The conversational sibling of
Emrahisik/rubric-dataset.
That one teaches a model to fill in a rubric. This one teaches an agent that has
just researched a subject on the web to do two things a small base model does
badly:
Cite what it was given, and only that. Handed five numbered sources, a
weak base writes a fluent verdict drawn largely from its own pretraining and
cites nothing — indistinguishable, to a reader… See the full description on the dataset page:
https://huggingface.co/datasets/Emrahisik/persona-dataset.
