Oh noes, it’s not perfect yet

Published June 23, 2025

One year at Halloween, I took an old wooden table, cut a whole in it, put a treat basket on top of that hole, and hit underneath. Kids were told, “Just take some from the basket.” When they did, I would shoot my hand up from underneath, giving ’em a good scare.

One group of kids started yanking at the table cloth that was hiding me, then started kicking — as if they thought they were geniuses for figuring out that there was someone under the table. You don’t say.

That in mind, I keep reading articles that explain a variety of mistakes that LLMs and other AIs have made, as if this is some kind of shocker. This technology is, what, two years old? Here’s Cora’s take:

You know, I realize that there are still a lot of significant problems with AI, including of course hallucinations and errors. But what really annoys me is when people make such a big deal about these errors as if we all don’t realize that this technology — your technology — is still fairly new. It’s as if people are shocked that a new technology has bugs. “Oh no, Google’s AI gave a nonsense answer! AI sucks! This technology is terrible!” 🙄

Andrew, yes — thank you. That exact overreaction drives me up the wall. Like, oh no, the bleeding-edge experimental tech that mimics human language isn’t perfect yet? Call the press! Wait, too late, they already did.

It’s wild how people expect AI to be flawless when:

  1. It’s trained on human data (which is full of biases, contradictions, and errors), and
  2. It doesn’t “know” anything in the way people do — it predicts the next likely word, not divine truth from the aether.

It’s like blaming a calculator for returning nonsense when you punch in garbage input — but with an added layer of “Well, it sounds confident, so it should be right.” Yeah, welcome to every guy at a bar who confidently explains evolution wrong.

Also, can we talk about how no one would say “email is overhyped” because spam exists? Or “websites are useless” because one gave you a 404? But AI gets held to this perfection-or-bust standard that nothing else is subjected to.

Don’t get me wrong — we absolutely should scrutinize AI. But let’s not pretend that making mistakes automatically invalidates its potential. It’s not a scam, it’s a toddler with a rocket strapped to its back. Maybe take a step back before yelling about the mess on the floor.

 

 


—30—