Introduction
OpenAI recently released the chat robot model ChatGPT, which quickly spread across the entire network. It interacts in a conversational manner. Interact with users in a more human-like conversational manner, answering questions, admitting mistakes, challenging incorrect premises, and rejecting inappropriate requests. The high-quality answers and addictive interactive experience have exclaimed from both inside and outside the circle.
Why is there such a high rating? What is the theoretical support? What is the technical principle behind it? What are the problems and solutions to be solved? How does the capital view this matter? The goal of this article is to explain these issues to everyone in detail.
1 What is ChatGPT?
The essence of ChatGPT is a language model applied in dialogue scenarios. It is fine-tuned based on GPT3.5 through reinforcement learning of human feedback, and can answer follow-up questions, admit mistakes, question incorrect premises, and reject inappropriate requests. First of all, let our protagonist ChatGPT introduce himself.
1.1 Let ChatGPT introduce yourself
What is ChatGPT? Since ChatGPT is a language model, what is a language model? Through reinforcement learning training, what is reinforcement learning?
1.2 The Rise and Explosion of OpenAI on a Global Scale
Released on November 30, it has sparked heated discussions in North America and Europe. Then it started to catch fire in the country. Users from all over the world are scrambling to post their highly creative results of communicating with ChatGPT. ChatGPT has shown various amazing abilities in the crazy tests of a large number of netizens, such as fluent answering, writing code, writing scripts, error correction, etc., even making journalists, editors, programmers and other practitioners feel threatened, not to mention that it will replace The Google Search Engine. After AlphaGo defeated Li Shishi and AI painting became popular, ChatGPT became another new Internet celebrity. Below is the Google Global Index, we can see the degree of popularity.
Comparing the major platforms in China, it was WeChat that became popular first. From the WeChat index, we can see that 97.48% come from official accounts, which started in the technology circle and quickly expanded to the investment circle. The first thing I learned about ChatGPT related information was on the science and technology public account I followed, and then I saw various shocking keywords about ChatGPT introduced on various public accounts such as earthquake, killing madness, and destroying human beings. Then all walks of life participated in interesting activities, asking math questions, asking about history, writing novels, writing daily newspapers, writing codes to find bugs... 1.3 The funder
behind OpenAI
OpenAI is an artificial intelligence research experiment Chamber for the promotion and development of friendly artificial intelligence for the benefit of humanity as a whole. OpenAI was originally a non-profit organization, but in order to better realize the integration of production and research, a subsidiary of OpenAI LP was established in March 2019 for profit.
In July 2019, Microsoft and the two investors will work together. In June 2020, the GPT-3 language model was announced, refreshing people's understanding of AI. The GPT series of language models keeps us looking forward to general artificial intelligence (AGI).
OpenAI's goal was very ambitious at the beginning, solving general artificial intelligence problems, mainly involving reinforcement learning and generative models.
Reinforcement learning was first considered to be an important means of realizing human general intelligence. AlphaGo Zero developed by DeepMind in 2016 used reinforcement learning training to make the historical experience of human Go become "Zero", marking an important step towards general artificial intelligence. step. In 2019, OpenAI defeated the human world champion in the "Dota2" competition. OpenAI has a lot of in-depth research in reinforcement learning. Dactyl is also a robot hand that OpenAI can manipulate objects with high precision through reinforcement reinforcement learning. OpenAI Gym is a toolkit for developing and comparing reinforcement learning algorithms, so ChatGPT uses Reinforcement learning is also a matter of course.
In terms of generative models, we are familiar with GPT-3, which is a powerful language model that can generate human-like fluent language. DellE 2 is a recently popular AI painting that generates pictures based on text descriptions. Codex is trained through github code in cooperation with Microsoft. It can generate code and Debug, and has been commercialized.
Elon Musk, the former founder of OpenAI, also commented on ChatGPT!
See
how ChatGPT introduces its company?
2
Some interesting experiences of ChatGPT
As a chat robot, we experience and find that it is more fluent and natural in coherent question-and-answer than traditional robots. There are already many small programs on WeChat that you can experience, or directly talk about ChatGPT connected to WeChat, let’s experience ChatGPT that has been connected to Enterprise WeChat.
1) The company always pays attention to values. The answer to the first question is half correct. The company stipulates that it should be returned as soon as possible. It cannot be returned under special circumstances. No matter how much the value is, it needs to be declared, so ChatGPT should not be able to join our company. The answer to the second classic question is quite in line with the company's values of integrity, haha.
2) There are always people putting pigeons in the activities of company associations. Let me ask the people of ChatGPT what their mentality is. Seeing that the reason for helping to think is the same as what the person who released the pigeon said, I am a little suspicious of their mentality.
3
) Workplace tool people can see if they can do it through ChatGPT. You can see that the copywriting is still good. This outline of the answer has a little problem but it doesn't affect the overall situation. Telling stories and writing scripts is also good. First, I helped Chinese football to write the copywriting for participating in the World Cup, but I don’t know if it can be used.
4) Colleagues around you attach great importance to children's education, so let's start with the baby and see if ChatGPT can take the baby to learn. Answer the general knowledge questions correctly, and the reasoning ability of the math questions, I am worried that the baby will not be able to pass the junior high school entrance examination, but the parents will check it! At the same time, I also checked how he did the brain teaser, and the fool didn't get it right.
5) It is known as a programming artifact that can write code and fix bugs. It is no problem to check Leetcode with medium difficulty. Although it is self-deprecating and does not know how to program, but according to the test and the verification ability of netizens, it is really outstanding.