I wrote an article about ChatGPT when it came out in December last year, and today I will talk about it again.
(1) High-quality data sources
I am doing an important research during this period, but unfortunately, fake information is flying everywhere on the Chinese Internet, and valuable data is not made public. The only thing I can rely on is the slightly credible data source of financial reports of listed companies.
When I research some benchmark companies, I often use industry contacts "inside sources" to understand the true ins and outs of some measures and results of benchmark companies, but unfortunately, these so-called "inside sources The understanding and understanding of the real situation of some of the company's measures and results is also a mess. Because I am a professional researcher and not for writing self-media articles, I can say whatever I want, so I often use the five-level questioning method, but often When asked about the third level, the "insider information professional" could no longer answer. His understanding of the matter and the logic of his understanding also contradicted himself and could not be explained.
But I also understand this phenomenon, which is normal. Because most people (including executives and even the company’s boss and founder) are busy with research and development, sales and delivery every day, version after version, customer after customer, and many decisions are made on and off the desktop. Mutual blocking, decision-making and results based on the current situation, so many things have no logic and no clear goals.
Therefore, in terms of professional research, my requirements for my research team below are:
It is forbidden to check Baidu. If you want to check a search engine, you should check Google, and you must check English Google. Moreover, artificial intelligence-machine translation is now so mature.
It is forbidden to use online content as research evidence. If you want to read online content, you should go to the official websites of the government and enterprises.
It is recommended to go to Dinghui Dingjournal to find literature and build a reference index.
Yesterday I heard that Baidu is going to follow suit and launch a Chinese version of ChatGPT. I was thinking about one thing: if you input garbage, you can get gold output?
(2) open
The Web Internet is a good thing:
HTML
URL
Link
HTTP
For the mobile Internet, Apple made a bad start, and now the mobile Internet is all App technology:
It is difficult to connect and jump between applications, all of which are isolated
The content is not open, all closed in the application
I think of the WeChat App, the content can only be searched by itself, and cannot be accessed by other apps.
ChatGPT crawls the high-quality data sources of the StackOverflow website. ChatGPT became popular this year, programmers don’t need to go to StackOverflow to find answers to questions, they can get the best answer with ChatGPT, so StackOverflow quickly disappeared 2 million monthly active users. So StackOverflow blocked ChatGPT, preventing ChatGPT from crawling its own information.
I remembered the mutual closure of Taobao and Baidu back then.
I was thinking that China and the United States are anti-globalization, and the mobile Internet is also anti-globalization-isolating and closing. This AI has pushed the isolation and closure a step further. This is what I don't want to see.
(3) Interaction
This time ChatGPT was able to go out of the circle, it really depends on having a dialog box that can directly ask and answer conversations. When ChatGPT first came out, I, like people in the industry, thought that ChatGPT was GPT3 wrapped in a shell. When GPT3 appeared in 2020, it was only reported by the industry for a few days before it was silent. But ChatGPT is out of the circle.
I wrote an article in the past about:
Demand-side amplification: interaction innovation-direct interaction-interaction frequency-interaction content-interaction value
Supply-side amplification: opening up, introducing-connecting-integrating more providers to meet the huge variety of user groups on the enlarged demand side
The reason why the popularity of PCs exploded is because the graphical and mouse-based forms of Windows replaced the character-based and command-based forms of DOS. The Windows API is open, replacing the difficult-to-use interrupts of DOS.
The Internet was born in 1969, why did it explode in 1993? It was because FTP was popular in colleges and universities in the 1970s, and Email was popular in the business world in the 1980s. It was not until the invention of HTML in 1990 and the invention of the HTML browser in 1992 that the Internet exploded among the general public. HTTP and Link of the Web Internet have surpassed the dedicated closed protocols of FTP and SMTP in the past.
Nokia and Motorola launched Symbian Smart phones and Linux Smart phones in 2001, but in 2007 iPhone Smart phones came out because the iPhone replaced the past point-touch interaction technology with multi-touch interaction technology. Moreover, the opening of the application store has surpassed the difficult application search-download-installation of Symbian mobile phones in the past.
Therefore, the innovation of interaction is a message tree for the real explosion of technology. Without interactive innovation, technology is small circle technology.
So I have also written in the past that if Ethereum wants to get out of the circle, not relying on digital currency to get out of the circle (ETH2.0 in September 2022, POS has replaced POW), it must have innovative breakthroughs in interactive technology to really explode and popularize. A few days ago, NoStr became popular a few days ago. In fact, it was exploring the interactive technology of Ethereum, but sorry, the exploration failed.
