Running out of data?
Mark hears that AI will soon have read all the text on the internet, and wants to know where it will get more data to learn from.
SYNTHETIC DATA
Data made by AI
Bit explains: one way is synthetic data, data made by AI itself. For example, AI writes millions of math problems with solutions.
FILTERING
Check before using
The key step is careful filtering: recheck the math, run the code, and only keep what's correct.
WHAT IT'S GOOD FOR
Filling gaps where data is rare
It's useful where real data is rare or sensitive: unusual situations, or made-up records so real people's information isn't exposed.
RISKS
A photocopy of a photocopy
The risk: if AI only learns from AI, over time it's like a photocopy of a photocopy, getting blurrier. This is called model collapse, so real data always has to be mixed in.
REMEMBER
When you see this word
When you see this word, think: data AI makes to teach AI, useful when it's filtered carefully.
This article is based on the video Synthetic data: AI making data for AI from the Mark học AI channel. Watch the video (in Vietnamese) to see the animations.