this post was submitted on 11 Oct 2024
713 points (99.4% liked)
Technology
59569 readers
4136 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Best case is that the model used to generate this content was originally trained by data from Wikipedia so it "just" generates a worse, hallucinated "variant" of the original information. Goes to show how stupid this idea is.
Imagine this in a loop: AI trained by Wikipedia that then alters content on Wikipedia, which in turn gets picked up by the next model trained. It would just get worse and worse, similar to how converting the same video over and over again yields continuously worse results.
A very similar situation to that analysed in this paper that was recently published. The quality of what is generated degrades significantly.
Although they mostly investigate replacing the data with ai generated data in each step, so I doubt the effect will be as pronounced in practice. Human writing will still be included and even curation of ai generated text by people can skew the distribution of the training data (as the process by these editors would inevitably do, as reasonable text could get through the cracks.)
AI model makers are very well aware of this and there is a move from ingesting everything to curating datasets more aggressively. Data prep is something many upstarts have no idea is critical, but everyone is learning about, sometimes the hard way.