Experimenting with creating my own LLM coding benchmark
The first test-run was already very interesting, but the tests need some improvements to make the results more meaningful. Later I plan to run it with different sub-agent combos. Maybe I'll also add the results to my website 🤔 #BuildInPublic