Digital Krypton (digital36kr)
Have you been using ChatGPT lately?
Text | Digital Krypton
Cover source: IC photo
ChatGPT, a conversational large-scale language model released by artificial intelligence laboratory OpenAI, has recently attracted hot attention, and its user base has reached one million in just 4 days, and the number of users has exceeded 100 million in 2 months. It's surprising how hot it is.
ChatGPT is prompting people to imagine a large number of application scenarios and trigger the transformation of the AI industry. At present, technology giants such as Microsoft and Google have come down and preemptively laid out a new AI track. Just this month, Microsoft announced that all its products will integrate ChatGPT, and Google officially released the next generation of dialogue AI system Bard. The tech world is about to usher in a series of major moves.
Against this backdrop, we caught up with some friends in the tech community to chat with them about the high-profile ChatGPT and their views on the future of adoption.
Huang Yukai, Deputy General Manager and Technical Head of Haitian AAC:
In the decades of AI development, AI was previously considered an efficiency tool that could help humans improve productivity in all aspects, and AI had not been involved in the field of content creation in the past. But with DALL· With the advent of E, Stable Diffusion, and the recent ChatGPT, AI has officially entered the field of content creation.
The field of content creation has brought many commercial breakthroughs from the PGC era dominated by professionals producing content in the past to the UGC era of mass production of content. Now we are experiencing the commercialization of ChatGPT applications.
He Wanyu, CEO of Xiaoku Technology:
From the end of '22 to the beginning of '23, ChatGPT captured a million users in five days at the fastest speed in history, and achieved 100 million activity two months later, so hot that schools must ban students from using ChatGPT to write homework.
Xiaoku tried ChatGPT for the first time from the end of November to the beginning of December, and shared some of the results at the annual meeting of Shenzhen Green and Intelligent Construction Society. We tested the definition of "smart building" in the smart building pilot city work that we were participating in at the time. Whether in Chinese or English, ChatGPT gives the following multi-round dialogue answers with certain reference value. Such results are amazing, but for connoisseurs, these answers are clearly not enough.
Source: Xiaoku Technology
Undoubtedly, the application scenarios of such large-scale language models are indeed very extensive, and they are very close to the "jack of all trades" in science fiction (it can give a direct and relatively accurate answer if you ask anything, rather than thousands of indirect answer clues like search engines). As you can imagine, this technology has broad prospects for use in many consumer areas. For example, write an advertisement, an introduction, a story, etc.
At present, we are also studying how to apply ChatGPT's API to achieve vertical applications in the construction field, especially based on domestic business scenarios. Maybe one day we're looking at a design description that came from its creation.
Source Code Capital Partner Yungang Huang:
ChatGPT was released in November '22, marking the industry's entry into the public eye, but it's still a very early stage. There is no doubt that everyone's interest is very high, and this interest may come from the fact that AIGC is a potentially huge technological change, there may be a very high ceiling, and this change is very scarce.
From a technical point of view, ChatGPT is not an obvious technical change, behind it is the Transformer model proposed in 2017 after several years of polishing and iteration by OpenAI, gradually expanding the scale of the model, to a productized output today. Every time an important technology in the history of AI is introduced to the public, it will cause a huge discussion, but most of the time, not long after the discussion, people treat the capabilities that AI has demonstrated as a function that AI can easily achieve.
But ChatGPT can still be seen as a landmark event, which represents the beginning of a new wave of AI technology, that is, generative AI or the industrial beginning of AIGC. Through the natural communication effect of ChatGPT, everyone's understanding of this industry will be deepened, which will obviously lead to an increase in investment and participation in this field. And then pull the research and development to form a new wave of technology.
What is certain is that more and more industries will improve efficiency with the empowerment of AIGC, and more pendant tools may gradually appear. People need to embrace, learn, and use these tools to keep up with the times. It may be expected that some more disruptive impact will be expected, but we are not sure about this, AIGC may have the potential to disrupt certain industries, and the speed may be fast and slow in different areas.
At the INFRA and model layers, everyone pays a lot of attention, hoping to dig into the best quality team to do upstream models. At the application layer, it is also actively exploring, and many scenarios have greater possibilities, which require more industry awareness and technical understanding to complete PMF.
In addition, everyone's understanding and attention to core technologies are also strengthening, and there are more thoughts on the nature of changes in the underlying technology.
