Using large language models (LLMs) for writing code has its pitfalls, such as encountering hallucinations where the model invents methods or libraries that don’t exist. However, these hallucinations are easily caught when the code is run, providing a free form of fact-checking. It is crucial to manually test code as LLM-generated code may look great but could still be incorrect. To reduce hallucinations, try different models, provide more context, and stick to well-established libraries. Reviewing code produced by LLMs can help improve your coding skills. Overall, effective use of LLMs for coding requires effort and practice, but the benefits are worth it.
https://simonwillison.net/2025/Mar/2/hallucinations-in-code/