web analytics

AI isn’t the only one to hallucinate

Behold the two big fat simplified Chinese characters that Google OCR dropped in the middle of the page of the Victorian history book I’m working on. They were about that size, too.

Google Translate tells me it means “Lulu” – but of course it doesn’t mean anything. It’s pure glitch.

I asked ChatGPT to explain and it said modern OCR is trained on a multilingual character set and isn’t smart enough to stick with Roman characters. It saw a shadow or pattern I can’t see and dropped a character in. And once it’s farted one out, it said it’s not uncommon for OCR to repeat the character.

Google OCR has a little AI-sauce mixed in – which is one of the reasons its OCR is so clean – but not enough to know there wouldn’t be two big fat Chinese characters in the middle of an English history book.

This stuff isn’t taking anybody’s job yet. Managers who are firing on that basis are kidding themselves (or using it as an excuse).

Comments


Comment from durnedyankee
Time: April 1, 2026, 6:53 pm

Excuse would be correct.

Well, that and making sure their bonus will buy them that new Maserati at Christmas time.


Comment from QuasiModo
Time: April 1, 2026, 9:07 pm

Oracle just laid off 30,000 people…CEO Larry Ellison is at the center of the push for AI and has Trump’s ear…don’t think it’s going to end well for him.


Comment from Uncle AL
Time: April 2, 2026, 2:11 am

@QuasiModo — Yes, when the out of work tech bros of Redwood City get hungry enough they’ll eat him.


Comment from Bob Mulroy
Time: April 2, 2026, 2:39 am

Just conferred with non-AI Chinese scholars. LuLu is the dew that collects on plants in the morning.
I hope that helps.

Write a comment

(as if I cared)

(yeah. I'm going to write)

(oooo! you have a website?)


Beware: more than one link in a comment is apt to earn you a trip to the spam filter, where you will remain -- cold, frightened and alone -- until I remember to clean the trap. But, hey, without Akismet, we'd be up to our asses in...well, ass porn, mostly.


<< carry me back to ol' virginny