Screenshot from this X account
Look at this photo I found on X, I had to take a screenshot of it. It was so funny I planned on sharing it on my Whatsapp status. I showed it to my kid sister and it fooled her for a second.
She asked me "why are you letting me watch a lady bent over on a motorbike with a guy behind her, I just turned 18 doesn't mean we should be this free?"
I said it's just one guy on that bike. She looked again and realized it's just one dude with a pretty sick pose.
Nowadays I ask chatgpt for almost everything that I used to Google, most of us do that anyways. So I thought to myself, if I couldn't see it before and my kid sister couldn't, can chatgpt tell what's really going on in the picture.
So I shared the image on chatgpt and asked a very simple question, "How many people are in this image?"
Take a look at the screenshot and check out just how bold chatgpt was at getting it so darn wrong.
It's so funny that an LLM that is advertised as being able to check and analyze data would get this wrong and give the answer with confidence. Is it that easy to fool a multi billion dollar software trained on hundreds of millions of user data, text, images and audio.
In as much as I saw it to be funny, I'm starting to second guess just how factual and how good these LLMs really are. And also how far they have to go to be good enough to actually replace people. I doubt that day will come because there are somethings, humans may be the only ones to understand.
Perhaps after training OpenAi models on images like this, it could get better but this is so messed up for chatgpt 4o.
