closed-source-models-used-in-experiments

Summary

The experiments under study rely on three specific commercial language models — an early ChatGPT, a first-generation GPT-4, and Google's PaLM 2 chatbot — rather than open-weight alternatives. This matters because every downstream finding in the system is scoped to these particular model versions, meaning results may not generalize to newer or open models that were not part of the test set.