PhD levels of intelligence, I seem to have read somewhere. These LLMs were released too soon.
this post was submitted on 20 Dec 2025
240 points (97.2% liked)
Fuck AI
5268 readers
2108 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
founded 2 years ago
MODERATORS
Do you have a PhD in biology? If not, how can you say it's wrong? Are you telling me that the x-ray fish picture is wrong? Or the Y place is? You can't Claim it doesn't have such intelligence level without said intelligence level, duh!
Having known people with PhDs, it's closer than you think. It says more about people with PhDs however!
PhDs tend to be specialists. At the extreme end, they can approach idiot savants. Idiot savant is a relatively good way to describe an LLM.
I LOVE THEM SO MUCH
THEY ARE CALLED TUGHGLE AND UGHGLE
(stolen from https://grok.com/share/c2hhcmQtMw_ae619099-f31c-4836-9520-99989351b031?rid=d40ed2b9-31d5-4d21-9bce-83271c137f65)
Edit: i keep coming back to look at them and they make me so happy 😭😭😭
Love the tiger for 's' as well!
These animals look so cursed and goofy.
Everything is cat
But on the bright side, we found a nice new mascot for Owlcat Games.
It was trained by stealing data from social media posts. As we all know, the entire point of the internet is sharing cat pictures, so that checks out.
We've reached the cat singularity
Truly horrifying
The only thing I ever liked about genAI is how hilariously it can mess up. Then people wanted to use it for their little side hustle.
There are tons of such images in the internet, but as ppl lost the ability to use search engine, now everyone seems to use ai as replacement.
what a fucking idiots
Have you tried using a search engine to find a picture lately? You'll get 75% AI results.
Well, yeah, that's part of mine job. And I put enough efford to make it from 70% to around 10 or less.
but indeed, it needs efford.
Well first of all, use a non shit one.
Super tempted to pay for Kagi. None of the free search engines I tried worked for me. There are differences, but each is bad in a different way.
You can self host, or use somebody else's SearXNG, without paying
I paid one year. Speaking of search quality, it finds the stuff most of the time, but the company is also heavily focused on developing AI.
IT FAILS AT EVERYTHING
It never fails to disappoint though
AI= ALWAYS INCORRECT
Why would you even use AI for something like this, though? Seems pretty pointless
No idea why OP did it, but for me it demonstrates that the claims of techbros that these LLMs are working on a reasoning level comparable to PhD, is wildly exaggerated. It puts into question if spending literal trillions of dollars for this crap is a good idea, when 250 billion (inflation adjusted) could build the large hadron collider, or a meager 25 billion a year could prevent world hunger.
When paying humans less for their work is your goal, no amount of computing power is too much to not justify the cost
While it's easy to demonstrate it falling at "real work", most of those examples only can be understood by people familiar with the niche.
So this is basically showing failure on what "should" be their turf. A strong point of the LLMs is supposed to be milquetoast stuff like this, where pretty much this verbatim sort of content should have appeared over and over and over in the training content.
For the supposed best of the best, worthy of billions upon billions directly and even more indirectly, this is just a way to be extra embarrassing.
The kicker is when the tech bros get confronted repeatedly with the same exact dodgy behavior, they just smile and laugh. Watched a WSJ video about a vending machine experiment by anthropic that would have just been brutally embarrassing but the company rep just laughed it off. Not when the usual "we will fix that right away" ir even "you used it wrong" just a general attitude that screened "it doesn't even matter, everyone is going to give us money anyway"
Totally not a bubble
🫧🫧🫧
U is for Ububbles
I rode my X-ray Fish to the store earlier. But he was in a bad mood. He kept barking at people.
Currently on mobile so cannot find the gif, but someone should post the gif for this here:
Starts with:
A is for:
Americans want to bomb me!
Remember to look both ways before an X-ray fish crossing!
This is kind of interesting because it's exactly the sort of "hard problem" that AI researchers have been trying to solve for a long time now. For a prompt like this to work, the inference step needs to be really fucking good at directing the rest of the model(s).
For reference, anyone that uses AI image generation regularly would view this anyone that tries this as completely ignorant of how the tech works or just, "wishful thinking."
Someone with experience would generate one letter image at a time. Otherwise you'll run into all sorts of problems such as all the animals looking like cats 🤣
Walter screaming in car.jpg
Don't do this, it'll become the next benchmark to test generative AI in a few months!