Learn AI & Machine Learning with any AI

AI is broad, but the path through it is fairly linear. Roughly in the order it makes sense to learn:

• Foundations & history. From Turing's question "can machines think?" through symbolic AI, the "AI winters," and the deep-learning revolution that changed everything.
• Machine learning. The core idea and its three modes — supervised (learning from labelled examples), unsupervised (finding hidden structure), and reinforcement learning (learning from reward).
• Neural networks & deep learning. How these models actually learn: layers, weights, and the gradient descent that tunes millions of parameters.
• How LLMs work. Transformers, attention, and next-token prediction — plus the pipeline (pre-training, fine-tuning, and learning from human feedback) that turns raw text prediction into an assistant.
• Generative AI. The same underlying ideas producing text, images, audio, video and code.
• Agents & tool use. Models that don't just answer but act — calling tools, browsing, and chaining steps toward a goal.
• Alignment, safety & interpretability. Making powerful systems reliably do what we intend — and understanding what's going on inside them at all.
• Governance & the economy. Regulation, the geopolitics of compute, and what automation does to work.
• Prompt & context engineering. The practical craft of getting the best from these systems — the skill this whole site is, quietly, teaching.

A field only three-quarters of a century old already has landmarks. These are the ones worth knowing — real people and real papers. (For anything past a model's training cut-off, reach for the Frontier prompt.)

• Alan Turing — his 1950 paper Computing Machinery and Intelligence posed the question and proposed the "imitation game" we now call the Turing test.
• The Dartmouth workshop (1956) — where the term "artificial intelligence" was coined and the field was founded.
• The perceptron — Frank Rosenblatt's early neural network, and the decades-long argument about what such models could and couldn't do.
• Backpropagation (1986) — Rumelhart, Hinton and Williams popularised the algorithm that makes training deep networks practical.
• AlexNet (2012) — Krizhevsky, Sutskever and Hinton's deep network crushed an image-recognition benchmark and kicked off the modern deep-learning era.
• "Attention Is All You Need" (2017) — Vaswani and colleagues introduced the transformer, the architecture underneath essentially every large language model since.
• The "godfathers of deep learning" — Geoffrey Hinton, Yann LeCun and Yoshua Bengio, whose long-unfashionable bet on neural networks set the field's direction.
• The scaling era — the run of ever-larger models that showed capability climbing with size and data, and set off the debates in the next section.

This is the most contested subject in the whole library, and a good tutor holds the disagreement open rather than quietly picking a side. The genuine fault lines:

• Is scaling enough for AGI? One camp holds that making models bigger keeps making them smarter, all the way up; another says something fundamental is still missing. Both have serious people and serious arguments.
• How hard is alignment? Can we reliably control systems that may become more capable than us — and would we even know if we'd failed? Views run from "manageable engineering problem" to "defining challenge of the century."
• Do these models "understand"? Are LLMs reasoning, or — in the famous phrase — stochastic parrots recombining patterns? Searle's old Chinese Room argument is suddenly very current.
• Open or closed? Should the most powerful models be released openly for scrutiny and access, or kept restricted for safety? Reasonable people disagree sharply.
• What happens to work? Whether AI mainly augments human workers or mainly replaces them — and what either would mean for how societies share out prosperity.

A sensible path, all driven from the panel above:

1. Run Orientation on "how large language models work" — the single highest-leverage thing to understand right now.
2. Turn on your assistant's web search and run The Frontier, so you're working from what's true this month, not just the training data.
3. Use Great Debates on the AGI or alignment question to get both sides at full strength.
4. Then get practical with the prompt & context angle — you're already doing it; the Method page and this site are a live worked example.

And do the thing only this subject allows: ask the tutor to explain a concept, then ask it to show how it does that very thing.