Microsoft has developed a large language model Gorilla, which is better than GPT-4 in API calls

According to a report from IT House on August 12, Microsoft Research recently joined forces with the University of California, Berkeley to develop a new large language model (LLM) called Gorilla, which is superior to GPT-4 in terms of accuracy and flexibility in API calls. . Gorilla LLM is said to use a technique called "retriever-aware training" to allow LLM to learn the semantics of API calls and generate semantically and grammatically correct text content.

Microsoft Research said that even if the API documentation changes, the correct semantics and syntax can still be generated when calling the API. Although GPT-4 is the most powerful LLM developed by OpenAI, it can control various scenarios such as text generation, translation language, and answer content. However, the API call is its shortcoming. It cannot understand the semantics of API calls and can only generate and API documents. similar text.

In a recent study, Gorilla outperformed GPT-4 on various API calling tasks. GPT-4 has an accuracy rate of 85% in API calls, while Gorilla's accuracy rate can reach 95%. In addition, Gorilla also has the skill that GPT-4 does not have, which can generate correct API calls for APIs without training data sets.