The realistic path

Most people fail to learn AI not because it is too hard but because they start in the wrong place. Textbook-first learning stalls; tool-first learning stays shallow. A middle path works better.

Stage 1: Use the tools

Spend two or three weeks using ChatGPT, Claude and Gemini for real work — drafts, summaries, plans, code, research. Notice what works, what fails and where you have to reword requests.

Outcome: a working intuition for what generative AI is and how to instruct it.

Stage 2: Learn the vocabulary

Get comfortable with:

  • Model, parameter, token, prompt, context window
  • Training vs inference
  • Supervised, unsupervised, reinforcement learning
  • Overfitting, generalisation, evaluation
  • Bias, hallucination, alignment

You do not need mathematical depth at this stage. You need to be able to read AI content without stopping every sentence.

Stage 3: One small project

Pick something narrow:

  • A script that classifies your emails into two categories
  • A tool that answers questions about a folder of PDFs
  • A chatbot for a small dataset you own

The point is not the artefact. The point is the practice of turning a fuzzy goal into data, code and evaluation.

Stage 4: Foundations, on demand

When a project stalls because you do not understand a concept — probability, gradient descent, embeddings, evaluation metrics — learn that concept then. Learning theory in isolation is slower and less durable.

Stage 5: A structured program

At some point, a structured program moves faster than self-study. It gives sequence, feedback and accountability. The Glintr Artificial Intelligence Program is designed for exactly this stage. For applied ML, see the Machine Learning Program. For a broader overview of computer-science paths, browse the Computer Science category.

Things to avoid

  • Buying five courses at once and finishing none
  • Only watching videos
  • Chasing every new model release
  • Ignoring evaluation — anyone can produce output; producing correct output is the discipline

Frequently asked questions

Do I need Python? Yes, at some point. Start without it, add it in Stage 3.

How long does this take? Three to nine months of consistent effort to reach a usable applied level.

Is it too late? No. The field is early enough that consistent learners catch up quickly.