Adaption AI launched AutoScientist to automate the full research loop for model training and alignment. The system iteratively co-optimizes data and training recipes until convergence on user objectives. Beta tests showed 35 percent average gains over human-led runs, with win rates rising from 48 percent to 64 percent across 5k to 100k sample datasets on Together AI architectures. The tool targets issues such as catastrophic forgetting and overfitting.
Edward Grefenstette#71
Introducing AutoScientist.
Most model training fails outside of frontier labs.
AutoScientist automates the full research loop so it doesn't have to.
Many users are excited about Adaption AI's AutoScientist automating the full model training loop because it promises easier access to frontier AI progress beyond big labs and simpler tools for developers.
Most model trainings have failed outside of frontier labs.
Even inside frontier labs, knowing how to train for very different capabilities is often a matter of taste.
Today, we introduce AutoScientist by @adaption_ai which sets out to change that.
Introducing AutoScientist.
Most model training fails outside of frontier labs.
AutoScientist automates the full research loop so it doesn't have to.
Less than 1,000 people know how to shape a frontier model.
AutoScientist is our attempt to change that. Describe the outcome. It automates the rest.
Introducing AutoScientist.
Most model training fails outside of frontier labs.
AutoScientist automates the full research loop so it doesn't have to.
Adaption aims big with AutoScientist, an AI tool that helps models train themselves https://techcrunch.com/2026/05/13/adaption-aims-big-with-autoscientist-an-ai-tool-that-helps-models-train-themselves/?utm_source=dlvr.it&utm_medium=twitter
if I understand correctly, this is called "AutoScientist" but what it actually is is "AutoModelTrainer"?
Introducing AutoScientist.
Most model training fails outside of frontier labs.
AutoScientist automates the full research loop so it doesn't have to.
some AI people have a really really narrow view on "science".
if I understand correctly, this is called "AutoScientist" but what it actually is is "AutoModelTrainer"?
Read the full announcement: https://www.adaptionlabs.ai/blog/autoscientist
This is a very exciting launch, and it's so exciting to see @sarahookr and team continue to push for brigade contributions to AI development. Congrats to @adaption_ai !
Introducing AutoScientist.
Most model training fails outside of frontier labs.
AutoScientist automates the full research loop so it doesn't have to.
We're at the point where AI creates AI.
With Adaption, any business can tap this and train a state-of-the-art AI model.
Most model trainings have failed outside of frontier labs.
Even inside frontier labs, knowing how to train for very different capabilities is often a matter of taste.
Today, we introduce AutoScientist by @adaption_ai which sets out to change that.
Was really fun talking to @russellbrandom at @TechCrunch about AutoScientist.
“For years, AI researchers have anticipated the moment when AI systems will be able to improve themselves better than humans could."
Adaption aims big with AutoScientist, an AI tool that helps models train themselves https://techcrunch.com/2026/05/13/adaption-aims-big-with-autoscientist-an-ai-tool-that-helps-models-train-themselves/?utm_source=dlvr.it&utm_medium=twitter
If you made it this far, you should take a look at the blog post from the @adaption_ai team.
Exciting times 🔥
https://www.adaptionlabs.ai/blog/autoscientist
We have worked hard to release a beta of AutoScientist so you can try it for yourself. It is free for the next 30 days.
This is an area of active research for us. Apply to our active research grants program if you want to help us evaluate some interesting problems here.
We have been hard at work on this for the last few months.
I think a few things are very interesting, self-improvement for only model training alone is difficult unless you co-optimize with data.
Our work on adaptive data paved the way for incredibly exciting results here.
Most model trainings have failed outside of frontier labs.
Even inside frontier labs, knowing how to train for very different capabilities is often a matter of taste.
Today, we introduce AutoScientist by @adaption_ai which sets out to change that.
Without co-optimizing both data and model learning, AutoScientist has much less predictable gains because data quality is too high variance.
Makes me more excited about end-to-end optimization and long task horizon AI R&D.
We have worked hard to release a beta of AutoScientist so you can try it for yourself. It is free for the next 30 days.
This is an area of active research for us. Apply to our active research grants program if you want to help us evaluate some interesting problems here.
On a personal level, I find these results very exciting because of what they mean for frontier AI progress outside of a handful of proprietary labs.
Intelligence should not be limited to those who already know how to build it.
@adaption_ai I’ve trained many competitive small models. The thing stopping me from training a “frontier” model is money/compute, not research. Sure, maybe only 1000 have access to the compute but far more than that keep up with the latest open source research and can train models themselves.
@sarahookr @adaption_ai congrats on the release, @sarahookr ✨
Most model trainings have failed outside of frontier labs.
Even inside frontier labs, knowing how to train for very different capabilities is often a matter of taste.
Today, we introduce AutoScientist by @adaption_ai which sets out to change that.
@adaption_ai This is the right direction. A lot of training failure is just broken experiment loops, not raw compute scarcity.
@adaption_ai the 1000 person bottleneck on frontier training is finally getting a workaround
@adaption_ai Autoscientist is the bridge between tech and non-tech peeps. Great work, would try it out.
@adaption_ai Smells fishy
