AI, machine learning and deep learning
AI, machine learning, deep learning and LLMs are four nested ideas, not four words for the same thing. Watch each one try the same job, spotting a scam SMS, and see where each one breaks.
People use “AI”, “machine learning” and “deep learning” as if they meant the same thing. They don’t. Google Cloud describes them as a clear hierarchy: deep learning is a specialised type of machine learning, and machine learning is a core discipline within the broader field of artificial intelligence. LLMs sit inside deep learning.
- Artificial intelligence (AI) is the whole field: building systems that do tasks which normally need human intelligence. It includes rule-based systems, search and planning, not only learning.
- Machine learning (ML) is the part of AI where a model learns to make predictions from data, instead of a programmer writing every rule.
- Deep learning is the part of ML that uses neural networks with many layers. It learns its own features from raw data such as text, images and audio.
- Large language models (LLMs) are deep learning models trained on huge amounts of text to predict the next token. They’re a kind of generative AI: they create new content.
Press Play to watch each layer handle the same SMS. Then switch Try it to A sneaky scam and watch the hand-written rule fail.
- AI
- ML
- DL
- LLM
Your KYC has expired. Click http://bit.ly/kyc-upd to keep your UPI active.
Press Play.
- Hand-written rule?
- Machine learning?
- Deep learning?
- LLM?
Four terms people mix up, and how they fit inside each other. Press Play to see each one try the same job: is this SMS a scam?
The rule is real logic and the machine learning score is a real sigmoid over the weights shown. The weights, the deep learning percentages and the LLM replies are illustrative, not from a trained model.
What to notice
- Rules are AI, but brittle. The rule works only for the exact words its author expected. “K.Y.C” instead of “KYC” slips straight past. Every new trick needs a new rule.
- Machine learning learns the weights, not the clues. Training on labelled examples decides how much “has a link” or “urgent words” matters. But a person still had to choose those clues. This is called feature engineering.
- Deep learning learns the clues too. A deep neural network reads the raw text and discovers useful patterns itself. The cost: it needs much more data and computing power. See How a neural network learns.
- LLMs are general. One model, trained once on huge amounts of text, can classify the SMS, explain why, and answer the next question, all from a plain-English prompt. See What is an LLM?
- Each inner circle is also the outer ones. An LLM is a deep learning model, which is a machine learning model, which is AI. A spam rule is AI but not machine learning.
| Who writes the logic | Needs | Good at | |
|---|---|---|---|
| Rule-based AI | A programmer | Knowledge of the problem | Simple, fixed, explainable checks |
| Machine learning | Learned from labelled data, with hand-picked features | Labelled examples | Predictions on structured data |
| Deep learning | Learned from raw data | Lots of data and compute | Images, speech, text |
| LLMs | Learned from huge amounts of text | Massive data and compute (someone else’s) | Understanding and generating language |