Does AI learn from my questions?
Mark thinks that every time he asks a question, AI learns a little more from it.
TRAINING
Training: learn once, at great cost
Bit explains: there are two stages. Training is when the model learns, running on tens of thousands of GPUs for months and costing a fortune. Once it's done, the knowledge in the model is frozen.
INFERENCE
Inference: every time you ask
Inference is when the model uses that knowledge to answer you. Each question takes just a few seconds of computing, and the model doesn't change itself.
EXAMPLE
Like school and then work
It's like four years of university being training, and answering customers every day at work being inference.
REMEMBERING YOU
So how does AI remember me?
AI remembers your name thanks to the Memory feature, notes sent along with each question, as in videos 9 and 27. The model isn't learning again.
WHY IT MATTERS
It can run on your home computer
Knowing this shows you: training needs a whole data center, but inference for a small model can run right on your laptop.
This article is based on the video How training differs from inference from the Mark học AI channel. Watch the video (in Vietnamese) to see the animations.