C2LLM is a new family of code embedding models that helps computers find the right code faster and more accurately.
The paper introduces UCoder, a way to teach a code-generating AI to get better without using any outside datasets, not even unlabeled code.