this post was submitted on 27 Oct 2023
301 points (96.6% liked)
Technology
59414 readers
3119 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
As long as LLM AI models are prone to hallucinating and there is no way to audit how they derive results (eg to verify accuracy), relying on them will have roadblocks/limitations. Once they solve this issue though, that will be a whole different story, I agree. As for other AIs such as image or video generation, I don't have enough experience to tell...
Hallucinations can be heavily reduced today by providing the LLM with grounding truth. People use naked LLMs as knowledge databases, which is prone to hallucinations indeed. However, provide them with verified data from the side and they are very, very good at keeping to the truth. I know, because we deploy these with clients to great avail.
Image, music, video models are making great strides and are already part of various pipelines, all the way up to the big boy tools like Photoshop (generative fill, for example).
The tech is being incorporated at a large scale by a lot of companies, from SME to megacorp. I don't see it going away any time soon, even if it doesn't improve from here on out (which it undoubtedly will).
The issue is that there are from time to time they still confidently hallucinate and there is no way to detect if they are right or not.
Hire 1 person to verify AI output instead of a dozen to make the content. If that one editor misses something, who cares when we live in a post-truth society where the media lies on purpose.
How many countries start with the letter K in Africa?
GPT-4:
Anecdotal evidence is useless because it can be contradicted with anecdotal evidence.
Hallucinations aren't the only issue with LLMs, they also have a limited amount of context they can recall and that problem won't go away.
That problem is very much being worked on
This mentality is the same as saying Wikipedia is not a source. I agree, there are better regulations that need to be implemented, but the speed at which these things are being churned out and tuned is mind blowing. It's like if Wikipedia started branching off into sub categories that each have their own specialty, which can be more easily moderated, and then folded back into the more general system.