Zhang Peng, CDO of Wanton Jiufang:
In fact, I am more optimistic about AI technology, because AI technology can actually change a lot of operation and interaction forms, ChatGPT can actually change a lot of experience and communication forms, in business, it is not only a chat robot, it can provide human-like customer service capabilities, including sales consulting capabilities, because we have a lot of channel consulting, in fact, we are also using a lot of labor to support this business. That part can actually replace people with ChatGPT to complete the answers to some general questions.
That another point, we can change the shopping guide service system and training system through this technology, the previous training is done manually, whether online, offline are people talking, if combined with this ChatGPT AI technology and virtual reality technology, in fact, can be completed online through AI. In this way, we can better operate and manage the trainers, because for AI, its capabilities are equal, as long as the AI is set up, it will be the same no matter how much training is done. This will help a lot to reduce the overall operating costs.
In addition, we used RPA to do the review list, which is actually relatively standardized, but ChatGPT actually does a lot of non-standardized actions. It's not formulaic, it can be technically learned and then executed. It's smarter and more human-like. In fact, we can transfer more process-based operations to the form of automation with the help of ChatGPT technology. This will greatly improve the synergy efficiency and operational efficiency, so we are actually quite optimistic about this aspect.
In addition, there are still a lot of use space and utilization scenarios, such as the quality inspection link on the back-end production line, which is currently partly done with the help of automated means, but some places are still done manually or semi-manually in the middle, in fact, maybe in the future, some ChatGPT This very powerful AI capability, we can also improve accordingly. Therefore, the aspect of supply chain optimization should also be able to be achieved through these technologies. So I'm also researching and learning how to integrate this with real business.
In summary, ChatGPT should have relatively large application space and scenarios in terms of user experience improvement and user experience model subversion and innovation, including channel operation and management system and intelligent level of terminal service and training system. At the same time, we can also use this ability to explore some capabilities for the back-end supply chain, I think this kind of human-like learning ability and knowledge graph learning ability, in fact, there is a lot of room for application.
Yuan Jinhui, CEO of First-class Technology:
Technological development is incremental most of the time, and step development is very scarce and occurs occasionally (such as AlphaGo, AlphaFold, etc.). Even with all the other step moments that have occurred in deep learning in recent years, ChatGPT is arguably the most significant. ChatGPT's breakthrough is not only academic, but also has grown to 100 million users in just two months, becoming the world's fastest-growing product, showing that the technology is mature enough to be productized. Even if there is still room for improvement in the future, ChatGPT's breakthrough is the first to show the world that the goal of computers understanding and generating languages is achievable, and it shows a path.
Its significance can be compared to mathematician Zhang Yitang's research breakthrough in the twin prime number conjecture, Zhang Yitang found a distance of no more than 70 million prime numbers, and the twin prime number requirement distance of 2 is far away, but this is not important, an American mathematician commented that the distance from 70 million to 2 is insignificant compared to the distance from infinity to 70 million.
In fact, soon after Zhang Yitang's paper was published, that distance was improved to several hundred, very close to 2. In short, the linguistic intelligence of computers is no longer just an academic research problem, but can become the infrastructure of the future.
The industry should have realized that ChatGPT is the so-called Game Changer, which can lead to a reshuffle of related industries in the medium term and reshape the industrial landscape (related inputs such as Microsoft and Google).
Bill Gates believes: "The birth of this artificial intelligence technology has great historical significance, no less than the birth of the Internet or personal computers, in 2023, artificial intelligence will become the most concerned hot plate, artificial intelligence technology in the future, the impact on society will be no less than personal computers and other technologies." ”。
ChatGPT represents a major breakthrough in general artificial intelligence (AGI), and it will soon happen that many kinds of work efficiency will be greatly improved, and even quite a few kinds of jobs will be replaced by machines.
For individuals, being able to get ChatGPT and not being able to use ChatGPT will have a decisive impact on the individual's development potential (initially a comparison of physical strength, then a competition for high ability to use tools, until today whether or not to use ChatGPT will make a difference in intelligence).
Similarly, companies and countries with ChatGPT capabilities will inevitably gain an asymmetric advantage in competition over companies and countries without similar capabilities.
Newsletter Tweet Yu Xinlie:
One of the reasons why ChatGPT is so popular is that ordinary people can use the interface casually, and it can also give eye-catching high-quality results. That feeling is very similar to the first time we used some smart products, wow. For example, the first time I clicked on my phone, the lock of the bicycle in front of me snapped open.
