On December 5th, Microsoft officially celebrated the one-year anniversary of Copilot and announced several new upcoming features.
Generated by unbounded AI.
As early as a few months ago, Microsoft had incorporated AI-assisted features from its extensive portfolio into Copilot, including the GPT-4 model. With OpenAI's release of an upgraded version at its inaugural developer conference, Copilot will also generate responses using the latest GPT-4 Turbo model to enable users to tackle more complex and longer tasks, such as writing**, etc.
Microsoft said the model is currently in open beta with select users and will be widely integrated into Copilot in the coming weeks. In addition to this, Copilot will also bring several major updates.
In addition to the GPT-4 Turbo, Copilot's text-to-image feature will also be upgraded to the new DALL-E 3 model, allowing users to create higher-quality, more accurate images based on prompts. Tested by Slashgear, the image visuals have improved significantly, and its understanding of prompts has become much better. The following diagram shows a set of comparisons presented by Microsoft:
You can do this by visiting Bingcom create or ask copilot to create an image to access this feature.
Another interesting feature coming soon from Microsoft is Multimodal Search Fundamentals, which combines GPT-4 capabilities, visual, Bing image search, and web search data to provide better image understanding.
According to the case showcase, Microsoft said that traditional multimodal systems can only describe what is in an image in an ordinary way, but combined with a search basis, users are able to accurately identify the space shuttle and its launch date.
As mentioned above, Microsoft is also working on a new interpreter. With this feature, users can perform complex tasks such as accurate calculations, coding, data analysis, visualization, math, and other complex tasks.
According to reports, Copilot's Python environment runs in a secure sandbox environment on top of Azure container applications, and the sandbox is pre-installed with many popular data science tools and libraries, such as pandas, numpy, matplotlib, sklearn, flask, etc., to solve complex problems.
During use, Copilot will write to answer the user's complex natural language requests, and will allow the user to upload and file files from Copilot to use their data and Bing search results. Microsoft plans to make it widely available in the near future.
Currently, users can also summarize or ask questions about what's being done in Edge. As you can see in the image below, if you're at Satya (Microsoft CEO)'s latest Ignite keynote, you can ask Copilot to wrap it up for you.
The biggest upgrade in the space is undoubtedly deep search when it comes to Bing search. In short, if AI is destined to reimagine the web search experience and make it more valuable, deep search is the answer.
This feature leverages the power of GPT-4 to provide optimized search results for complex topics, breaking down a simple search query into multiple directions and providing relevant answers for each direction, eliminating the hassle of subsequent search steps.
So, how can deep search help?
According to Microsoft, deep search is not a replacement for traditional web search. Rather, it is a complementary system that tries to ask questions related to the user's initial query and proactively find answers.
For example, if you're searching for something like "how to find good coffee shops in Los Angeles," a deep search will split your query into an intelligent, multi-step journey and find answers to other related questions you might ask, such as:
Top coffee shops in Los Angeles.
The closest coffee shop to your current location.
Reviews and ratings for Los Angeles coffee shops.
Ranking of the average coffee shop spend in the area**.
Opening and closing times of different coffee shops.
Related coffee shop *** and menu.
Microsoft says that deep search is about reading search intent. Once it outlines all the expected search queries related to your primary question, it lists the best answers from the resource that don't appear when you perform a regular Bing search query.
The company emphasizes that deep search is 10 times deeper than regular Bing search and can analyze tens of millions of web pages. Not only do the results produced by this method be more detailed than those found through standard search rankings, but they are also more precise. Once relevant search results are gathered, it conducts topic matching to provide the most appropriate answers. At the same time, it also considers factors such as trust and authority, the freshness of the material, and whether it provides the right amount of depth and detail.
At the heart of deep search is OpenAI's GPT-4 language model. When a user enters a single-sentence search query, the AI model leverages its generative capabilities and expands the search query by adding multiple contextual extensions to it. The easiest way to understand this behavior is to assume that if your search contains a "why", deep search will add "how", "when", and "where" sections to it in an extended form.
"Deep search leverages GPT-4 to find all possible intents and compute a comprehensive description for each intent," Microsoft explains. ”However, this is at best context-aware speculation, and even Microsoft admits that the extended work of deep search can be problematic from time to time. To ensure that users are guided in the right direction and are not overwhelmed by useless answers, the deep search section will provide a disambiguation panel where users can select the right extension for their search query, while the remaining AI-generated search context will be removed.
It's a complex, multi-step background process. As a result, deep search results take more time to compile and render than regular Bing searches.
Microsoft noted in the blog post that depending on the complexity of the query, a deep search can take up to half a minute to show results. However, users don't use deep search for every web search, especially if you want a straightforward answer. So, think of deep search as a chatbot, such as ChatGPT or Bing Chat, but it's essentially about internet searching and trying to find answers before you ask more questions.
Although Copilot is now publicly available, deep search is still largely an experimental feature. As a result, Microsoft took a cautious approach at the time of rollout to ensure that all issues were addressed at an early stage before a broader rollout.
Currently, the company notes that deep search is only "available to a small group of randomly selected Bing users worldwide." At the same time, Microsoft would not disclose whether it would prioritize specific users or paying users when accessing deep search. But considering that it's largely a Bing search technique, the best option is to access it before anyone else uses the Edge browser.
When it's available, you'll see a new deep search button next to the text search box. Until then, you can turn to Microsoft's web browser, which currently has a dedicated copilot: in the top right corner of the screen
Clicking on it will open the "Copilot with Bing Chat" window, where you can enter your query;
It will take the answers from the web in the form of paragraphs with ** citations;
Once you've provided an answer, the interface displays a set of three related queries based on your original search.
You can simply click on these preset queries to get more relevant answers.
So, which of the new features of Copilot mentioned above would you like to experience?
Reference Links: