Microsoft’s Guidance Project Aims to Streamline AI Language Models
Language models, such as those used in AI, often have a tendency to interpret user prompts relatively loosely, which can lead to inefficient results when trying to capture them with longer prompts. However, Microsoft’s Guidance Project aims to improve the efficiency and accuracy of AI language models by tailoring them more effectively to user requirements.
Guidance is a separate AI template language that has its own syntax, which outlines specific rules for the models that are primarily intended for development teams that rely on reliable results when using AI. With Guidance, developers can easily write programs that implement text generation, prompting, and the plausibility check of results in one streamlined step. External tools can also be integrated for further precision.
Unlike other normal template languages, Guidance’s syntax is order-specific, which aligns with the token processing order of the AI models. Using the {{gen}} command, developers can generate AI responses and output them at specific points while initiating the next logical steps. OpenAI and Hugging Face’s language models, along with Facebook’s leaked open source model Llama, can work with Guidance, making this project universally applicable.
Microsoft Guidance, along with suitable Jupyter Notebooks for developers to try out the functions themselves, is readily available in a Github repository. This project aims to streamline AI language models, making them more accurate, user-friendly, and efficient, while providing a standardized solution for reproducible outputs.
