Why?

There are very few paper that have moved me as much as AlphaEvolve. The paper improved on the RL pilled frontier of the current AI research landscape, and proved - by discovering novel algorithms, just how much we can still squeeze performance out of LLMs by improving solutions to hard problems and evolving them to discover new solutions. Here's what @karpathy has to say about it, and is more or less what AlphaEvolve does:

Scaling up RL is all the rage right now. I'm fairly certain RL will continue to yield more intermediate gains, but I also don't expect it to be the full story. ... There's significantly more bits of supervision we extract per rollout via a review/reflect stage of human mechanism of improvement along the lines of "what went well? what didn't go so well? what should I try next time?" etc. and the lessons from this stage feel explicit, like a new string to be added to the system prompt for the future, optionally to be distilled into weights (/intuition) later a bit like sleep. ... Example algorithm: given a task, do a few rollouts, stuff them all into one context window (along with the reward in each case), use a meta-prompt to review/reflect on what went well or not to obtain string "lesson", to be added to system prompt (or more generally modify the current lessons database). Many blanks to fill in, many tweaks possible, not obvious.

"Cadence" is an attempt to understand evolutionary problem solving by studying how solutions evolve over generations of improvement, and whether or not it could lead to newer, improved solutions for NP Hard problems. For the purpose of this blog, I chose the Travelling Salesman Problem(TSP), and analyzed how the solutions evolves. TSP is a classical optimization problem, with approximate solutions traditionally requiring exponential time to solve.