ModelFactory sounds like the dream. Fine-tune an LLM through a clean GUI, no terminal, no API keys, just permissioned data from the @OpenLedger network.
I want to believe it.
But here's what keeps nagging me. The whole pitch rests on "approved" datasets. Who approves them?
If a contributor uploads biased medical data and it passes review, every model fine-tuned on it inherits that bias at scale. The GUI doesn't show you that. It just shows a checkbox.
Imagine a hospital team spinning up a diagnostic model in an afternoon, trusting the green light. They never see the dataset's lineage, only its label.
Convenience is a feature. It's also a blindfold.
So my question is simple. When attribution is automated, who actually stays accountable for what the model becomes?