A hallucination is fluent or confident AI output that is factually wrong, nonsensical, or unsupported. Coding models can produce code that looks correct while hiding serious problems. section 1.5 section 1.5.2
LLMs predict likely words and code patterns rather than retrieving answers from a truth database. Faulty training data, tricky prompts, or misleading learned patterns can therefore produce confident errors. section 1.5 section 1.5.2
Watch for hallucinations when generated Python code:
- Uses a familiar but suboptimal pattern, such as a list where a set would be more efficient. section 1.5.2
- Chooses a popular library unnecessarily or works only for common cases while failing on edge cases. section 1.5.2
- Repeats outdated Python practices from codebases written before newer language features. section 1.5.2
Review generated Python
A generated solution may look polished while selecting a list instead of a set, ignoring an edge case, or following an outdated Python pattern. Check the implementation against Python best practices and project architecture before using it. section 1.5.2
