A friend sent this link:
https://ravenation.club/@bearsong/111232965922163404
Oh hey look, LLMs are dumb.
So I tried it with GPT4:
And:
Which is IMO actually really impressive.
I actually see a lot of this where people post screenshots saying either “Wow, look how amazing GPT is” or “Wow, look how dumb GPT is”.
Its really important to check up and see what you get. Here’s another
https://mastodon.social/@MattHodges/111273866044538701
Here’s what @MattHodges got:
Which is really impressive. So I tried it:
Its possible some differences in the original image occurred due to compression etc so I’m not using the exact same input maybe. Still In my version it reads the situation in the image correctly, it just gets the wrong answer.
After clicking regenerate, I get the correct one:
Interestingly looking closer at the original:
Leave a Reply