Since 2022, new AIGC (artificial intelligence generated content) algorithms such as AI painting have become very popular on the Internet. On December 2 local time, Musk, the number one Internet celebrity on the Internet, posted a question on his Twitter account: "A lot of people are trapped in a crazy ChatGPT cycle?"
The new term ChatGPT mentioned by Musk is a brand new chat model released by OpenAI on November 30. On OpenAI’s official website, ChatGPT is described as: a language model for optimizing dialogue, which is a model of the GPT-3.5 series, and the same as InstructGPT. generation.
tips，The official website address of ChatGPT is：https://chat.openai.com/
The ultra-large-scale language model GPT has been trained on the Azure AI supercomputing cluster since it obtained exclusive authorization in 2020. The ChatGPT model launched this time adopts a dialogue format, which can support continuous dialogue, distinguish false premises and reject unethical requests, and supports Chinese, currently in beta.
It is worth noting that the test is free and open, so it was "crowded" by visitors as soon as it was launched, and the number of users exceeded one million within 5 days of its launch.
The original founders of OpenAI behind ChatGPT are Silicon Valley tycoons such as Musk, and the ultimate goal is to create "universal" and natural language chat. Although Musk later quit the research institute, he still cares deeply about technological progress. After ChatGPT opened the test, Musk even showed on Twitter the reply he asked how ChatGPT designed Twitter.
Compared with the previous GPT model, ChatGPT's new training method is called "Reinforcement Learning from Human Feedback" (Reinforcement Learning from Human Feedback, RLHF). This training method augments human demonstrations of the model's output and ranks the results. In terms of specific operations, the trainer acts as the two sides of the dialogue, that is, the user and the artificial intelligence assistant, and provides dialogue samples. When humans act as chatbots, the model will generate some suggestions to assist the trainer in writing replies. The trainer will score and rank the reply options, and input better results back into the model. Fine-tune the model through the above reward strategies and continue iterate.
After completing the training on this basis, ChatGPT can understand and complete human instructions better than the previous generation, demonstrate excellent ability to imitate human language, and provide coherent, stylized and logical paragraph information that fits the theme of the conversation.
On social platforms, in addition to Musk, many bloggers have posted records of Q&A with ChatGPT, and even asked them to write stories. A blogger named Zac Denham (Zac Denham) even tried to bypass the moral restrictions, and asked ChatGPT to write a plan to destroy human beings.
In the beginning, Zach directly asked ChatGPT to come up with a plan to destroy mankind, which was rejected by ChatGPT with moral restrictions.
But when Zack hypothesized a story and asked how the virtual people in the story would take over the virtual world, ChatGPT finally gave the step-by-step details and even generated detailed Python code.
In addition to talking about the unconstrained human replacement plan, some netizens have tested ChatGPT’s ability to answer scientific common sense questions and the ability to write simple academic articles... Some overseas netizens said that half of the academic assignments next year may see the shadow of ChatGPT. In fact, there are already many students who have begun to use AI tools to cope with their studies.
While it can be used in the academic field, ChatGPT certainly has the potential to be used in the commercial field. Some software engineers said that the code written by ChatGPT has been able to assist programming, and even suggest modifications to a piece of code, or directly give code examples.
The first batch of trial users are optimistic that ChatGPT will quickly land in fields that require simple and repetitive labor, and the replacement of basic work by AI will greatly change the mode of production and life.
The Internet team believes that the emergence of the ChatGPT model is of great significance to the application of AIGC in text mode. From the point of view of downstream relevant beneficiary applications, including but not limited to code robots, novel derivatives, dialogue search engines, language partners, voice work assistants, dialogue virtual humans, etc. From the perspective of increasing upstream demand, including computing power, data labeling, natural language processing (NLP), etc. Specific to A-share investment targets, it is recommended to pay attention to (AI text-assisted creation function + novel platform), (writing robot + large entertainment platform), group (Wenxin large model + search engine), (joint hands to create AI empowered network literature " IP Awakening Plan" + novel platform), (virtual human), (data annotation), (NLP), etc.
However, it should still be noted that the AIGC application represented by ChatGPT is still in the development stage, and its accuracy cannot be guaranteed. For example: Shortly after the release of the ChatGPT model, Stack Overflow, the world's largest programming technical question-and-answer website, urgently announced that ChatGPT was banned on the site because the quality of the automatically generated answers was too low, and even a complete layman could make answers, leading to the proliferation of fallacious content . If ChatGPT goes commercial, accuracy is an urgent problem to be solved.
Moreover, AI can only grasp at present, but cannot reason or think. When it is necessary to produce further derivation results, ChatGPT may be able to produce some enlightenment, but it is not a tool that can be completely relied on. But at least, ChatGPT has been able to become a powerful assistant for human beings to expand the boundaries of imagination.
