In the past two days, there has been a craze for sharing chat records on the Internet in China and abroad, and the chat partner is the same "person", an AI robot that can chat, write copy, type code, and draw pictures - ChatGPT.

ChatGPT is the latest generation of conversational robots developed by OpenAI, an American artificial intelligence research organization. Compared with previous AI software, the intelligence of ChatGPT far exceeds people's expectations. No matter what questions users input in any field, ChatGPT can basically answer them fluently and logically. There is almost no trace of AI in the content, which has attracted a large number of netizens to "train" it tirelessly.

In less than a week, the number of users on ChatGPT has exceeded 1 million. At its peak, it was even crashed by enthusiastic netizens.

What is the origin of ChatGPT that has caused such a carnival in the technology industry?

First, let ChatGPT introduce himself.

Chat, as the name suggests, is positioned as a chatbot. GPT stands for Generative Pre-trained Transformer, a text generation deep learning model trained based on available Internet data. It is used for question and answering, text summary generation, machine translation, content classification, code generation, conversational AI, and so on.

The introduction mentioned GPT-3, which means that the model has now developed to the third generation. Its predecessor, GPT-1, was born in 2018, and its functions were more limited to understanding user input text rather than compiling and interacting; GPT-2 came out in 2019. It recorded a more complete data set based on GPT-1, and in addition to the ability to understand, GPT-2 has a strong ability to produce content.

The current GPT-3 has a higher amount of data and larger parameters. The amount of data of GPT-3 is about 1,000 times that of GPT-2, and the model parameters are about 100 times that of GPT-2. The English Wikipedia containing 6 million articles only accounts for 0.6% of its total learning data set. Simply put, it is an encyclopedia covering knowledge such as literature, politics, medicine, law, and religion.

The results show that GPT-3 learns more deeply and faster than the previous generation of models, and also performs well in problems other than natural language processing. It can generate more natural and diverse responses based on the input it receives, that is, it is closer to human language habits and will not be mechanically repeated; it can even automatically output a series of simple codes according to instructions, and it performs extremely well in foreign language translation.

Faced with such a product at the forefront of technology, crypto users who are “tech trendsetters” will naturally not miss it. In fact, Web3 users may be one of the most active groups in testing ChatGPT. People like Binance boss CZ and Ethereum founder Vitalik have already participated in person.

For example, CZ asked ChatGPT how to run a good crypto exchange, and ChatGPT’s answer was “ensure platform security, provide a user-friendly interface, support a wide range of cryptocurrencies, provide high-quality customer support, and maintain regulatory compliance.”

From @cz_binance

CZ seemed quite satisfied with the answer. He believed that "these points are better summarized than most crypto exchange founders, including himself," and said that "I will print out the answers and post them on the wall."

Vitalik Buterin published an article after conducting a coding experiment on ChatGPT. According to the results of his experiment, ChatGPT would make errors during the coding process, but it also proposed some new coding patterns and would speed up the coding process. Vitalik believes that "ChatGPT is very helpful as a programming auxiliary tool."

However, he also made it clear that "AI is far from replacing human programmers."

As Vitalik said, ChatGPT does not have human imagination and creativity. The content it outputs is based on the data it already has, so it is impossible to completely replace human programmers. However, ChatGPT is still very outstanding in its ability to integrate existing knowledge.

Based on this, we can try to discuss what capabilities of ChatGPT can bring changes to the Web3 industry.

knowledge base

ChatGPT has powerful information summarization and induction capabilities. By inputting information from the relevant knowledge base during the training process, when the user enters questions related to the "trigger word", ChatGPT can provide corresponding answers based on the trained knowledge base information.

Therefore, ChatGPT can be used as a tool for learning the basics of Web3. When you enter keywords such as "Bitcoin", "Web3", and "blockchain" in ChatGPT, the answers it outputs are of high quality in terms of relevance and accuracy.

By @billtheinvestor

However, the data of ChatGPT's training model is currently up to 2021, which means that you will not get a correct answer if you ask about things that happened after 2021 (such as the Ethereum merger). After all, it is only in the testing phase now. I believe that after the official version is launched, as long as ChatGPT is regularly trained and updated, its answers will be able to reflect the latest content of the knowledge base in a timely and accurate manner.

Coding ability

The most surprising innovation of ChatGPT is its ability to understand and generate code. For example, if you tell it what function you want to achieve, as long as you describe it in enough detail, it can accurately understand and generate code.

For example, if you ask it to write a Solidity code to publish NFT, ChatGPT can do it easily.

This is of great significance to the Web3 industry. For example, ChatGPT can automatically write some basic codes, assist programmers in writing codes, troubleshoot code bugs, optimize codes, etc. It can even act as a security guard to identify phishing attacks and warn of hacker attacks. This process may still require programmers to intervene and debug, but it will also effectively improve the work efficiency of Web3 technicians and provide a lower entry threshold for some users with low coding skills, making Web3 technology more popular.

writing ability

ChatGPT can generate human-like text responses based on input information. In fact, ChatGPT has passed the Turing test in the field of writing, which means that you can no longer tell which articles it created were created by ChatGPT and which were created manually.

Therefore, ChatGPT can basically handle some simple copywriting needs easily.

For example, write a project slogan:

Write an NFT airdrop event announcement:

Analyzing the market is no problem (it can be said that you have mastered the essence of being an analyst):

The more detailed your input is, the closer ChatGPT's output will be to what you want, and you can basically use it directly with just a few modifications. However, ChatGPT is still a little lacking in long copywriting, and is not yet competent for some creations that require strict logical reasoning.

Conversational Ability

Conversation is the "traditional craft" of ChatGPT. Its original positioning was a chat tool, so the direct application of ChatGPT is to be used as a chat robot for social media such as Discord and Telegram.

GPT-3, which has undergone a lot of training, can be said to be very "learned and talented". Simply sending a welcome message to the group and "GM" is obviously a waste of talent.

Since ChatGPT absorbs a lot of knowledge related to different fields, it means that it can call on a lot of knowledge to talk to you. Whether you have professional knowledge confusion or emotional needs, even if ChatGPT cannot give an ideal answer, it can still give you inspiration through other related answers, and it will also learn during the conversation. This kind of AI plus personal character customization will undoubtedly improve the service level of customer service robots by more than one level.

For example, Web3 game developer Webaverse has previously tried to connect OpenAI's GPT-3 model to the NPCs in the game. Compared with traditional NPCs, these NPCs will be smarter. In addition to answering questions in the game, they can also answer some simple questions about the real world.

Summarize

It has to be said that the innovation of a technology has an impact on all aspects of an industry. The above speculations about ChatGPT's transformation of the Web3 industry are limited to the current information and personal imagination.

Technology is developing rapidly, and according to information revealed by OpenAI, ChatGPT is actually just the "appetizer". GPT-4, which will be released next year, is the real highlight. GPT-4 can pass the Turing test and is so advanced that it is indistinguishable from humans. I believe that by then there will be more points of integration with Web3 that are beyond our imagination.