Key Concepts Of Generative AI. explained to a 12 year old
3 minute read
How to Teach a Robot to Speak and Understand Human Language This robot’s brain is made up of a complex network, similar to how our brains are made up of neurons. This complex network is called a neural network. To make the robot learn, you need to adjust its settings or “knobs” called parameters.
You teach the robot by giving it lots of examples, and this process is called training. Now, let’s say you want the robot to not only understand language but also create new sentences, stories, or even poems on its own. This is called generative AI. To do this, you first give the robot lots of data to learn from, which is called pre-training.
After pre-training, you might want the robot to become better at specific tasks, like writing poems. You can further adjust the robot’s settings to make it better at these tasks, and this process is called (fine) tuning.
Now, when you have a big, fancy robot brain that’s been pre-trained and can be fine-tuned for specific tasks, you can call it a foundation model. When your robot works specifically with text, it’s called a large language model (LLM).
To make your robot create something specific, you give it a starting point or idea, which is called a prompt. Your goal is to make the robot do exactly what you want, and this is called alignment. Sometimes, the robot might create things that aren’t accurate or don’t make sense; these are called hallucinations.
The robot is designed to understand and create natural language, or language that feels normal to humans. To do this efficiently, it breaks down words into smaller parts called tokens through a process called tokenization.
If you want your robot to learn about a new topic or become better at a specific problem, you can adapt or tune it for that purpose. You can also adjust a setting called temperature to make the robot’s output more or less varied — imagine the robot becoming more creative or more predictable.
To help the robot understand the relationships between words or sentences, you can represent them as lists of numbers called embeddings. These lists have unique properties that help the robot understand how similar or different words and sentences are.
Finally, if you want to make your advanced robot brain smaller, faster, and cheaper without losing its performance, you can use a process called distillation. This involves creating a new, smaller robot brain (the “student”) that learns to mimic the bigger, original robot brain (the “teacher”).