Matt discusses his results from testing Phind-CodeLlama-34B-v2, which beat the GPT-4 zero-shot HumanEval test by a good margin.