How do LLMs bring victory? - AI as a Personal Racing Coach.
I ride my bike on Zwift every day. It is its own form of racing, so strategy and execution have a big impact on results depending on the course and the characteristics of the riders in the field.
In the past, I built models based on riders' absolute power, weight, and FTP to predict race outcomes. Even that was useful enough to judge which race made sense for me on a given day depending on my condition.
Then I added an LLM on top. Instead of merely showing numbers, I used it for course analysis and strategy building as well. The effect was huge. Sometimes it pointed out angles I had not thought of. Even when it only reorganized things I already knew, that helped me focus on execution without wasting mental energy.
In the race from the screenshots, Zwift ranking points would have put me around fifth, but I never overpaced in the middle and kept riding my own game all the way to the end, then won it with a bunch sprint. I knew that was how I win, and yet so many times I had been dragged into the tactical games of riders whose strengths were different from mine.
Using AI like this, trying to live every day as the best version of myself, has become one of the big joys of my life lately.
(As an aside, if you use Upstage's Solar Pro 2 model with reasoningEffort set to high, I find the price-performance and speed for this kind of conversational analysis to be quite solid.)
