HomeWinBuzzer NewsMicrosoft Research Gorilla LLM Beats GPT-4 in API Call Accuracy and Flexibility

Microsoft Research Gorilla LLM Beats GPT-4 in API Call Accuracy and Flexibility

Microsoft Research has developed a new Gorilla large language model (LLM) called Gorilla that outperforms GPT-4 in API call generation.


Researchers at UC Berkeley and Microsoft Research have developed a new large language model (LLM) called Gorilla that outperforms GPT-4 in making API calls. Gorilla is able to generate semantically and syntactically correct API calls, even when the API documentation changes.

GPT-4 is a powerful LLM from OpenAI that can generate text, translate languages, and answer questions. However, it has been shown to struggle with API calls. This is because GPT-4 is not able to understand the semantics of API calls. It can only generate text that is similar to the text in the API documentation.

Gorilla addresses this problem by using a technique called “retriever-aware training.” Retriever-aware training involves training the LLM on a dataset of API calls and their documentation. This allows the LLM to learn the semantics of API calls and to generate text that is semantically and syntactically correct.

In a recent study, Gorilla was shown to outperform GPT-4 on a variety of API call tasks. Gorilla was able to generate correct API calls with 95% accuracy, while GPT-4 was only able to generate correct API calls with 85% accuracy. Gorilla was also able to generate API calls for APIs that were not in its training dataset, while GPT-4 was not able to do this.

Building Accuracy through Different Training Methods

The key difference between Gorilla and GPT-4 is that Gorilla is trained on a large corpus of code snippets and natural language descriptions, while GPT-4 is trained on a general text corpus. This allows Gorilla to learn the syntax and semantics of different programming languages and APIs, as well as the common patterns and best practices of coding. Furthermore, Gorilla uses a novel attention mechanism that can capture the long-term dependencies and structural information in the code generation process.

Here are some of the key benefits of using Gorilla:

  • Accuracy: Gorilla is able to generate correct API calls with 95% accuracy. This is significantly higher than the accuracy of GPT-4, which is only 85%.
  • Flexibility: Gorilla can generate API calls for APIs that are not in its training dataset. This makes it a valuable tool for developers who need to work with new APIs.
  • Ease of use: Gorilla is easy to use. Developers can simply provide Gorilla with a natural language description of the API call they want to generate, and Gorilla will generate the correct API call.

Microsoft and OpenAI’s Relationship Means Collaboration

Of course, Microsoft is the biggest investor in ChatGPT developer OpenAI and is reportedly holding 49% of the AI leader. Microsoft’s multiple billions invested into OpenAI allow the company to access LLMs and other solutions. Since the start of the year, Microsoft has been mainstreaming AI services across its ecosystem by combining its own models with technology from OpenAI. GPT-4 has been at the heart of that evolution. Microsoft has integrated OpenAI’s generative AI in the following ways:

With such close collaboration, Microsoft’s own extensive research into large language models is unlikely to hamper its partnership with OpenAI. In fact, it is more likely that OpenAI will become a collaborator on such projects, or at least that Microsoft will use AI platforms like Gorilla alongside integrations from OpenAI.

Luke Jones
Luke Jones
Luke has been writing about all things tech for more than five years. He is following Microsoft closely to bring you the latest news about Windows, Office, Azure, Skype, HoloLens and all the rest of their products.