News for hackers. Stuff that matters.
Re: by thaw13579 on Sunday February 12, 2023 @03:07PM (#34766350)
The sources are there in the training dataset, they are just not linked to the response. I don't think this is an inherent property of LLMs though, and I imagine future iterations will have some sort of attention mechanism that highlights the contributing source materials.

There are no comments on this comment yet.