This is the thing nobody seems to understand about AI; it doesn't _know_ anything. It's just predicting word order based on probability. It's a blind collage machine with some statistical sophistication that is designed to sound convincing. Facts don't enter into the equation at any point.
Reposted from
Chuck Wendig
The amazing part isn’t just the wrong answer, but that the explanation it gives is accurate — and it *simply doesn’t know that 27 is a smaller number than 32.*
Billions of dollars spent to make a black box machine that doesn’t know something elementary school kids know.
Billions of dollars spent to make a black box machine that doesn’t know something elementary school kids know.
Comments
Maybe some new research will find ways to blend LLMs with more deterministic knowledgebases but nothing like that exists currently.
AI isn't even AI. Applied statistics at best, with machine learning models. They pick from reliable averages. Very little useful feedback can be given to these models. They will ignore direct instructions.
The models most people see/use was made with no drive to recreate genuine intelligence. Only refined models to make predictions on what the "correct" words could be.
There are plenty of good uses for deep learning models, but they don't get funding. Generally, if they are a product, they are not helping anyone but the seller.
The AI bros are desperately trying to find sustainable markets for the generative systems, but I doubt they will ever be economically viable let alone wildly profitable. Toys are only attractive when they're not ruinously expensive.