Mengchen Yang Jing from the Temple of Concave Fei
Qubits | Official account QbitAI
It is said that ChatGPT is the "iPhone moment" of AI, and now its "app store" is here.
OpenAI just threw out another king bomb -
Announced the launch of a plug-in feature that gives ChatGPT the ability to use tools, network, and run computations.
For example, in the official demo, once ChatGPT is connected to Wolfram Alpha, a mathematical knowledge engine, it no longer has to worry about inaccurate numerical calculations.
It also opens up two plugins of OpenAI's own: a web browser and a code interpreter.
In the past, because of the lack of Internet access, users could only query messages before September 2021; But now not only can you directly retrieve the latest news, but even mathematical calculations have been solved!
(No more worrying about ChatGPT!) ）
In addition, it also supports individuals or enterprises to feed private data (documents, notes, emails, etc.) to ChatGPT, becoming the "second brain" of people or intelligent assistants of enterprises.
A flash-by-3-page list of third-party plugins also includes:
Let ChatGPT query the meaning of words and phrases in Chinese words and phrases in the world
Let ChatGPT check flight and hotel information to help you plan your trip
Let ChatGPT access major e-commerce data to help you compare prices or even place orders directly
Connect ChatGPT to the automation platform Zapier and connect with almost any office software to create your own intelligent workflow
……
And this is just the beginning, the permissions and methods for making plugins will also be open to developers.
A senior programmer lamented:
I've developed plugin systems for many software, and the ChatGPT plugin interface is probably the craziest and most impressive I've ever seen in my life.
Andrej Karpathy, Tesla's former head of AI, who has just returned to OpenAI, said:
The GPT class model is a new type of computer architecture that runs on text and can "talk" not only to us humans, but also to existing software infrastructure, with APIs being the first step and plugins being the second step.
ChatGPT has eyes and ears
According to the official introduction, this plug-in function is mainly a language model design tool with security as the core principle
(Well, in other words, it won't let you use it that quickly.) ）
OpenAI still has to put users on the waitlist first.
However, they also said that while it is now in alpha, prioritizing a few developers and ChatGPT Plus users, they also pack tickets:
It is only a matter of time before it opens up on a larger scale.
In terms of the "preliminary functions" that have been implemented so far, ChatGPT can already have three capabilities:
Access the latest information such as sports scores, stock prices, latest news;
Retrieve knowledge base information such as company documents, personal notes, running calculations;
Perform actions on behalf of users, such as booking airline tickets, ordering meals.
Currently supported third-party plugins, you can see restaurant recommendations, travel planning, shopping, AI language teachers, online stores, as well as academic knowledge application Wolfram, application library (interact with more than 5,000 applications, including Google Sheets)...
In this way, it seems that almost covers our life, food, clothing, housing, work and study.
In addition to this, OpenAI has its own two plugins: a web browser and a code interpreter, and open-sourced a knowledge base to retrieve the plugin's code.
As mentioned earlier, OpenAI places great emphasis on security principles this time, and they have implemented a number of safeguard measures, all of which are one-to-one for functions.
1. Third-party plug-ins
OpenAI even gives a whole set of building processes: "How to build your plugin on ChatGPT", which is open sourced on Github.
The display on ChatGPT is like this, users can choose the plugin and download it by themselves.
It can then be used on the chat interface.
2. Networking.
Can you tell me, which person, which film won the Oscars in these categories?
As you can see, ChatGPT has more "browsing" results in the chat interface, and directly gets the latest 2023 results.
OpenAI's security measures include: users can only use the plugin to retrieve information, and do not include "transactional actions", such as form submission; The use of the Bing retrieval API inherits Microsoft's reliability and authenticity in terms of information sources; Run in a stand-alone server; Show source, etc.
3. Code interpreter: You can use Python to handle the experimental ChatGPT model uploaded and downloaded.
In simple terms, it provides a Python interpreter that works in a sandbox, firewall execution environment, and some temporary disk space.
Code run by the interpreter plugin is evaluated in a persistent session, and subsequent calls can build on each other. Currently, the feature supports uploading files to the current conversation workspace and downloading work results.
However, OpenAI said that this code interpreter mainly provides some of the most basic functions of computers, allowing some junior programmers and enthusiasts to provide a smoother interface.
Currently recommended features include:
Solve quantitative and qualitative mathematical problems
Perform data analysis and visualization
Convert files between formats
One More Thing
For ChatGPT's new plug-in function, some developers said that this basically provides Apple App Store-style services.
But others have also suggested that there is no monetization model for developers yet.
Critics argue that OpenAI has just killed a startup, LangChain, and that going like this will hinder innovation.
LangChain, which provides a Python framework for quickly building applications using the Language Model API, has just secured $10 million in seed funding.
Do you think OpenAI will become the Apple of the AI era? Who will be Android in the AI era?
Reference Links:
[1]https://openai.com/blog/chatgpt-plugins
[2]https://platform.openai.com/docs/plugins/introduction