Anders Hejlsberg, the father of C# and TypeScript, today announced a new open source project - TypeChat , which uses AI to build a "bridge" between natural language and application schema (application schema), and API.
Simply put, TypeChat is a tool library that can use "human language" to build type-safe UI .
In the past few months, we have witnessed a wave of discussions around large language models.
Many people communicate with large models in a dialogue way, which is also the most direct way. But this approach faces a problem: how to integrate these models into existing APIs in the best way.
In other words, how can traditional UIs be enhanced with natural language interfaces? How can AI be used to take user requests and turn them into something an application can run ? How do you ensure that applications are secure and are doing work that both developers and users can trust?
The TypeChat library attempts to address these issues by using type definitions in the codebase to ensure that structured AI responses are type-safe.
Example of use
The user asked to convert the phrase "Could I get a blueberry muffin and a grande latte?" into JSON and replied in JSON format only:
The above example does not specify the type. When the type is specified, TypeChat's reply is as follows:
TypeChat supports usage in many different ways. The way Microsoft is currently discussing is mainly to use data patterns to convert user intent into structured responses. In fact, TypeChat can also use API patterns to build basic programs.