hi friends!
here at La Vie en Code—and in general—I strive to be an equal opportunity opportunity proponent AND critic of artificial intelligence and machine learning. and as such, I must draw attention once again to the major… major limitations of AI.
I present to you this video from the subreddit r/chatgpt. I frequent the forums to see what’s going on, and this one had me on the floor laughing.
(unfortunately, it’s not my own content, but i absolutely would have thrown my laptop across the room by the time he got results)
I will be speaking soon on why these types of 100% wrong-ass responses occur, because it feels a lot sometimes like a glitch, right?
it’s not.
it’s a feature, not a bug. ????
see, we train these models to behave the way they do. not intentionally… it’s not by design but by the very nature of AI.
AI and ML needs to be shown the path to understanding the things it knows, so it can branch off and logically adapt that knowledge to other situations.
like a child, it’s ALWAYS learning. neural networks are remniscent of our child brains, actually, as we soak up knowledge and build new neural connections every second of every day.
AI and Machine Learning are much the same—but quite frankly, it doesn’t know it’s wrong until you tell it, or it learns better from other sources.
and as it’s added to the database from which it learns, it is then able to apply that knowledge to future interactions.
the best you can hope for is that those connections and logic already exist, or it’s going to draw conclusions that are just plain WRONG.
(and it still might draw incorrect conclusions)
so the next time you tell DALL-E to draw a duck and it draws godzilla, you know why. it just don’t know.
which begs the question: should we be happy it doesn’t know?
is HUP the key to unlock the extinction of humanity?
maybe some day it will know how to do some basic things it struggles with today. it might be tomorrow, it might be next year. but the end goal is for it to learn, and FAST.
the lesson today is: AI is useful, but not perfect (another post topic coming up!)
happy AI chattin’, friends ????
N
