The discussion on SOTA (State of the Art) code retrieval with efficient code embedding models raises pertinent points about the advancements in AI-driven code analysis and retrieval. Concerns are voiced regarding the impact of training on synthetic data generated by large language models (LLMs). There's a fear of creating a linguistic feedback loop that may lead to biased language in programming and documentation texts, as future models could reinforce certain lexicons over others. This pattern may perpetuate unintentional bias in the types of language used, risking innovation and clarity in code-related communication.