"Chat" is just an interactive interface, and the essence of the service provided by ChatGPT technology is to efficiently process a large number of text messages. Through massive text training, it can quickly summarize, summarize, adjust the structure, cite historical stories, and list real examples.
Why is it suitable for combination with search engines? Search engines used to just list a bunch of content links. Users need a page to open and browse. The advantage of ChatGPT technology is the ability to collect, integrate, and summarize output. It helps you take care of the dirty work of the search process and directly gives you content that can guide your next move. (The more mature the technology, the better the experience)
In 2008, a senior vice president at Google showed Larry Page some of Gmail's new features, and he asked, Why can't Gmail automatically write me emails? Today, technology like ChatGPT does just that.
This morning, at the Bing conference with integrated ChatGPT technology, Microsoft showed an interesting application scenario: use the Edge browser to open Gap's 15-page earnings PDF, send a "refinement point" to the AI assistant, and get a bullet point summary in less than 10 seconds. What's more, knock on the sentence "list a table and compare with lululemon", the AI assistant can also give you a comparison table of financial reporting points.
So what ChatGPT technology is good at is text information processing, helping you summarize, integrate information, and help you write small essays. But be careful not to blindly believe the factual content it provides.
Technologies such as ChatGPT will become a basic technical service similar to cloud computing, and more and more companies will embed it in different application scenarios, which can improve efficiency and user experience.
In the field of daily office/content creation, technologies such as ChatGPT will provide some help in text creation, improve content creation efficiency, and reduce the cost of regular content creation. There are already small companies and teams that use ChatGPT to automatically generate content that doesn't require high quality.
Technologies such as ChatGPT will be embedded in many fields as a basic service, such as search and office software. Optimistic about its commercial prospects, but the limitations are also obvious.
AI popular science author Yu Meow:
ChatGPT is a very large language generation model, which can be simply understood as containing human language and the internal logic of language. Therefore, ChatGPT can predict the language very well, give it a word, it can predict what the next word will be, by predicting what the next word is, and then predict what the next word will be, which is the basic principle of ChatGPT.
But the popular saying now is that ChatGPT has creative power, and I think this is very controversial.
ChatGPT is a lot like a search engine that contains too much of the corpus that humans already have. After asking a question, it looks for information like a search engine, except that it integrates the information and outputs it to humans. It's more of a generator, and it's hard to call it creativity.
If ChatGPT is particularly amazing, it's the excellent natural language processing power: it can generate a piece of language that looks real, much like a human, enough to fake it. But the problem is that it lacks common sense, and the generated content often has errors. That is to say, it looks like it is "right", but it actually imitates the tone of "right", and there are many errors in the content. Especially for some scientific questions, if humans have not written the correct answer to this question on the Internet, it will not be able to give the correct answer, or it will confuse different questions in order to give the answer. Previous tests have shown that ChatGPT's answers are questionable and difficult to believe.
ChatGPT is a language model, because the training data contains images, code and other content, so that it can not only generate articles according to problems and requirements, it can also generate pictures, web pages, code according to requirements, these are its applications.
If you use it to refer to commercial applications, it is actually a bit unsolvable. Currently it only exists as a ChatBot or generator, which can help gather information for research efforts. It can charge users as a tool like Grammaly, and ChatGPT does exactly that. But other forms of commercial application may be more difficult, which is also the common problem that AI technology needs to solve: finding business scenarios.
Interactive topics
What do you think of ChatGPT and its application prospects? Welcome to leave a message to communicate with us
Digital Intelligence Foresight - ChatGPT series content (can be viewed on the digital Krypton public account (ID: digital36kr)):
"Behind the popularity of ChatGPT: five years of suffering, three iterations of AI routes|Industry Watch"
Behind ChatGPT: A Genius, Tens of Billions of Financings and 100 Million Monthly Lives | Digital Intelligence Foresight
《The fastest monthly life exceeds 100 million, but ChatGPT may eventually be defeated by open source| Digital Intelligence Foresight
THE END
We are building a "TOB field" theme communication community, welcome to scan the code to add digital Krypton assistant, and send business cards or personal introductions, we will continue to recommend relevant industry information and activities for you.
For more event information and industry details, please pay attention to the WeChat public account "Digital Time Krypton" (WeChat ID :d igital36kr), looking forward to your arrival.
Scan the code to join the group
Digital Krypton (digital36kr)
This article is from the WeChat public account
"Digital Krypton" (ID: digital36kr)
, author: Digital Time Krypton, 36 Kr published with permission.