this post was submitted on 17 Dec 2024
578 points (92.5% liked)

memes

10668 readers
2283 users here now

Community rules

1. Be civilNo trolling, bigotry or other insulting / annoying behaviour

2. No politicsThis is non-politics community. For political memes please go to [email protected]

3. No recent repostsCheck for reposts when posting a meme, you can only repost after 1 month

4. No botsNo bots without the express approval of the mods or the admins

5. No Spam/AdsNo advertisements or spam. This is an instance rule and the only way to live.

Sister communities

founded 2 years ago
MODERATORS
 
you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 57 points 5 days ago* (last edited 5 days ago) (3 children)

Ugh. Don’t get me started.

Most people don’t understand that the only thing it does is ‘put words together that usually go together’. It doesn’t know if something is right or wrong, just if it ‘sounds right’.

Now, if you throw in enough data, it’ll kinda sorta make sense with what it writes. But as soon as you try to verify the things it writes, it falls apart.

I once asked it to write a small article with a bit of history about my city and five interesting things to visit. In the history bit, it confused two people with similar names who lived 200 years apart. In the ‘things to visit’, it listed two museums by name that are hundreds of miles away. It invented another museum that does not exist. It also happily tells you to visit our Olympic stadium. While we do have a stadium, I can assure you we never hosted the Olympics. I’d remember that, as i’m older than said stadium.

The scary bit is: what it wrote was lovely. If you read it, you’d want to visit for sure. You’d have no clue that it was wholly wrong, because it sounds so confident.

AI has its uses. I’ve used it to rewrite a text that I already had and it does fine with tasks like that. Because you give it the correct info to work with.

Use the tool appropriately and it’s handy. Use it inappropriately and it’s a fucking menace to society.

[–] [email protected] 7 points 4 days ago* (last edited 4 days ago) (2 children)

I know this is off topic, but every time i see you comment of a thread all i can see is the pepsi logo (i use the sync app for reference)

[–] [email protected] 9 points 4 days ago (1 children)

You know, just for you: I just changed it to the Coca Cola santa :D

[–] [email protected] 3 points 4 days ago (1 children)

Spreading the holly day spirit

[–] [email protected] 1 points 4 days ago (1 children)

We are all dutch on this blessed day

[–] [email protected] 2 points 4 days ago

We are all gekoloniseerd

[–] [email protected] 2 points 4 days ago

Voyager doesn't show user PFPs at all. :/

[–] [email protected] 8 points 5 days ago (2 children)

I gave it a math problem to illustrate this and it got it wrong

If it can’t do that imagine adding nuance

[–] [email protected] 11 points 5 days ago (1 children)

Well, math is not really a language problem, so it's understandable LLMs struggle with it more.

[–] [email protected] 11 points 5 days ago (1 children)

But it means it’s not “thinking” as the public perceives ai

[–] [email protected] 5 points 5 days ago (2 children)

Hmm, yeah, AI never really did think. I can't argue with that.

It's really strange now if I mentally zoom out a bit, that we have machines that are better at languange based reasoning than logic based (like math or coding).

[–] [email protected] 1 points 4 days ago

Not really true though. Computers are still better at math. They're even pretty good at coding, if you count compiling high-level code into assembly as coding.

But in this case we built a language machine to respond to language with more language. Of course it's not going to do great at other stuff.

[–] [email protected] -1 points 4 days ago* (last edited 4 days ago)

Ymmv i guess. I've given it many difficult calculus problems to help me through and it went well

[–] [email protected] 2 points 4 days ago* (last edited 4 days ago) (1 children)

Wait, when did you do this? I just tried this for my town and researched each aspect to confirm myself. It was all correct. It talked about the natives that once lived here, how the land was taken by Mexico, then granted to some dude in the 1800s. The local attractions were spot on and things I've never heard of. I'm...I'm actually shocked and I just learned a bunch of actual history I had no idea of in my town 🤯

[–] [email protected] 2 points 4 days ago (1 children)

I did that test late last year, and repeated it with another town this summer to see if it had improved. Granted, it made less mistakes - but still very annoying ones. Like placing a tourist info at a completely incorrect, non-existent address.

I assume your result also depends a bit on what town you try. I doubt it has really been trained with information pertaining to a city of 160.000 inhabitants in the Netherlands. It should do better with the US I’d imagine.

The problem is it doesn’t tell you it has knowledge gaps like that. Instead, it chooses to be confidently incorrect.

[–] [email protected] 1 points 4 days ago

Only 85k pop here, but yeah. I imagine it's half YMMV, half straight up luck that the model doesn't hallucinate shit.