The Berkeley team points out, however, that these approaches are based on prompting the LLM with examples of API calls. By contrast, the Gorilla approach focuses on "systematic evaluation and ...