Keynote / Experiments

It's that time of year again—keynote season. We're about to see a firehose of new features presented as perfectly executed, inevitable advancements.

This is the polar opposite of how progress in AI research actually happens. A keynote is a highlight reel of the 1% of ideas that worked. The lab, on the other hand, is a graveyard for the 99% of experiments that went nowhere.

The real skill isn't just having the one idea that makes it to the stage. It's developing the intuition, from thousands of failed runs, to know which ideas to kill early. That accumulated wisdom is the most valuable asset a research team has, but it never gets a slide in the presentation. A keynote is a map of the destination; real research is the messy, handwritten journal of the journey.