"The popularity of ChatGPT is declining," this is not nonsense, but data published by professional web analytics company Similarweb. In their recent data, in the first five months of 2023, ChatGPT's global visits increased by 131.6%, 62.5%, 55.8%, 12.6%, and 2.8% respectively, while in June, ChatGPT visits fell 9.7% month-on-month, the first decline since its launch last fall.
This data from Similarweb also reflects the popularity trend of ChatGPT on the Internet, as the fastest growing consumer application in history, only two months after the launch of ChatGPT, its monthly active users have exceeded 100 million at the end of January this year. At the beginning of this year, ChatGPT was widely discussed among netizens.
Although ChatGPT has cooled down at the consumer level, it does not seem to change the optimism of enterprises about generative AI.
Around the same time, IT solutions integrator Insight Enterprises and research firm The Harris Poll conducted a survey of 405 respondents aged 25 and older in full-time executive-level positions or higher at companies with more than 1,000 employees in the United States. The results show that 72% of business leaders from Fortune 500 companies plan to adopt generative AI in the next three years to improve employee productivity.
It's not just the leaders of overseas companies who are optimistic about generative AI, but also domestically. A few days ago, NetEase CEO Ding Lei said at the Zhejiang Platform Economy High-quality Development Conference that exploring the innovation and application of AI large models at the fastest speed is NetEase's top priority, while accelerating innovative applications, NetEase Fuxi Youling crowdsourcing platform will also provide 100,000 people with AI new careers this year. Behind such disparate data, is the artificial intelligence bubble coming again, or is the future of generative AI promising?
Today, the most suitable description of artificial intelligence should be "the future is bright, but the road is tortuous". Now that consumers are not interested in ChatGPT, Similarweb is not the first research agency to discover, and ChatGPT's popularity is even a public fact. The reason why consumers lose interest in ChatGPT is also very simple, at this stage it is indeed a bit exaggerated, its role is exaggerated by the outside world.
In fact, ChatGPT is not strong artificial intelligence (StrongAI), nor general artificial intelligence (AGI), but a large language model (LLM), a natural language processing model based on artificial intelligence technology, just because the parameter scale of ChatGPT is very large, so that the phenomenon of intelligent emergence, can process large-scale natural language text, has the ability to generate natural language, and obtains certain reasoning ability. On the whole, ChatGPT is actually just the result of a big brick fly, not a new breakthrough in the basic level of AI.
In essence, ChatGPT pushes deep learning to the extreme on large language models, but this is only the extreme of deep learning algorithms, not the extreme of artificial intelligence. ChatGPT and other LLMs can only generalize in large-scale data and deduce a subset of common problems, but as long as the user's input is relatively unfamiliar in the database, they can only start talking nonsense.
The limitations of ChatGPT mean that in practice, it is far less usable than everyone thinks, even if it is the best conversational bot ever, but nothing more. ChatGPT can indeed write copy, but do you dare to use the results directly in commercial scenarios? Obviously don't dare, so you'll still check the content of ChatGPT output to ensure that copyright risks are excluded and legal compliance. Similarly, ChatGPT can also write code, but programmers are afraid to deploy the generated code directly.
For example, there is news that in March this year, NetEase's internal departments, including NetEase Games, NetEase Cloud Music, NetEase Youdao and other departments, began to try AI to improve efficiency, access ChatGPT's API, and use ChatGPT for testing, but the results were mixed. In fact, due to the shortcomings of reasoning ability, ChatGPT's ability to produce content in complex scenarios is greatly limited. And most fatally, because some users use ChatGPT's networking function to bypass the paywall of other websites, OpenAI has urgently taken this function offline, further limiting ChatGPT's ability to obtain information.
So in the final analysis, the scenarios that can exert the power of ChatGPT are actually very limited, and after the user's novelty has passed, they naturally dare not be interested in it. The reason why enterprises value ChatGPT or generative AI is also very simple, even if they still have very big limitations at this stage, but replacing labor with machines is undoubtedly the most perfect solution for reducing costs and increasing efficiency. Therefore, when users value the experience of the present and enterprises look to the future, this is why generative AI is now hot.