History & Timeline

Lecture 2 · The key milestones that led to the GenAI revolution

1950s–80s — Perceptrons & Rule-Based AI

McCulloch & Pitts neuron (1943), Rosenblatt perceptron (1958). Expert systems dominated; neural nets mostly abandoned in first "AI winter".

1986 — Backpropagation

Rumelhart, Hinton & Williams popularise backprop, enabling multi-layer networks. Second wind for neural nets.

2014 — GANs

Ian Goodfellow introduces Generative Adversarial Networks — the first practical framework for high-quality image synthesis.

2017 — "Attention Is All You Need"

Vaswani et al. at Google introduce the Transformer. Self-attention replaces recurrence — parallelisable, scalable, and foundational to all modern LLMs.

2018–20 — BERT, GPT-1/2/3

Pre-train on massive corpora, fine-tune on tasks. GPT-3 (175B params) demonstrates in-context learning at scale.

2022 — ChatGPT & Stable Diffusion

RLHF-aligned ChatGPT reaches 100 M users in two months. Open-source diffusion models democratise image generation.

2023–Present — Multimodal & Agents

GPT-4V, Gemini Ultra, Claude 3. Models reason across text, images, code. Autonomous AI agents emerge.