this post was submitted on 02 Nov 2025
1008 points (98.7% liked)
Programmer Humor
27720 readers
84 users here now
Welcome to Programmer Humor!
This is a place where you can post jokes, memes, humor, etc. related to programming!
For sharing awful code theres also Programming Horror.
Rules
- Keep content in english
- No advertisements
- Posts must be related to programming or programmer topics
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
I love how accurate this image is. As they hammer more the structure becomes more and more unstable.
I've noticed the AI search suggestions from Bing and Google are just becoming more and more inaccurate as they feed off themselves. I can't even imagine what it's doing to systems we can't see.
Everything is fine. It will become god and fix all our problems any second now, and if you say that's mathematically impossible you're just a hater.
"We will create God then ask him for money."
"It doesn't matter that we are burning 1 trillion dollars with no pat to revenue, because we will create God and it will make money meaningless."
"Also, it will cure cancer and solve Global Warming. I know dumb people like you that can't even get a trillion dollars to burn care about those things.
TIL Catholics invented AI
Exactly!
i mean, we have the solutions to a lot of problems already
traffic? public trains
hunger? just like, feed people
global warming? reduce fossil fuel usage and stop poisoning the oceans
homeless? literally so many empty homes, put people in them???
and we've had many smart people create step by step plans for all of these!
the thing is... i guess rich people want a solution that doesn't involve them paying for something that's good not just for them but for others? so all those plans fail at step 1 - have empathy
I mean, nitpick, but if you blamed mathematics you actually would be. The observation that AI/LLMs are highly unreliable and don't appear to be getting any better is empirical.
We ha e known since the 50s. The math was done.
Hmm. I'm not sure what you're referring to. Do you have a link?
No but Turing was involved, and the guy who wrote ELIZA
The tech isn't new. That's all the effort I'm willing to put in for this trash.
And that paper's name? Albert Einstein. I can't find anything on Weizenbaum and Turing authoring together. Weizenbaum seems to have written mostly prose and code, even - he's not really thought of for his mathematical innovations, although obviously math was his original field.
Back in the 50's people thought conventional algorithms, like everybody here has worked with, were going to reach human intelligence. They could play chess, and chess is smart guy stuff, so obviously recognising a bird should be easy, right? Well, they figured out that wasn't right, and so began the first AI winter.
The tech of deep neural nets is in fact fairly new. Like, arguably it didn't become a thing until the Cold War was ending, although there were a lot of precursors, and it kind of arrived gradually.
'Deep neural nets' that's not a thing, dude. If it was, it was out on ice for this trash.
Any comments on how you attempted to lie to us all there? To win an internet argument?
It is. It's one that has hidden layers, as opposed to a shallow neural net which does not. Shallow neural nets aren't really a thing anymore, so it's usually omitted, but historically things like the perceptron go back further, and they're conceptually simpler to update during training. They also can't really deal with anything nonlinear.
Okay boomer. You understand technology I guess. This technology that's so good and has so many applications and isn't a trillion+ dollar ponzi scheme, and is totally gonna summon us up god any minute now.
Thank you, yes I do. And like I said in OP it seems to be crapping out pretty good at it's current level of ability - maybe it's not a Ponzi scheme, but it is a giant overvalued bubble.
The internet can't hurt you, you don't have to lie to us. And it kinda pisses me off when people do anyway, because it makes it worse for everyone.
noai.duckduckgo my beloved, glad that's the best free option rn (although I think Kagi is getting popular too, but that's paid).
Kagi pays yandex for image search ewww. Use a searxng instance and thank me later. Seriously give it a try
Glad to know that option is available, I've been using DDG for years and use the AI assist like once a week when I have a really specific search about a movie or scene or actor or event happening 14 years ago going off memory. After about the 3rd search I actually get useful information from the AI assist.
Most recently I searched for "movie scene with slow motion cannonballs flying around while ship explodes" because I could remember that gif but not the movie. No search results were helpful, nor images or videos but the AI assist said it was a scene from the 3rd pirates of the Caribbean.
Also it's building on top of existing fragility (the thin pillar below), and only making it worse
I looked for some troubleshooting for my podcast player, and was really mistrustful of the AI summary, but it was right. But of course the solution was pulled from a forum, so could easily have been from a different product, or an outdated version. The last thing I looked for gave a well out dated answer for fingerprint readers under Linux
They don’t feed from each other. AI is trained with data from the before times. They only use good old regular search engine algos to get context data.
Simplicated:
Prompt->analyze->genereate search terms->analyse search results->generate response
They just make way more loops in real scenarios and even more if you let it “think longer”