Elektrine lite

← Feed

@michaelmaex@mastodon.online

2026-09-10 21:07 UTC

@antoinechambertloir@mathstodon.xyz It is almost tautological that if a mathematical work is in the training data, the LLM will reproduce it. The entire point is to imitate the training data. Worse is that it happens by nature without attribution. After all, if the training data claims to invent a new technique so will the output. This is fundamentally different to a text that is in the context, where the LLM is at least “aware” it is using something else, and it did not “reason” itself to that point.

Replies (2)

  • @michaelmaex@mastodon.online 2026-09-10 21:07

    @antoinechambertloir@mathstodon.xyz When I went through KU Leuven’s AI regulations towards the end of my phd there, there were very explicit warnings that LLMs can plagiarize without the user knowing through this very mechanism. It seems like open AI does not care about this warning and is very happy claim everything the LLM produces as something they deserve all credit for. This is essentially the same kind of theft that many artists have been mad about since stable diffusion models have gotten good.

    Open ##5153707

  • @michaelmaex@mastodon.online Indeed it had been put forward very early that this technology is not good as a search engine, precisely for the reason that it loses all references to the training data. However, “agentic models” are supposed to remediate such issues by launching independent search, adding annotations, etc. Does this actually work well, and efficiently?...

    Open ##5153710