this post was submitted on 12 Mar 2025
1146 points (99.2% liked)
Technology
66067 readers
5439 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 other!
- 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, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
99.999% accurate would be pretty useful. Theres plenty of misinformation without AI. Nothing and nobody will be perfect.
Trouble is they range from 0-95% accurate depending on the topic and given context while being very confident when they’re wrong.
The problem really isn't the exact percentage, it's the way it behaves.
It's trained to never say no. It's trained to never be unsure. In many cases an answer of "You can't do that" or "I don't know how to do that" would be extremely useful. But, instead, it's like an improv performer always saying "yes, and" then maybe just inventing some bullshit.
I don't know about you guys, but I frequently end up going down rabbit holes where there are literally zero google results matching what I need. What I'm looking for is so specialized that nobody has taken the time to write up an indexable web page on how to do it. And, that's fine. So, I have to take a step back and figure it out for myself. No big deal. But, Google's "helpful" AI will helpfully generate some completely believable bullshit. It's able to take what I'm searching for and match it to something similar and do some search-and-replace function to make it seem like it would work for me.
I'm knowledgeable enough to know that I can just ignore that AI-generated bullshit, but I'm sure there are a lot of other more ~~gullible~~ optimistic people who will take that AI garbage at face value and waste all kinds of time trying to get it working.
To me, the best way to explain LLMs is to say that they're these absolutely amazing devices that can be used to generate movie props. You're directing a movie and you want the hero to pull up a legal document submitted to a US federal court? It can generate one in seconds that would take your writers hours. It's so realistic that you could even have your actors look at it and read from it and it will come across as authentic. It can generate extremely realistic code if you want a hacking scene. It can generate something that looks like a lost Shakespeare play, or an intercept from an alien broadcast, or medical charts that look like exactly what you'd see in a hospital.
But, just like you'd never take a movie prop and try to use it in real life, you should never actually take LLM output at face value. And that's hard, because it's so convincing.