OpenAI connects GPT-4 and GPT-3.5 chatbots to tools with function calling

OpenAI has proposed function calling, which will allow users to call functions in third-party tools using natural language. That was not possible until now. GPT-3.5 can now also handle more tokens of context for input.

GPT-3.5 and GPT-4 work with arguments within a prompt to call a function, reports OpenAI. A prompt such as ‘send Arnoud an e-mail that he has to write a news item about function calling’ can therefore automatically lead developers to an e-mail program with a name and content pre-filled in it. The function can also work, for example, to request weather reports. The function is not included in ChatGPT, but can be included in chatbots based on GPT-3.5 and GPT-4.

The function works with a json object containing the arguments to the function and the models automatically detect when generating that json is necessary. In addition, GPT-3.5 can now handle contexts of 16,000 tokens, compared to 4,000 in the current version. This will be a more expensive version to use. The current GPT-3.5 with 4000 tokens of context will become slightly cheaper to use.