AI Today: From Chatbots to Agents
Lecture 1

The Generative Explosion: How We Got Here

AI Today: From Chatbots to Agents

Transcript

One hundred million users. Two months. That is how fast ChatGPT grew after its launch in late 2022, making it the fastest-growing consumer application ever recorded, according to Reuters. No app in history had scaled that quickly. Not Instagram. Not TikTok. That single data point tells you something profound: people were not just curious about AI. They were hungry for it. The world had been waiting for something it did not know how to ask for, and suddenly it arrived. To understand why that moment hit so hard, Viraj, think of AI's older role as a sorter. For decades, AI was discriminative — it looked at data and made a judgment. Is this email spam? Is this photo a cat? Useful, but invisible. Generative AI flipped the script entirely. Instead of sorting what already exists, it creates something new. Text, images, code, audio — produced on demand, from a simple prompt. That shift from sorting to creating is the central revolution of this era, and it did not happen overnight. It was built on a foundation laid years earlier. That foundation has a name: the transformer. In 2017, Google researchers published a paper called "Attention Is All You Need," and it quietly rewired the future of computing. The key idea is attention — a mechanism that lets a model weigh the relationships between every word in a sentence simultaneously, rather than reading left to right like older systems did. For example, when you read "the bank by the river," your brain instantly knows "bank" means a riverbank, not a financial institution, because you process context all at once. Transformers do something similar, at massive scale. That architecture became the engine under every major language model you have heard of. Training these systems is not cheap. Gartner and IEEE Spectrum have both reported on the staggering infrastructure required — clusters of over twenty-five thousand specialized GPUs, running continuously for months, just to train a single frontier model. The compute alone costs hundreds of millions of dollars. Now, language models are only part of the story. A separate but equally powerful breakthrough gave machines the ability to generate images. Diffusion models work through a fascinating process: they learn by studying how to reverse noise. The model is trained on millions of images that have been progressively scrambled into static. It learns to run that process backward — starting from pure noise and gradually resolving it into a coherent picture. The takeaway is that the machine is not copying an image from memory. It is reconstructing structure from chaos, guided by your text prompt. That is why the outputs can feel almost dreamlike — because in a technical sense, the model is dreaming them into existence. This capability, combined with transformer-based language models, created the multi-modal AI systems that now handle text, images, and code within a single interface. By 2026, Viraj, the landscape has moved again. The progression has been fast and it has been steep. Early chatbots answered questions. Then models started writing essays, generating code, producing art. Now the frontier is agentic AI — systems that do not just respond but act. They can autonomously execute multi-step digital workflows: browsing the web, writing and running code, booking appointments, synthesizing research across dozens of documents. The key idea here is autonomy. These are not tools you use. They are systems you deploy. That distinction matters enormously, for individuals, for businesses, and for society. The transition from discriminative AI to generative AI took decades. The transition from generative to agentic took just a few years. The pace is not slowing down. What started in research labs has become a daily utility, embedded in search engines, productivity software, and creative tools used by hundreds of millions of people. The generative explosion was not a trend. It was a permanent restructuring of how humans and machines interact — and understanding that shift, the transformer architecture that powers it, the scaling that enables it, and the agentic systems now emerging from it, is the essential foundation for making sense of everything happening in AI right now.