Les Défis de ChatGPT Face aux Cadrans de Montre selon Ned Block
Bien sûr, je serais heureux de vous aider à rédiger un résumé journalistique en français. Cependant, vous n’avez pas fourni la transcription de la vidéo YouTube. Pourriez-vous, s’il vous plaît, partager la transcription ou des points clés que vous souhaitez inclure dans le résumé ?
Source : Robinson Erhardt | Date : 2025-01-04 14:00:55 | Durée : 00:01:14
reinforcement training does not work on people either btw. that's why we have streets lined with addicted and homeless
Wow, just tried it and he's right.
Narendra Modi brought this up at the AI conference in Paris😂😂😂😂
to get a left hand picture you need to tell it to create a mirror image of someone writing
It also can't draw a full wine glass.
Grok 3 can’t either. I tried over and over with precise prompts.
Confirmed on March 12, 2025. Even guy the same right handed photo shown here.
I asked it draw a picture of an Orange Clown. Worked perfect for me.
Manually photograph a watch or clock with the hands in every position (YES it will be a lot of images and YES it will take some time – hey! free pun!) then add them to the training data fully labelled – problem solved. The issue isn't with GPT it's with the fact that the basic model can't conceptualize the idea that the hands can "move" (or more accurately are not static objects within the paradigm of the original object)
I JUST tried this and what I got was a group of watches, ALL of them showing a different time!!!!
That's why intelligence is supposed to work without countless examples of the same thing. Even if you fix this there will always be mistakes in any data. Even in the scientific data. So an intelligence must draw its own conclusions and make its own experiments and generate thoughts only based on the perceived sense of what's correct. Correct can never be associated to the most occurrent thing. Almost always new information comes from unexpected ideas, due to the nature of "new" and so learning by the sheer power of numbers will never work. You have to cross check everything, including your own thoughts, and then some.
That's why if you ask AI to make pictures of a doctor or some manager or something like that you will almost always see a man and not a woman
ChatGPT is a piece of crap. Outside of basic searches, it can't be trusted, and you never know what will set off if its policy guidelines are based on what is considered respectable because the leaders of chatGPT are spineless.
AI is plagiarism thats all! Zero creativity, 100% plagiarism
AI is utter trash, ugly, disgusting trash
You also can't get midjourney to give you a blank white image. It always has to draw something. Most of the time you ask for a color, like "imagine a pure yellow image, nothing but yellow", it'll draw a bunch of women wearing yellow. I did finally get it to listen to me about a blank white image the other day, but it gave me images of white canvases on a wall, like some modern art about nothingness, but not an all white image, like I asked for, in many different ways.
Also unable to draw a glass of wine completely full to the top.
I just did it for fun, and yeah. Cant get either one. I even tried to prompt it saying make a man with a missing right hand, draw a circle with his left hand, and it said it was graphic content lol
Damn! He’s right. I just tried it. Twice. 10:10 came up each time no matter what time I asked for.
Same thing if you ask to see a full glass of wine. AI can't conceptualize, it mearly copies what it sees.
Same with "A glass full of wine". No matter how you change the prompt, it will always be half-full
It’s like if you ask for a glass of wine filled to the brim, it’s won’t ever do it
The final boss of gaslighting
Same with a full glass of wine
It cannot draw a glass of wine full to the very top.
It says that mRNA vaccines are safe and effective.
So what you're really saying is that it isn't actually intelligence and it's just copying shit on the internet.
If it were intelligent, you could tell it to only refer to images of left hands, or to actually understand what a clock face does as time moves and just generate it.
This really highlights that these supposed AIs aren't true intelligence – they don't take what they're told, comprehend it, and supply a response. They take all the words they're given and create something that matches those words. It's not actually thinking. It's a pattern-recognition machine.
I just tried the left hand image gen and nothing worked yet.
this will NOT age well.
Ask ChatGPT to draw a peaceful protest. I dare you. 😅
So is it just a search engine with different name.
Isn't this supposed to be a super intelligence in like 2 years? I wonder if this is a more foundational problem with the underlying architecture and not an issue of feeding it more data?
I tried both experiments as instructed in the video and the results were similar to what was explained. The pen was still in the right hand and the clock displayed the time as 10:08.
Then I asked it to generate an image of holding a feather in the left hand and a stick in the right hand and it got it correct I tried it with several of objects such as a fork and glass, a book and a bird, etc. And each time it was able to generate the image successfully. There was some "artistic license" taken with the objects themselves as I did not specify a canary or a open vs closed book, but that is ok for this experiment.
My conclusion is the more data the prompt contains the more accurate or closer the image will be to the desired results. Teach yourself how to be better at prompting and you will achieve better results. Think of it this way. You can use a shoe as a hammer or you can use a hammer; they both achieve the results, but one is much better than the other.
This is not different to me asking a Daily Mail reader to see my point of view. Just give up already.
Its cause right hand writing, drawing is right 😉 or cause Im Right Handed mainly, or its because 10:10 the angel numbers ! 🙏❤️ 100
Its cause right hand writing, drawing is right 😉 or cause Im Right Handed mainly
Its cause right hand writing, drawing is right 😉
Is this why i heard dimensional clocks in my house in three different locations and i recorded the sound of a dimentional clock No Joke !!!! But there was no clocks anywhere In sight or in my house !!! I promise on Everything im probably the first person To have a dimensional clock following them
I got a more entertaining one. Try to get it to draw Discworld.
I see a picture of a left hand and someone writing
Called hallucinating🤪 data servers don’t have those features in data centers, never fed so can’t produce 🤷♀️
AI is theft
AI still sucks.