(4) study
A certain CTO group is also talking about ChatGPT during this time, but the content of the cold talk is outrageous, what level are they talking about:
Transfer some self-media articles
Turn some venture capital analysts' reports
Turn some jokes that make fun of ChatGPT
I want to discuss three questions:
1. ChatGPT principle: what is the backbone network architecture. In addition to massive & high-quality data sources, manual annotation and manual reinforcement learning feedback correction training, what innovations and progress has ChatGPT made at the grassroots level? Just like when NoStr became popular a few days ago, someone asked me what I think about NoStr. I said that I didn’t know much about NoStr. Is the number of paths correct?
2. The ChatGPT API is open, so let’s discuss what innovative applications can be done with ChatGPT? In 2015, when artificial intelligence was booming, I said that apart from search, promotion and vision, I couldn’t think of anything else artificial intelligence could do. Sure enough, 8 years have passed, and there are indeed no eye-catching applications in AI applications at home and abroad. Think about AI, something with such a high technical threshold and such a large resource investment, that the output is so low.
3. Bing, like Google, used to rely on affiliate ads to survive. Now New Bing has called ChatGPT, which has become a conversational question-and-answer style, and only one best answer appears (although it is linked to the original data), but what is the commercial realization of New Bing? It seems difficult to survive by advertising, what is the new commercial realization? Many friends said that it is charged by subscription. I am more skeptical. After all, there are only a small number of people who want to spend money to ask questions and find answers, and a small number of people cannot make a search giant move forward.
Unfortunately, no one is interested in discussing these three issues.
I was thinking, this is still a top-level technical group. As far as this virtue is concerned, the commercial application of AI in China - commercial realization - reappearance and reinvention of the large model wheel - innovation and improvement - breakthrough, the road is long.
(5) Commercial Realization
In 2018, I decided on three intelligences:
Intelligent hardware (Smart Device)
Smart Contract
artificial intelligence
In terms of smart hardware:
Now that Microsoft and FB are laying off VR R&D personnel, FB will definitely turn around this year and enter the AI field to compete with Google, Microsoft & OpenAI. After all, FB still holds the AI platform Pytorch, while Google and OpenAI compete with each other in model innovation. OpenAI is currently leading in large-scale model engineering innovation, and Microsoft is rapidly embedding OpenAI's engineering achievements to enhance productization.
At present, Silicon Valley venture capital has also withdrawn its investment in Metaverse and VR, which has delayed the emergence of Metaverse (next-generation applications) and VR (next-generation computing platforms). This is what I don't want to see.
In addition, Google's new operating systems that I am most concerned about: Fuchsia, Kata, I heard that they are also being abolished. This is what I don't want to see.
In terms of smart contracts:
Although the ETH merger is completed, I know that performance optimization is a long way to go, just like Java performance optimization back then, it is a continuous improvement. It cannot be OK with the completion of the ETH merger. Various rollups are just one of various performance optimization solutions.
It is also a good thing that POS has replaced POW. At present, I am paying close attention to the commercial realization of Ethereum applications and the impact on chips.
After V God merged with ETH, he started to work on the unified soul account. This is something that should be done and must be done with a high priority. But I think smart contract programming languages, virtual machines, oracles, and interactive access technologies may also be important. Therefore, the popularity of Ethereum-smart contracts-commercial applications has exploded, and this road is also a long way.
Finally, we talk about the focus of our article: artificial intelligence (artificial intelligence).
If it weren't for the recent popularity of ChatGPT, I would have almost shaken my confidence in the artificial intelligence research I set up 5 years ago. I heard that in the past few months, Silicon Valley VCs are not interested in metaverse-VR, Ethereum-digital currency-Web3, and they have begun to look at the AI track again.
In addition to the AI embedded in Soubantui that we discussed 8 years ago to enhance Soubantui and realize monetization by Soubantui, what other ways are there for AI to monetize? It has been 8 years, and it has not appeared at home or abroad. The only one who benefits may be Nvidia, which is a dry chip, and no one else has benefited (A16z, a professional venture capital investor in the digital intelligence track, seems to have this view).
I always said in the past: the demand that does not pay the bill is a pseudo-demand. Whether the wool comes from sheep or the wool from dogs, someone has to pay.
How artificial intelligence can be monetized commercially, not just monetized on search, promotion and promotion, is what I am most concerned about.
