Adaption has released AutoScientist, a fully automated system for model training and alignment. It runs the entire research loop on its own, simultaneously optimizing data and training recipes — meaning even non-technical users can fine-tune custom models without hand-holding.
In testing, AutoScientist outperformed Adaption's own internal researchers' manual configurations by an average of 35%. Across eight vertical benchmarks using the Together AI architecture and dataset sizes ranging from 5,000 to 100,000 samples, the system boosted fine-tuning win rates from the human-configured 48% to 64%.
Most private-data fine-tuning efforts fail due to catastrophic forgetting and overfitting, but AutoScientist delivers predictable gains. Adaption says it not only lowers the barrier to building custom AI but also frees machine learning engineers from babysitting parameter searches. The system is available now for a 30-day free trial, and the team promises future real-time adaptation technology that won't require training at all.