LLM Performance for Code Generation on Noisy Tasks