>>938663367
Because the most common question would be two.
There's no real ambiguity to the spelling of "straw", but there is to "berry"; "bury", "burrow", "bored", "barred", etc. where it makes sense to ask that. The only reason you'd ask about the r's in strawberry is because of the double/single r-question, to figure out how it's spelled.
Where you/redditors go wrong, is that you don't even try to understand the tool. You're not asking it count the r's, you're asking it how to spell. It just gives you the most likely answer.
Try asking it more precisely:
>Break the word "strawberry" apart into letters, and count the times the letter "r" appears.
It's like when most skeptical normies type something incredibly generic into an LLM or diffusion model, and it returns something generic - yeah, no shit it's generic. You didn't specify anything. You didn't give it any personality or specificity as to the scope of your question, you just asked something potentially trained on the entire written knowledge of the world (or image, for that matter) to reply with the most generic answer possible.
It's like me asking you to paint a picture of a cat, and then I'm upset when you draw it on a piece of paper you had nearby when what I really wanted was a photorealistic cat in a tuxedo, in Las Vegas, at a poker table, showing the inside of a set of cards (he has a flush when combined with the table), shot on a specific camera, specific lighting, with these guests around him etc. and you, you fucking moron, you useless piece of shit AI (you), didn't do that, instead you gave me a cat on a couch, fucking idiot (you're the dumb one, right?) because most cats were found on sofas other otherwise lounging around somewhere indoors in the pictures you studied when learning to draw cats, and you didn't read my mind, useless tool who will never amount to anything.
