[Will AI take your job!?] ChatGPT finally releases AI agent "Operator". It can automate simple operations by automatically operating the browser. 《A thorough explanation of how to use it, the dangers, and use cases》

◾️Introduction to the channel "Chaen [AI Research Institute]"

Digirise Co., Ltd. operates "Chaen [AI Research Institute] ~The latest AI information that can be used at work~". The mission of this channel is to provide the fastest and most easy-to-understand AI learning content in Japan, and we post the latest AI information at least once or twice a week.

Main features:

X AI-focused channel boasting 130,000 followers
No.1 in corporate AI training in Japan
AI advisory track record for GMO Group, Persol Group, etc.
The main lecturer is AI expert Chaen, who has appeared on many TV programs.

Regularly distributed content:

AI Basics Course
Practical use of the latest AI tools
How to use ChatGPT
Commentary on the hot topic of AI news
Examples of AI use in business

Watch the video here

◾️Video content

Overview and potential of ChatGPT's new feature "Operator"

Our company, Digirise, conducted a detailed review and explained how to practically use the new feature "Operator" announced by ChatGPT in January 2024. This new feature is an innovative AI agent that combines ChatGPT-4's image recognition function with reinforcement learning.

Key features of the Operator include:

Allows automatic browser operation
Screen recognition for appropriate action
Automate routine tasks such as filling out forms and online shopping
Security design including user approval process
Automate your website without an API

Currently, the beta version is only available to ChatGPT Plus plan users ($20/month). Access is limited to the US due to geographic restrictions, but it can be accessed from Japan by using a VPN.

In our testing, the accuracy of robots compared to humans was about 58%, meaning that humans still have an advantage in complex tasks. However, simple routine tasks can be automated with sufficient accuracy, and this is a function that is expected to develop further in the future.

Demonstration of automatic browser operation and basic functions

In order to verify the practicality of Operator, Digirise conducted multiple tests based on actual business scenarios. Of particular note are the following basic operations that have been confirmed:

Specific operation examples:

Automated Post Creation and Distribution with X
Creating and sending emails in Gmail
Searching and collecting information on websites
Auto-fill forms
Attaching and parsing files

In actual operation, when you enter instructions on the Operator screen, a new browser window will open and the AI will automatically perform the operation. For example, if you give the instruction "Search for today's AI news, summarize it, and post it to X," it will automatically execute the following steps:

Search in the browser
Collection of related articles
Summary
Logging in to X (formerly Twitter)
Writing and posting a post

One notable feature is the implementation of a "takeover mode" that requests human approval midway through an operation, so user confirmation is always required before important operations. It also has a task saving function, so frequently performed tasks can be registered and reused efficiently.

Security features and precautions

Digirise has conducted detailed verification of the security aspects of Operator, assuming its adoption in companies. Due to the nature of automated operation by AI, responding to security risks is a particularly important aspect.

OpenAI implements the following key features to ensure safe use in the enterprise:

Confidential information protection function
- Disable screen capture when entering password
- Excluding input information from AI learning
Operation Approval System
- Important: User approval required before operation
- Automatically detect and stop suspicious behavior

When adopting Operator, companies should first check whether it complies with their own security policies. Of particular note is that Operator places restrictions on its use on financial-related sites and on destructive operations such as file deletion.

In addition, the sending of direct messages and DMs is also restricted, and the system is designed to prevent abuse such as spam. In this way, OpenAI has realized a robust security design with an eye toward business use, but it is also important for the adopting company to build an appropriate security management system.

At Digirise, we provide implementation support in line with your company's security policies, as well as security advice during operation.

Specific use cases in companies

As a result of our verification, we at Digirise found that Operator is particularly effective when used in the following business areas:

Sales support operations
- Automatic collection of company information
- Auto-fill contact forms
Information gathering and analysis
- Automated market trend research
- Gathering competitive information

In fact, in one case where a major company used the system, the time it took to research companies and prepare for business negotiations was reduced to one-third of what it used to be. In addition, the time it took to respond to routine inquiries and enter data has also been significantly reduced.

One task in which it was particularly effective was the collection of information from press releases. Previously, staff members had to manually check each company's press releases and extract the necessary information, but by using Operator, this process could be automated. As a result, the man-hours from information collection to analysis were reduced by approximately 701 TP3T.

Operator is also highly effective in searches and comparisons across multiple websites, for example, comparing product prices or analyzing service details, a task that would have taken hours to complete manually can now be completed in minutes.

However, at present it is difficult to fully automate all tasks, and final confirmation and fine-tuning by humans is necessary. We provide implementation support that takes into consideration the optimal division of roles between humans and AI.