Trying to make sense of how we got here with AI

Something that used to bug me is how AI gets talked about like it just showed up in the last few years. ChatGPT launches, everyone loses their minds, and suddenly "AI" is a household word. But the field is genuinely decades old, and I wanted to actually understand the path that led here instead of just experiencing the recent explosion as if it came out of nowhere.

It started way earlier than people think

The term "artificial intelligence" itself goes back to the 1950s. Early researchers were wildly optimistic — some genuinely believed human-level machine intelligence was maybe 10-20 years away. That optimism ran into a wall pretty quickly, and the field went through what's now called an "AI winter" — funding dried up, progress stalled, and a lot of the early promises just didn't pan out with the compute and data available at the time.

This happened more than once, actually. There's a rough pattern in AI history of hype cycles followed by disappointment followed by quiet progress in the background.

Then the quiet part started mattering

While AI as a broad idea kept struggling to deliver, a narrower thing was steadily improving in the background: machine learning, and specifically neural networks. For a long time neural nets were seen as a bit of a dead end compared to other approaches. What changed that wasn't some single breakthrough insight — it was mostly three things lining up at once:

• Way more data became available (thanks, internet).
• Way more compute became available and affordable (thanks, GPUs — originally built for video games, of all things).
• A handful of real algorithmic improvements that made deep networks actually trainable at scale.

None of these alone would've done it. It's the combination that mattered.

The transformer moment

The next big turning point I keep coming back to is the 2017 "Attention Is All You Need" paper, which introduced the transformer architecture. This is the thing basically every modern LLM is built on. Before this, models processing language mostly worked sequentially, word by word, which was slow and struggled with long-range context. Transformers let models look at an entire sequence at once and figure out which parts of it mattered most to each other. That architectural shift is a big part of why progress accelerated so fast afterward.

Why it suddenly felt so fast to everyone else

From the outside, it looked like AI went from "not very useful" to "genuinely useful" almost overnight around 2022-2023. But from what I can tell, that wasn't a sudden leap — it was more like a long, slow climb that finally crossed a threshold where the output became good enough for regular people to actually notice and use in everyday life. The underlying research had been building for years before that.

Where I've landed, for now

The way I think about it now: AI didn't "arrive" recently, it just crossed a visibility threshold recently. The actual progress has been continuous, uneven, full of dead ends, for over 70 years. That reframe made the current moment feel a lot less like magic and a lot more like the result of a very long, very human process of trial and error.

I'm curious where the next "quiet decade" is happening right now, the one nobody's paying attention to yet that'll look obvious in hindsight in ten years.