Eat. Sleep. AI!

Why now

Every generation of researchers gets one tool that quietly resets the floor. The microscope did it. The spreadsheet did it. Search engines did it. For your generation it is AI, and it arrived faster than any of them — going from a lab curiosity to something running in the browser tab next to your homework in about three years. That speed matters, because a skill adopted this quickly stops being a specialty and becomes a baseline. A decade ago, knowing how to search well was what separated a decent student project from a memorable one. Today the separation is whether you can move from a question to a working answer while everyone else is still opening tabs. Learning this at fifteen instead of twenty-five is not a ten-year head start — it is a head start on every single project you take on in between.

What it changes

For a researcher specifically, AI collapses the distance between having an idea and testing it. You can absorb a stack of background papers in an evening, write the analysis script you did not know how to write, label a dataset that would have eaten a whole semester, and build a prototype good enough to show someone by Friday. But look closely at what does not change. Someone still has to decide which question is worth asking, whether the output is actually true, and what the result means for the next experiment. The model is fast, confident, and sometimes completely wrong. So the real skill was never prompting — it is judgment: knowing what to ask for, knowing how to check it, and knowing when to throw it out and start over. That judgment is exactly what a research program like Rainflower exists to build in you.

Demo — Wiring an AI model to a live camera feed so it reacts to what it sees, frame by frame. No company product, no special hardware — one afternoon and a few false starts.

How to start

Build something. Reading about AI teaches you roughly as much as reading about swimming teaches you to swim, which is why the clip above is a demo instead of a diagram — that is a model watching a live camera and responding in real time, put together out of curiosity rather than expertise. You can do the same thing this week. Pick one small annoyance inside a project you already care about and point a model at it: the graph you keep redrawing, the notes you keep re-reading, the code that will not run. Get it wrong, figure out why, fix it, and then do it again next week with something slightly harder. That loop is the entire method, and it is why the title of this post is a joke that is not really a joke. Eat, sleep, AI, repeat.