DEV Community

Shreya
Shreya

Posted on

Day 1: Unsupervised Learning

When you ask AI “What is 9+10”, how should it respond?

  • A calculator app would respond with 19.
  • A printer app would respond with “What is 9+10?”.
  • An AI app that’s been trained only using data from social media might respond with 21.

We know that AI apps are different from more conventional apps. If you give a conventional app a certain input, it will follow the code or instructions or algorithm to do what the programmer tells it to do.

Although chatgpt was trained on a seriously large dataset beyond just social media content, it does have the ability to use context from a conversation to make decisions about how to respond to new input. So if you ask chatgpt what is 9+10 based on social media viral content, it will say 21. Obviously that’s not the same as just asking the question by itself, but that’s how our brains work too. If you ask a stranger on the street with no additional context they would probably say 19. But if you ask your friend, they’d probably have enough context to say 21.

ChatGPT is trained using a method called unsupervised learning, which according to itself "is a type of machine learning where the algorithm is given unlabeled data and must find patterns, relationships, or structures within it without explicit guidance.”

So you have an algorithm that has no prior information but it has a task. That task can be to either “cluster” the data into groups or use “dimensional reduction” to simplify it. You give it unlabeled data and it will do one of these tasks. The output is then evaluated for quality and if it’s an iterative process, it can be refined and adjusted and the process repeated until the results are good enough according to people.

Top comments (0)