Shulou(Shulou.com)11/24 Report--

This article comes from the official account of Wechat: deep Burning (ID:shenrancaijing), author: Wang Min, Editor: Xiang Xiaoyuan

Does AI want to "turn against the sky"?

On December 1, OpenAI, a Microsoft-funded AI lab, released a chat robot model, ChatGPT, which can simulate human language behavior and interact naturally with users.

As soon as it was published, ChatGPT was crazily "flirted" by users. Some people used it to write small essays, some people tested it with college entrance examination questions, and some people asked it to write code. Later, ChatGPT became more and more versatile. It was not only "articles can write articles, martial arts can change Bug", but also was evaluated by Musk as "terrifying", but also sparked heated discussion because of "whether it is possible to replace search engines in the future."

After the deep burning experience of ChatGPT, it is found that ChatGPT can quickly become popular all over the world because it can give high-quality answers, provide efficient access to information, have a strong language organization ability, and bring users an interactive experience beyond expectations.

However, the drawback was soon exposed. The strength of ChatGPT is to organize a complete answer, but the content given is not necessarily accurate, and even for some common sense questions with accurate answers, there will be loopholes, so that the well-known programmer question and answer site Stack Overflow issued an emergency announcement to temporarily ban ChatGPT.

But that doesn't stop curious users from trying. Just five days after its launch, Open AI CEO Sam Altman revealed on social media that the number of users had exceeded 1 million.

Before the popularity of AI painting went down, the AI dialogue model ChatGPT came, adding a handful of wood to the hot AIGC (Artificial Inteligence Generated Content, that is, the content of artificial intelligence production) since the second half of 2022. After the accumulation of technology in the past few years, AIGC has reached the tipping point of explosion.

Why is ChatGPT suddenly so popular? ChatGPT has become popular quickly because it is easy to operate and easy to use, and you can get the answer by typing the question in the dialog box after opening it. Moreover, it is like an all-round player who can often give unexpected answers.

In order to deal with daily conversation, ChatGPT can distinguish the description errors in some questions and refuse unreasonable and immoral requests directly.

One of the surprising features of ChatGPT is the ability to create literature. For example, give ChatGPT a topic and it can write a novel framework. When Deep Burning asked ChatGPT to write a novel framework with the theme of "AI changing the World", ChatGPT clearly gave the background, protagonist, plot and ending.

ChatGPT according to the proposition to write a novel framework / deep burning screenshot did not finish, after reminding, but also under the "training", continue to answer, complete, ChatGPT already has a certain memory ability, can have a continuous dialogue.

After having the user experience, the evaluation said, "the language organization ability, text level and logic ability of ChatGPT can be said to be amazing." Some users even plan to hand over the written work of daily newspaper, weekly newspaper, summary and reflection to ChatGPT to assist in the completion.

Ordinary text creation is only the most basic. ChatGPT can also find Bug for programmers' code. Some developers have indicated in the trial that ChatGPT provides a very detailed solution to their technical problems, which is more reliable than the answer of some search software. Amjad Masad, CEO of US code hosting platform Replit, tweeted that ChatGPT is an excellent "debugging partner". "it not only explains errors, but also can fix them and explain how to fix them."

ChatGPT has more business logic, not only knows its own strengths and weaknesses very well, can analyze its own competitive products and write marketing reports for itself, and even knows the world economic situation like the back of its hand and can express its own opinions.

Some people compare ChatGPT to a combination of "search engine and social software", which can get answers to questions in the process of real-time interaction. Many users interact with ChatGPT in a playful mindset, but with the deepening of the experience, they also worry about whether they will one day lose their jobs because of AI.

Guosheng Securities blockchain Research Institute said in the research report that ChatGPT has partially surpassed today's search engines in finding answers and solving problems. ChatGPT may change the way we obtain information and output content in the future. AIGC is expected to become a killer application that drives the outbreak of demand in the digital economy era.

ChatGPT can not achieve the current interaction without the accumulation of OpenAI in the field of AI pre-training model. OpenAI is an AI laboratory invested by Microsoft, and it is also one of the leading AI laboratories in the global AI field. Because of its technical leadership, it has always been regarded as an industry "beacon" in the AI field.

From 2018, OpenAI began to release the generative pre-training language model GPT (Generative Pre-trained Transformer), which can be used to generate articles, code, machine translation, question and answer and other content. At that time, the number of parameters was only 117 million.

Next, the number of parameters of each generation of OpenAI GPT model has exploded. The number of GPT-2 parameters released in February 2019 was 1.5 billion, while that of GPT-3 in May 2020 reached 175 billion, which directly led to the development trend of pre-training model in the field of AI in the past two years, setting off an arms race of spelling parameters and computing power.

GPT-4 is not here yet, and this release of ChatGPT is considered to be a fine-tuned version of GPT-3, or GPT-3.5.

OpenAI said that ChatGPT's model uses "human feedback reinforcement learning (RLHF)" training, including: human questions machine answers, machine questions human answers, and iterations, so that the model gradually has the ability to judge the generation of answers. Compared with GPT-3,ChatGPT, the main improvement point is the memory ability, which can achieve continuous dialogue and greatly improve the user experience in the dialogue interaction mode.

Be aware that ChatGPT's "random fabrication" ChatGPT does not always surprise people, there is no denying that it still has many limitations.

When you ask ChatGPT to create, it is free to play, but when your goal is to find the right answer, ChatGPT may sometimes fail to meet expectations.

