On December 20, OpenAI’s 12-day press conference has entered the 11th day, which is the penultimate day. , the company released MacOS desktop applications and their interoperability features with various applications. This will lay the foundation for the future of "agent-based artificial intelligence" (Agentic AI, or agents), making ChatGPT more powerful and seamlessly integrated into users' daily workflows.

At the 11th day of the press conference, OpenAI Chief Product Officer Kevin Weil joined two colleagues wearing Christmas-themed suits to discuss the company’s latest MacOS desktop application program. They highlighted the transformation of ChatGPT from a simple conversational assistant to a more powerful agent tool, which means that ChatGPT can now perform more tasks on behalf of users, bringing users an unprecedented convenient experience.

Currently, users can view and automate their ChatGPT work through the MacOS desktop application. Although more similar versions will come out in 2025, before that, OpenAI has taken the lead in launching the following three major features:

First, with the "Work with Apps" function, users can now use ChatGPT Access more coding applications, including BBEdit, MatLab, Nova, Script Editor, TextMate, Android Studio, AppCode, CLion, DataGrip, GoLand, IntelliJ IDEA, PHPStorm, PyCharm, RubyMine, RustRover, WebStorm, Prompt and Warp, etc.

In a demo of the MacOS desktop app, OpenAI showed how artificial intelligence can dig deep into an application to obtain and understand its contextual information. Once the user selects an application through the "Work with Apps" function, ChatGPT can immediately access it, gain insight into the application, and provide immediate help.

Of course, ChatGPT is more than just a simple viewing tool. It relies on a powerful artificial intelligence model to perform a variety of functions. In Warp's demonstration, ChatGPT can not only capture the content on the user's screen, but also drill down into the application to browse more information. For example, when processing long strings of code, ChatGPT can achieve scroll-free browsing, which greatly improves work efficiency.

Compared with the Windows Recall function, ChatGPT focuses more on real-time collaboration with applications, rather than just recording and building a memory library. In another demo, the OpenAI team put ChatGPT is tightly integrated with XCode, allowing it to work within Apple's development applications. Users simply make a request and ChatGPT can generate code or solve programming problems.

It is worth noting that OpenAI also demonstrated a new skill of ChatGPT: it can embed the generated code directly into XCode, a feature that is expected to greatly simplify the workflow. Although the ChatGPT code failed twice during the live demonstration, the OpenAI team successfully got the code running on the third try.

Second, for users who use ChatGPT to write, OpenAI announced that the MacOS desktop application now supports Apple Notes, Quip and Notion. During a live demo, the OpenAI team was browsing a document designed to develop guidelines for hiking activities in Notion.

With this new feature, ChatGPT can work seamlessly with Notion. Live demonstrations focus on specific passages of text in the document and are tasked with "supplementing these conversation points." In addition, users can also use ChatGPT's search function to generate responses. For example, in the demo, it generated conversation points about "Emperor Norton (Norton I)" based on the selected text, with quotes and sources.

Third, in addition to the traditional operations of text selection and copy and paste, the MacOS desktop application supports advanced voice mode and can work together with other applications. In this mode, users can set a "holiday party playlist" in Apple Notes and ask Santa Claus for his opinions on candidate songs through ChatGPT. ChatGPT can even point out user errors, such as miswriting the Christmas song "Frosty the Snowman" as "Freezy the Snowman."

These features are now officially released, users just need to make sure they have the latest version of the MacOS application and subscribe to any one of ChatGPT Plus, ChatGPT Pro, ChatGPT Team, ChatGPT Enterprise or ChatGPT Edu, that is Available for immediate experience.

In terms of privacy protection, OpenAI particularly emphasizes that ChatGPT will only interact with the application when manually triggered by the user. Once the feature is activated, users will know exactly what content will be attached to messages, effectively alleviating privacy concerns.

Since December 5th, local time in the United States, OpenAI has started an intensive new feature release cycle, planning to New products and features will be launched through 12 live broadcast events in the next 12 days. Prior to this, OpenAIIt has successively released a number of innovations, including ChatGPT Pro plan, enhanced fine-tuning technology, Sora, interactive interface Canvas, advanced voice visual functions, Projects function, ChatGPT search, full-blooded version of o1 model, and opening large models to third-party developers through API o1 series and interact with ChatGPT via phone and WhatsApp, etc.

As the press conference comes to an end, people are paying more and more attention to AGI (artificial general intelligence). OpenAI said at the end of the 11th day conference: "On the 12th day, we have prepared extremely special content, don't miss it!"

In the corner of the demo screen, you can see a message titled " AGI_Interface.swift" folder. This is not the first such surprise in the past 12 days. A few days ago, OpenAI also unveiled a calendar event Easter egg called "Super Secret AGI", which undoubtedly further raised people's expectations for this 12-day series of announcements. Everyone speculated whether these announcements jointly painted a road to general purpose. The big picture of intelligence.

OpenAI also revealed that a Windows application for ChatGPT will also be released soon. But even more shocking news is that they confirmed the existence of a new agent, which is expected to be released in 2025. OpenAI said: "As our models become more powerful, ChatGPT will demonstrate increasing autonomy."

A few weeks ago, there were rumors that OpenAI was developing a software called “Operator” agent-based artificial intelligence, a plan the company only confirmed at its Day 11 launch event. Perhaps there is pressure from competitors behind this move.

Recently, Google announced Project Mariner, an agent that can navigate and perform operations on web browser tabs on behalf of users. Likewise, Microsoft has launched Copilot Vision, a feature that views content and provides relevant information in users' web browsers. Of course, Anthropic released the Computer Use feature earlier, which is ahead of other similar tools in time.

Now, there is only the last day of OpenAI’s 12-day series of events, and they seem to have saved the best part for the end - a new and powerful cutting-edge model is about to be unveiled. We will wait and see to see what exactly OpenAI brings to the table and how this new model differs from the previous o1 model.

It is worth mentioning that some benchmark tests have shown that the o1 model is one of the most powerful artificial intelligence models to date, even surpassing Claude 3.5 in coding tasks. Recently, a user of the X platform allegedly discovered the GPT-4.5 model, although the model currently only provides limited preview capabilities.

Now, all eyes are focused on OpenAI, and everyone is eagerly waiting to see what surprises they will bring on the last day of the press conference.