First published on Jiemian on December 13, 2022.
Copyright belongs to the original media and author. Please do not reprint without authorization.
By Jingwen
In early December, Open AI’s artificial intelligence laboratory released a natural language generation model named ChatGPT. According to official data, the number of ChatGPT users exceeded one million on December 5th, just five days after its launch.
Aside from the skyrocketing number of users, ChatGPT has become a new traffic hotspot. As of the time of writing, discussions about ChatGPT are ubiquitous on professional and social media. According to several popular trend indexes in China, such as Baidu Index and WeChat Index, ChatGPT’s popularity has even surpassed the peak of AlphaGo in the past two weeks.
This has reignited a widespread public discussion about the topic of “AI potentially replacing humans.” Among these discussions, the viewpoints that “AI will replace a large number of human jobs” and “Google is facing its demise” are the most widely circulated, suggesting that the era of AI competing with humans is looming.
What makes ChatGPT able to attract one million users in just five days?
On December 5th, the founder of ChatGPT announced on Twitter that the number of ChatGPT users had officially surpassed one million.
To some extent, ChatGPT can be considered the most widely adopted AI application since the concept of AI was proposed. Sharing screenshots of conversations with ChatGPT on social media has become a trend. But what is driving so many people to interact with ChatGPT?
Analysis of ChatGPT’s structure reveals that it is a large pre-trained language model based on the Transformer algorithm. It can perform a wide range of common text output tasks, such as coding, recipe creation, homework assistance, novel writing, and commercial copywriting. Unlike other chatbots that use predefined responses or rules to generate text, ChatGPT can generate more natural and diverse responses based on the input it receives after being extensively trained on a vast amount of data.
ChatGPT is arguably the most advanced chatbot currently available, and its “smarter interaction” is the main reason for its global attention. For example, Open AI has added code comprehension and generation capabilities to this model, which greatly expands its range of applications. At the same time, ethical principles have been introduced, enabling ChatGPT to identify malicious information and refuse to provide valid answers, effectively enhancing the user’s interactive experience during conversations.
The term “artificial intelligence” was formally proposed in the summer of 1956, marking the official birth of the science of “artificial intelligence.” Since then, IBM’s “Deep Blue” supercomputer defeated the world chess champion, further expanding the imagination space of AI technology. It can be said that this science has been given the ambition to surpass human intelligence since its inception.
However, more than sixty years have passed since the concept was proposed, and ordinary people have not had the opportunity to directly interact with AI technology, nor have they had a suitable channel to understand this profound technology beyond various imaginations in science fiction films. Although various content platforms have already started using intelligent algorithm technology for personalized recommendations, and online shopping platforms have been using artificial intelligence technology to improve advertising reach rates, these applications have not yet enabled users to participate in the interaction process, and users have played only a passive role.
The release of ChatGPT provides users with an opportunity to participate in the AI technology workflow and offers a way for users who are not familiar with AI applications to truly experience the development of AI technology. This has a certain milestone significance in the development process of AI, which means that AI technology is moving from “behind the scenes” to “center stage.”
Data scarcity and high cost are common challenges for the AI industry
Multiple sources indicate that in less than two weeks since its launch, ChatGPT’s functionality has expanded beyond just being a chatbot, and many users are using it to assist with learning and work. In addition to user feedback, ChatGPT’s practical value has also been recognized by the capital market. Dongwu Securities released a research report stating, “Overall, ChatGPT’s dialogue effect is satisfactory, and its novelty and practical value have made it quickly popular.”
Given ChatGPT’s current level of intelligence and the overall level of development in the AI industry, it may follow a similar path to AlphaGo – garnering widespread attention before gradually fading away. Google’s AlphaGo sparked a nationwide AI fever back in 2016, but so far, there have been no mature AI commercial applications based on AlphaGo.
Analysis of ChatGPT’s current level of application and intelligence suggests that it still has a long way to go to replace human workers.
Firstly, ChatGPT’s completeness and level of intelligence are not yet sufficient to support mature thinking and independent work. There is ample evidence to suggest that ChatGPT’s strength lies in organizing seemingly logical and well-rounded answers, but there is no guarantee of the truth or accuracy of those answers. From this perspective, ChatGPT still requires a significant amount of machine learning and data input before it can replace human workers.
Secondly, ChatGPT’s database is still limited to 2021, and obtaining broader data licenses may face a variety of difficulties, including privacy and security concerns, economic issues, and technical challenges. Open AI explicitly states on its website that, as ChatGPT is not connected to the internet, it may occasionally produce incorrect answers. “It has limited knowledge of the world and events after 2021 and may occasionally generate harmful or biased content.”
Given the increasing stringency of privacy and data protection, acquiring large amounts of fresh data for model training presents significant policy risks and economic costs for Open AI. According to an article by the Chinese Academy of Sciences Institute of Physics, training a GPT-3 model conservatively estimated to cost between $5 million and $20 million, depending on the number of GPUs used, with more GPUs resulting in higher costs but shorter training times, and vice versa.
Furthermore, how to obtain large amounts of data for machine learning at low cost and with security is a general problem that the AI industry faces in development. Some people in the industry have proposed solutions such as privacy computing and blockchain, but in terms of technological maturity, these solutions cannot fundamentally solve the problem.
ChatGPT’s is not here to replace humans
Analyzing the current AI industry in China, it can be seen that the products that have been successfully implemented are highly concentrated in the To B industry, such as intelligent algorithm recommendations, smart cities, and intelligent monitoring. Apart from the intelligent algorithm recommendations that can continuously optimize models using their own platform’s massive user data, both smart cities and intelligent monitoring have relatively low data requirements, with their core being large computing power.
ChatGPT’s development direction is different from the aforementioned types. Some people currently believe that the appearance of ChatGPT will squeeze out the survival space of search engines, and as intelligent language generation models represented by ChatGPT continue to develop, search engines will become a thing of the past.
Analyzing the modes of search engines and ChatGPT shows that the idea of replacement is biased. Search engines are based on the aggregation of massive amounts of data, while ChatGPT is based on interactive text output, and the two have significant functional differences. However, search engines are still a direction for the development of ChatGPT.
With the increase in bandwidth and intelligent devices, online data scales are growing exponentially, which poses significant challenges for individuals and businesses alike. How can we find valuable information for ourselves in massive amounts of data? And how can businesses make reasonable decisions based on massive amounts of data?
In the background where not everyone can be a data scientist, ChatGPT, which has the ability to process massive amounts of information, output complete text, and further modify results according to instructions, has become a “bridge” to help individuals and businesses deal with the challenges of the big data era.
Although the current ChatGPT cannot fundamentally change our production methods, it represents that the development of artificial intelligence has entered a new stage. In this stage, users will no longer be passive recipients but rather active participants in the production process, and AI will also enter the lives of users at the practical level.