2026-07-21 19:04 UTC
My response:
I think that including your content in LLMs is a very bad idea, and I am against it. LLMs are most accurately described as "sparkling autocorrect", using mathematics to approximate a statistically likely best response to a user's question. Unlike other numerical models, there are no bounds, no guarantees on the accuracy of the output, because it's not a numerical output, it's words with meanings that can't be quantified. (1/3)
Replies (1)
-
@hpcgal@mast.hpc.social 2026-07-21 19:05
As such, LLMs are not effective stewards of the information provided by ACM papers, books, and other materials. Knowing that companies that develop these models do not have a clear understanding of consent and the fact that no means no, I am concerned that ACM content will still make it into the models. In particular, government-sponsored research will definitely make it into their corpus. (2/3)