Show HN: Catch AI code hallucinations without asking a model (github.com)

A new Python tool, *hedgemony*, has been introduced to detect AI-generated code errors without relying on the language model that produced the code. Instead of asking the model about its outputs, *hedgemony* directly queries the Python interpreter and package registries to verify the validity of packages, methods, and function calls. This innovative approach circumvents the biases and inaccuracies typically associated with language model hallucinations, enabling developers to identify inaccuracies like non-existent packages, invalid function arguments, and inconsistencies in code behavior. This tool is significant for the AI/ML community as it addresses a notable challenge in AI code generation by providing an objective mechanism for validating code correctness. *Hedgemony* categorizes errors into types such as "fabrication," "invention," and "contradiction," emphasizing clarity and verifiable claims over vague terminology like "hallucination" or "lying." With clearly defined criteria to assess code correctness, developers can ensure that their outputs meet intended specifications, ultimately enhancing the reliability and usability of AI-generated code. By offering an easy-to-install package that integrates seamlessly into existing workflows, *hedgemony* empowers users to maintain high coding standards and mitigate errors from automated systems.