OpenAI Releases AI Browser ChatGPT Atlas, Challenging Chrome Supremacy

October 21, local time.OpenAIChatGPT Atlas, an independent AI-powered web browser, was officially launched via live streaming, which is not only OpenAI's first foray into the field of Internet terminal tools and a landmark product for laying out the core entrance of users“ access to the Internet, but also regarded by the industry as a direct challenge to Google Chrome's dominance as a browser for more than a decade and to the traditional web model of ”clicking on links to find answers". "It is also regarded by the industry as a direct challenge to Google Chrome's dominance of the browser for more than 10 years, as well as the traditional web model of "clicking on a link to find an answer.

“The browser is a central scenario in modern work and life, and AI gives us a once-in-a-decade opportunity to redefine what it means to ‘use the browser,”" OpenAI CEO Sam Altman said bluntly at the the launch event bluntly stated.

He said the ultimate goal of ChatGPT Atlas is to end repetitive and tedious “digital labor” on the Internet, allowing users to move from “manually manipulating web pages” to “telling AI what to do”. The goal is to end repetitive and tedious “digital labor” on the Internet, so that users can move from "manually operating web pages" to "telling AI what to do", and to drive the network into a new phase of "intent-driven". He mentioned that in the future, using the Internet may be as simple as chatting with a browser.

01.Core breakthrough: “Intelligent Body Mode” allows the browser to change from “see” to “do”.”

Unlike traditional browsers that are only positioned as “content display windows”, the core competency of ChatGPT Atlas is “Agent Mode”. It is an AI intelligence that can accomplish complex tasks on web pages on behalf of the user, directly upgrading “browsing” to “execution”.

The demonstration at the launch event visualized the value of this feature. After the user gives the command “Prepare a beach picnic”, the AI smart body can automatically recognize the list of ingredients in the recipe page, jump to Instacart and add sunscreen, picnic mats and other items to the shopping cart; in the workplace, it can read the project plan from Google Docs and automatically synchronize it to the Linear task management tool. Linear task management tool, the whole process does not require users to copy and paste or switch operations.

There are two ways for users to activate “Smart Body Mode”. One is to click on the function button in the browser; the other is to ask ChatGPT whether it needs to be turned on when asking a question. However, this feature is currently only available for preview to paid users of ChatGPT Plus, Pro and Business, and OpenAI says it will optimize the speed and success rate of the task based on user feedback.

02.Experience innovation: AI deeply embedded, subvert traditional browsing habits

ChatGPT Atlas interface

In addition to the “Intelligent Body Mode”, ChatGPT Atlas also restructures the basic browsing experience in an “AI-first” way, so that the AI assistant becomes the core bridge between the user and the webpage.

Search Reversal

The most obvious change is Search Reversal. When the user enters “best travel destinations in 2024” in the address bar, the browser will not list a bunch of links first, but will directly pop up a consolidated answer generated by ChatGPT, which contains 3-5 destination recommendations, reasons, and links to related official websites, while the traditional list of links, images, and videos are relegated to the secondary tabs of “Search Results” and “Images”. Traditional link lists, images, and videos are relegated to sub-tabs such as “Search Results” and "Images".

This “give the answer before the source” model significantly reduces the time from the user's question to obtain effective information, in contrast to Google Chrome's “link-first”.

Contextual Sidebar Features

“Context Sidebar solves the pain point of ”copying and pasting to read web pages and ask questions“. After opening any web page, the ChatGPT panel on the right side will automatically load the page content, and users can directly ask questions ”summarize the core conclusions of this report“, ”explain this terminology“ without additional operations, and the AI will give answers based on the context of the current page. The AI will give the answer based on the current page context.

The “CursorChat” (CursorChat) function is even further, in writing emails, sending documents, selecting the text can trigger AI suggestions, real-time adjustments to the wording, change the tone of voice, for example, from the “colloquial” to the For example, you can change the tone from “colloquial” to "formal business style", or complete unfinished sentences.

03. Security Tout: Multiple restrictions superimposed on user control to reduce the risk of AI operations

User-controlled browser security settings

“Security must come first when letting AI operate web pages on behalf of people,” OpenAI repeatedly emphasized in its announcement. In order to prevent AI from overstepping its boundaries or being used for malicious purposes, ChatGPT Atlas has set strict technical boundaries for the “Intelligent Body Mode”, which prevents AI from running code, downloading files or installing extensions in browsers, or accessing the local applications and file systems of computers; and when it comes to sensitive websites such as those related to banking and payment, the Intelligent Body will automatically pause and wait for confirmation from the user to prevent unintended operations. When it comes to sensitive websites such as banking and payment sites, the AI will automatically pause and wait for confirmation from the user to prevent unintended operations.

