Update January 23, 2024: At Microsoft, we are always updating and testing features to offer our users the best possible experiences as we experiment with new approaches to functionality. To improve the user experience and streamline our tools that empower creativity, Bing Image Creator is now Designer and Bing Chat is now Copilot. Create wow-worthy images with your words and AI with Designer, and try Copilot, your AI-powered search assistant for the web.
In the fast-evolving digital era, Copilot stands as a testament to Microsoft’s prowess at harnessing the potential of artificial intelligence. To truly appreciate its capabilities, it’s vital to understand the foundational elements such as the large language model, or LLM, AI. Get to know the tech behind Copilot’s conversational, generative expertise and learn about key features you can enjoy.
LLM AI: The brain behind Copilot
The term large language model might sound intricate, but its essence is simple. Picture a colossal digital library brimming with billions of words, ideas, and contexts. Now, imagine having a super librarian who can fetch the most fitting piece of information almost instantly. That’s how LLM AI works.
AI types and categories
Artificial Intelligence can be divided into different types based on its capabilities and functionalities. How does generative AI work, and what are other types of AI? Let’s delve into that.
Narrow AI vs. general AI
Narrow AI, also known as weak AI, is designed to perform specific tasks or functions. It excels in a particular area but lacks the ability to generalize beyond its specialized domain. Conversely, General AI, or strong AI, aims to replicate human-level intelligence, capable of understanding, learning, and performing any intellectual task a human can. Understanding how does AI work in these contexts is crucial.
Supervised, unsupervised, and reinforcement learning
AI systems learn and improve their performance through various learning methods. Supervised learning uses labeled data for predictions or classifications. Unsupervised learning identifies patterns and relationships within unlabeled data. Reinforcement learning trains an AI model through trial and error, learning from feedback and rewards.
Natural language processing (NLP)
At its heart, AI like Copilot operate on neural networks that mimic the human brain’s structure. These networks allow AI to learn from vast volumes of data. Based on each person’s interactions and new online content, this continuous learning ensures that the platform becomes smarter and more efficient over time—understanding, interpreting, and generating human language. That means that Copilot’s functionality increases every time you use it!
How does AI work in crafting responses?
When you prompt Copilot with a question or statement, it performs the following actions to give you a response:
- Interpret the input. The LLM AI analyses your words, understanding context and nuances.
- Search the knowledge base. Copilot searches for relevant patterns, data, or context.
- Craft the answer. The AI crafts a contextually fitting, coherent reply, ensuring clarity and relevance.
- Deliver the response. Copilot offers you a well-articulated answer almost instantly.
The AI powering Copilot: OpenAI’s ChatGPT-4, Prometheus, and DALL-E 3
Copilot is powered by a one-of-a-kind AI trio. Each brings unique functionality to Copilot’s conversational, generative, AI-powered searches:
- OpenAI’s ChatGPT-4. This model of Open AI’s ChatGPT LLM is customized for search, optimized for speed, and equipped to generate creative text in any form you can imagine.
- Microsoft’s Prometheus model. This is Microsoft’s proprietary model for working with the OpenAI LLM. It’s specially designed to increase the timeliness, relevance, and safety of your AI-powered search responses.
- DALL-E 3. This is OpenAI’s generative, text-to-image neural network. DALL-E 3 goes beyond the language-bound LLMs and takes AI-generated content into the visual realm.
