET AI Awards 2026
Disclaimer Statement: This content is authored by a 3rd party. The views expressed here are that of the respective authors/ entities and do not represent the views of Economic Times (ET). ET does not guarantee, vouch for or endorse any of its contents nor is responsible for them in any manner whatsoever. Please take all steps necessary to ascertain that any information and content provided is correct, updated, and verified. ET hereby disclaims any and all warranties, express or implied, relating to the report and any content therein.
