Post #1450549

It's going to become increasingly challenging to identify when, in the authorship and publishing process, AI hallucinations were introduced. Mass-scale publishing leaders are as keen for the technology as other CEOs and it means that reference parsing, for instance, could go through an LLM. Rigorous versioning is, of course, the answer so that we can do the equivalent of the software engineering "blame" command, to find out where. But most industrial publishing manuscript systems do not store the MS in any diff-able format. It's just a binary. This will, for now, make hallucination hunting, when needed, a very slow and manual process, involving a human trawling through each version of a file, to see who introduced the nonsense...