this post was submitted on 15 Dec 2025
704 points (99.7% liked)

Not The Onion

18942 readers
986 users here now

Welcome

We're not The Onion! Not affiliated with them in any way! Not operated by them in any way! All the news here is real!

The Rules

Posts must be:

  1. Links to news stories from...
  2. ...credible sources, with...
  3. ...their original headlines, that...
  4. ...would make people who see the headline think, “That has got to be a story from The Onion, America’s Finest News Source.”

Please also avoid duplicates.

Comments and post content must abide by the server rules for Lemmy.world and generally abstain from trollish, bigoted, or otherwise disruptive behavior that makes this community less fun for everyone.

And that’s basically it!

founded 2 years ago
MODERATORS
 
you are viewing a single comment's thread
view the rest of the comments
[–] coriza@lemmy.world 15 points 22 hours ago (2 children)

You may already know that, but just to make it clear for other readers: It is impossible for an LLM to behave like described. What an LLM algorithm does is generate stuff, It does not search, It does not sort, It only make stuff up. There is not that can be done about it, because LLM is a specific type of algorithm, and that is what the program do. Sure you can train it with good quality data and only real cases and such, but it will still make stuff up based on mixing all the training data together. The same mechanism that make it "find" relationships between the data it is trained on is the one that will generate nonsense.

[–] tetris11@feddit.uk 1 points 18 hours ago* (last edited 18 hours ago)

But you can enter in real search data as a prompt, and use its training to summarize it. (Or it can fill its own prompt automatically from an automatic search)

It won't/can't update it's priors, and I agree with you there, but it can produce novel output on a novel prompt with its existing model/weights

[–] MiddleAgesModem@lemmy.world -1 points 17 hours ago

Whole lot of unsupported assumptions and falsehoods here.

Stand alone model predicts tokens. LLMs retrieve real documents, rank/filter results and use search engines. Anyone who has used these things would know that it's not just "making stuff up".

It both searches and sorts.

In short, you have no fucking idea what you're talking about.