AI Fundamentals· Lesson 2 of 10
- 1What Even Is AI? AI vs. ML vs. Deep Learning6 min
- 2How Models Learn: Training vs. Inference6 min
- 3What Is an LLM?8 min
- 4Tokens — the Currency of Everything7 min
- 5Why AI Output Is Probabilistic (and Why It Changes How You Build)7 min
- 6Hallucination — Why Models Make Things Up7 min
- 7What AI Is Genuinely Good and Bad At — a Capability Map7 min
- 8AI Systems vs. Traditional Software — the Mindset Shift7 min
- 9Where AI Fits in a Product (and Where It Shouldn't)7 min
- 10Responsible AI Basics for Builders7 min
How Models Learn: Training vs. Inference
In one line · Training teaches the model once (expensive); inference is using it afterward (cheap and fast).
As an AI engineer you live in inference: using a ready model via API. Training is done by teams with big GPUs and huge data.
Knowing you only do inference stops you chasing 'how to train a model' and focuses you on what matters: your prompt and the data you feed in.
Training is like years of study; inference is asking one exam question and getting an answer. You only ask.
Like a chef who trained for years (training); when you order, they cook in minutes (inference) — not relearning each time.
GPT was trained once on trillions of words. Each message you send is inference — costing cents, not millions.
In training the model sees examples and adjusts its parameters to reduce error; then the parameters freeze. In inference we push input through that fixed network and read the output.
An OpenAI/Anthropic API call is inference on a pre-trained model; your message doesn't change it. Customizing it is fine-tuning (an advanced topic).
🧪 Try it: hit any model's API with the same question twice. It doesn't learn from the first — each call is independent inference.
Beginners assume they must 'train a model' to build an AI product. No — 99% of products just use a ready model via inference.
Inference is what you'll use in nearly every project: API + prompt + your data.
Training/fine-tuning only makes sense with a very specific case and lots of data — rare early on.