That just sounds like you’re unaware of the other times you asked AI for an answer and got the wrong one, but accepted it as fact anyway.
A while back someone suggested asking AI about some niche hobby or interest that you’re already incredibly educated or experienced in, and see how well it does. It’ll usually get the main idea, but definitely starts falling apart within the details and nuances. Thing is, you wouldn’t know where to draw those lines if you weren’t already versed in that matter.
Suggesting to my friends that they should look up a topic that they’re well versed in and know inside and out has been a great way to get some friends of mine to start being far more critical of LLMs. Suddenly it wasn’t seeming that smart anymore to them.
The thing is, I have indeed done this many times, in fact I do it every time a new model comes out. Last year it was an intern at best, but now? It’s easily a coworker. Which would be nice, only this happened in less than a year. What happens in 5?
We’ve run tests with Astra this week for a 3d artist. Turns out the output quality is slightly better than most juniors in the field. Two weeks ago, the best model in the world couldn’t do beat a child.
You don’t need AI for this. You can just ask a human, they’ll induce you in error just as often.
When was the last time a movie box office told you the wrong time for a movie?
The box office, I wouldn’t know, but friends? At least once. Which isn’t much, but it’s one time more than AI.
That just sounds like you’re unaware of the other times you asked AI for an answer and got the wrong one, but accepted it as fact anyway.
A while back someone suggested asking AI about some niche hobby or interest that you’re already incredibly educated or experienced in, and see how well it does. It’ll usually get the main idea, but definitely starts falling apart within the details and nuances. Thing is, you wouldn’t know where to draw those lines if you weren’t already versed in that matter.
Suggesting to my friends that they should look up a topic that they’re well versed in and know inside and out has been a great way to get some friends of mine to start being far more critical of LLMs. Suddenly it wasn’t seeming that smart anymore to them.
The thing is, I have indeed done this many times, in fact I do it every time a new model comes out. Last year it was an intern at best, but now? It’s easily a coworker. Which would be nice, only this happened in less than a year. What happens in 5?
We’ve run tests with Astra this week for a 3d artist. Turns out the output quality is slightly better than most juniors in the field. Two weeks ago, the best model in the world couldn’t do beat a child.