From Data to Discovery: Automating the Model Adaptation Loop
Adaption Labs · online, with Prayag Dwivedi
A one-hour technical introduction to Adaption for researchers and Master's students, given as an Adaption ambassador. Why the dataset, not the model, is usually the bottleneck in adapting an LLM, and how Adaptive Data, Invent and AutoScientist close that loop. 30 minutes of talk, 15 of live demo, and 15 where the audience did the work.
- Why the dataset is the bottleneck, and what Adaption automates across data, training and evaluation
- What I measured: a claim I had published, withdrawn on stage after 176 fine-tuning jobs over 13 seeds went 13 of 13 the other way
- Hands-on exercise: everyone wrote a spec for a model they wish existed, covering behaviour, how to judge it and what a wrong answer looks like



















