When I first looked at ModelFactory's no-code fine-tuning dashboard, it felt like a genuinely useful accessibility decision.
Then I started thinking about what no-code actually abstracts.
Fine-tuning parameters are not cosmetic choices. Learning rate, epoch count, data composition ratios, these decisions determine whether a specialized model improves on its domain or degrades on tasks the base model handled well before fine-tuning touched it.
A user who does not understand why those parameters exist can produce a model that passes basic evaluation and fails in deployment in ways that are genuinely difficult to diagnose.
The Proof of Attribution records where the model's knowledge came from. It does not record whether the fine-tuned model is performing better or worse than the base model on the tasks the user actually cares about.
Those are different kinds of information. The gap between them seems worth understanding before treating no-code accessibility as the same thing as reliable specialized model development.