Users also have full control over the entire process, and can pause, interrupt, or even take over the AI operation at any time via the interface buttons; if they are concerned about data security, they can choose to run the AI in “logged out mode”. If you are concerned about data security, you can choose to run the AI in "logged out mode", where the AI will not use cookies from your existing account and will not automatically log into any platform.

However, OpenAI also admits that there are still risks. Intelligent bodies may make mistakes in complex tasks and may be induced by “malicious commands” hidden in web pages (i.e., “indirect prompt injection”), which may lead to data theft and incorrect operations. Therefore, the platform suggests that users try to monitor the behavior of AI and prioritize sensitive operations to be done manually.

04. Personalization and privacy balance: “Browser memory” can be controlled and deleted.

ChatGPT Atlas Adds Optional “Browser Memory” Feature

To make AI more user-friendly, ChatGPT Atlas includes optional “Browser Memories” while safeguarding data privacy through multiple designs.

After opening this feature, ChatGPT will record the user's browsing preferences, historical content key points, such as last week's view of the job information, collection of renovation cases, and then directly ask the question, “Organize the last week to look at the requirements of the three product manager positions,” AI will be based on the memory of the rapid generation of summaries. However, these memories are only owned by the user's personal account and will not be used for OpenAI model training by default.

The user has full control over the “Browser Memory”. You can view all memories in the settings, manually archive useless records, or delete associated memories with the “Clear Browsing History” button; even if the memory is turned on, you can individually disable ChatGPT from reading the content of a particular web page with the “Visibility Toggle” in the address bar. Even if the memory is turned on, you can disable ChatGPT from reading a certain webpage through the "Visibility Toggle" in the address bar to prevent sensitive information from being recorded. In addition, Atlas also supports stealth mode, when it is turned on, chat history and browsing history will not be saved, and the AI will not be able to recall past memories.

05. macOS users experience first, Windows and other versions are under development

Apple Mac Users Can Complete ChatGPT Atlas Installation in Six Steps

Starting on the day of release, ChatGPT users worldwide can download the macOS version of ChatGPT Atlas at chatgpt.com/atlas. ChatGPT's Free, Plus, Pro, and Go users have free access to basic browsing features such as AI Search, Sidebar, Cursor Chat, and Browser Memory, but the “Smart Body Mode” preview is limited to Plus, Pro, and Business users. "Smart Body Mode" is previewed for Plus, Pro, and Business users only.

When opening the browser for the first time, users can import bookmarks, saved passwords, and browsing history from existing browsers (e.g. Chrome, Safari) with one click, without having to reconfigure them.OpenAI reveals that Windows, iOS, and Android versions are already under development, and the exact time of going live will be notified through official channels.

The core team of Atlas includes a number of browser veterans, including former Chrome engineer Darin Fisher and Ben Goodger, who was involved in the development of Chrome and Firefox, which is seen as an important part of OpenAI's efforts to ensure product compatibility and stability.

06. Market Jockeying: AI Browser Track Heats Up, Atlas Challenges and Opportunities

The release of ChatGPT Atlas has made the competition in the already hot AI browser track more intense. Currently on the market, Google Chrome with more than 60% share occupies absolute dominance, and is accelerating the depth of Gemini AI embedded in the browser, plans to launch a similar “intelligent body” of the automatic task function; Microsoft Edge, although the early integration of Copilot, but the market share is less than 6%. 6%, it is difficult to form a scale effect; AI search startup Perplexity previously launched Comet browser, although the main focus of “AI Q&A + real-time references”, but in the execution of complex tasks is far less than the ability of Atlas “intelligent body mode ”.

For OpenAI, the advantage of Atlas lies in ChatGPT's weekly active user base of more than 800 million people, these users have knowledge of the AI interaction mode, the conversion threshold is low; but the challenge is equally obvious, Chrome's user habits have been deeply rooted, and relying on the Google search ecosystem has formed a strong viscosity, Atlas to pry this part of the user, but also need to be in the stability of the function, Cross-platform experience on the continuous optimization.

Industry analysts believe that the significance of Atlas is not only “a new browser”, but also marks the layout of OpenAI to the “AI Internet layer”. In the future, if more and more users complete web operations through AI intelligence, advertising, e-commerce, service recommendations and other core business models of the Internet may be reconfigured. The Internet will also move from the era of “user-initiated content reading and writing” to the “executable” era of “AI executing tasks according to intent”, and ChatGPT Atlas is the first key testing ground for this shift. ChatGPT Atlas is the first key testing ground for this shift.

This article is from WeChat“Tencent Technology”, by Joyless.