undefined

points

[-]

> LLMs (I believe) _can_ be trained to not plagiarize. They do not have to be original, nor to use public domain training-data, they just have to give attribution (which, as far as I can tell, requires little more than a corpus and a stack of similarity-measures).

And something like attributions can be determined even if the LLM is paraphrasing, if I'm understanding Cosma Shalizi correctly[1]. He gives Grosse et al.[2] as a reference for doing so in a scalable manner.

[1] https://bactra.org/research/2026-03-10.html

[2] https://arxiv.org/abs/2308.03296

by fhdkweig2 hours ago|

prev|

[-]

In addition to giving credit where credit is due, attribution is important for verifying the trustworthiness of "facts". When someone tells me they asked a LLM and it says xxxxx, I want to know did it get it from a bad source, or worse, hallucinated it from thin air. And if it turns out to be good information, I might want to check out other things the guy has written.

At least with the old school web search engines, you got a hyperlink that you could use to find out more. Now the only way to learn more is to keep asking the LLM more questions. There is no way to get from the LLM to a source outside the LLM.

by fhdkweig1 hours ago|

parent|

[-]

And I hope I am just being a Chicken Little, but if LLMs become the go-to place for searching, we will lose traditional web search engines. They won't be profitable if not enough people use them. For those who don't remember how the web was before Google's search engine, we had search engines that really didn't work, web rings, and people's list of random bookmarks. The web is much too complex and big to ever be able to go back to those methods. The web will be effectively unnavigable and dead. All we will have left will be the unattributed knowledge distilled in the LLM with no way back to the old methods. And please tell me I'm wrong.

by novaRom2 hours ago|

prev|

[-]

> It clearly memorized the code

This code could appear many times in training data; but I wonder what decoding params did you actually used, because that effect appears likely with close-to-greedy way. Also, if it is a coding assistant model, chances are this skill was increased by additional sampling from that domain during training? Or maybe RL has teached model to provide verbatim answers for certain questions? So it was rewarded if that kind of answers is really helpful (I guess it was?).