For example, ChatGPT was asked to recite the poem "viewing the Sea", but it improvised one.

ChatGPT "improvisation", "viewing the sea" / deep burning screenshot and asked ChatGPT to do a primary school math problem, it seems to write a long list of calculations, but gives the wrong answer.

ChatGPT can not give correct answers to some questions / deep burning screenshot depth experience will find that ChatGPT is not very good at some cultural common sense problems and math problems. Moreover, ChatGPT's answer is often long, too long, seemingly logically self-consistent, but sometimes in a serious "bluff", careful thinking will find that there are certain loopholes. If non-professionals are unable to tell the accuracy of ChatGPT answers, they are very likely to be seriously misled.

As a result, Stack Overflow made the decision to temporarily ban it. It is reported that after the introduction of ChatGPT, a large number of answers generated by the model appeared on Stack Overflow soon. These answers usually require people with some professional knowledge to read carefully to tell if they are wrong.

This situation has had an impact on the quality management of Stack Overflow. ChatGPT also said: "because the answers generated by AI are not always accurate or relevant, it may confuse the wrong answers on Stack Overflow and mislead users who seek help." And also stated: "it is reasonable for Stack Overflow to prohibit the use of AI to generate answers."

OpenAI also mentioned that ChatGPT sometimes gives plausible but incorrect answers. On this issue, Sam Altman said: "We are trying to stop ChatGPT from fabricating randomly, and it is tricky to balance it with current technology at this stage. Over time, it will be improved based on user feedback."

As a result, some users' positioning for ChatGPT is as good friends who chat and relieve boredom. When you encounter a dilemma or need to find meaning, ChatGPT can give you the answer like the Book of Destiny.

Some users use ChatGPT as an efficiency tool, looking for references on copywriting or some professional issues.

After experiencing ChatGPT, a practitioner in the field of Web3 feels that ChatGPT is more beneficial to creative content creators, and that content based on research and interview types is difficult to be replaced by ChatGPT. As AIGC surges forward, content creators of a conventional nature may be eliminated. However, from auxiliary indexing to content presentation, ChatGPT not only greatly improves the efficiency of reading and collecting materials, but also stimulates its own thinking and creation system.

After the advent of ChatGPT, the hottest discussion is whether it can replace search engines. For this question, ChatGPT gives a "negative" answer: the core technology of the search engine is the search algorithm, which can quickly index, sort, filter web information, and provide relevant search results according to the user's query words, while ChatGPT can only generate natural language text and can not deal with the search algorithm.

ChatGPT's answer / Screenshot of subverting the search engine but ChatGPT can be combined with the search engine to improve the search experience. For example, ChatGPT can be used to build intelligent robots, explain and guide users' query words, and help users find the information they need quickly.

ChatGPT out of the circle, the AIGC era is also early praise and doubt Qifei, ChatGPT out of the circle, but also let its AIGC plate attention soar.

The so-called AIGC is the use of artificial intelligence technology to generate content. In 2022, the era of Web 3.0 accelerates, and there is a higher demand for the quantity and quality of content production. Compared with the previous Web1.0, Web2.0 era of UGC (user production content) and PGC (professional production content), AIGC, which represents artificial intelligence production content, is a new round of change in the mode of content production, and AIGC content will also grow exponentially in the Web3.0 era.

At the Baidu World Conference in July this year, Baidu founder Robin Li said, "AIGC may subvert the existing content production model and generate AI original content at a '1/10 cost' and a hundredfold production speed."

A recent research report by Guosheng Securities also pointed out that AIGC will be a productivity tool in the Web3 era. AIGC represents the beginning of a new round of paradigm shift.

Since August this year, the text generation image model Stable Diffusion officially open source, all of a sudden gave birth to a large number of AI painting applications, which directly promoted AI painting into the public view, recently widely experienced by the public, familiar with, become the AIGC track Deep-Fried Chicken.

However, the AIGC craze can not be detonated without an article in September entitled "generative AI: a New Creative World". It is mentioned that the areas that generative AI can deal with include knowledge work and creative work, involving billions of manual workers. Generative AI can increase the efficiency and creativity of these workers by at least 10%, with the potential to generate trillions of dollars in economic value.

There has been a boom in AIGC investment overseas, and even two unicorns have emerged. The post-investment valuation of Stability AI behind Stable Diffusion has climbed to $1 billion, while Jasper, a text-based AIGC unicorn, is currently valued at $1.5 billion.

Zhou Wenjing, founding partner of non-consensus capital, said to Shenjing that AIGC is not a new concept. Before that, AI generates text and AI generates pictures, but the technology is not yet mature and widely used. However, with the accumulation of technology in the past two years, AI has become mature in generating pictures, videos, music and text, and the quantitative change is ushering in a qualitative change.

AIGC is still in its early stage, there are some changes in the user experience at present, and it will be commercialized slowly in the future. She pointed out that at present, AIGC is oriented to the B side, and there is already some commercial cash space. In the context of reducing costs and increasing efficiency, AIGC can help many companies improve their efficiency in content production, which is consistent with the development demands of enterprises in the post-epidemic era.

There are also some problems that need to be solved urgently in the hot AIGC field. Take ChatGPT as an example, the content is inaccurate, or it will cause trouble to users. The previous popularity of AI painting is due to the copyright of the painting, whether it belongs to the AI painter or to the platform. Has been criticized by the outside world.

Although the hidden worries are there, AIGC is in the early stages of development, and the subversive changes that can be brought about in the future are still full of imagination.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.