ChatGPT is so popular, OpenAI itself did not expect it.
According to the latest disclosure from the New York Times, the next generation of GPT-4 was almost ready within OpenAI and was just waiting to be released.
In mid-November, senior management suddenly announced a new task: to release a free chatbot application in two weeks, codenamed "Chat with GPT3.5".
A group of technicians made a temporary U-turn, but even some of them doubted whether it really made sense.
In terms of underlying technology, they have been exposed to the most advanced GPT4 every day and have become numb. To them, GPT3.5 is an old thing from two years ago.
In terms of product form, no one paid attention to the chatbot released by Meta, another scientific text generation model, and Galactica was removed from the shelves after being criticized for being inaccurate within 3 days.
Informed sources revealed that the idea of OpenAI's senior management was originally very simple:
The AI painting that became popular last year started with DALL·E released by OpenAI, but due to copyright, ethics and other concerns, it was chosen After entering the internal beta invitation system, the free and open source Stable Diffusion stole most of the limelight.
Regarding chatbots, they are worried about a similar situation happening and being preempted by other companies. In addition, launching a product quickly using an old model can help them gather feedback to improve the new model.
In just 13 days, ChatGPT was officially born.
What happened next was beyond everyone’s expectations.
Users exceeded one million in 5 days, and active users exceeded 100 million in 2 months. ChatGPT has become the fastest growing consumer application in history.
Students use it to write homework, workers use it to write weekly reports, programmers use it to write code...
But the significance of ChatGPT seems to be not limited to Consumer-level applications are so simple.
Merchants and brands also found it "really good" after trying it. For example, real estate agencies in the United States used it to generate listing copywriting, and the original one-hour workload was shortened to 5 seconds.
The heat is coming so fast, like a tornado, and OpenAI itself is afraid.
OpenAI President Greg Brockman once celebrated the 2 million user milestone on Twitter. As CEO, Sam Altman believed that such publicity was unwise and asked him to delete it.
Altman worries that excessive hype may trigger a regulatory backlash or lead to users having too high expectations for future versions.
Although Altman wants to suppress the heat, the outside world continues to boil.
Microsoft has made additional investments, Google is facing a formidable enemy, and the stock prices of various related and unrelated AI companies have skyrocketed.
Bill Gates praised in an interview with Forbes: The significance of ChatGPT is no less than the birth of PC and the Internet.
From this perspective, what is truly unique about ChatGPT is that it changes the mode of interaction between humans and machines.
Answering follow-up questions in context, admitting mistakes, questioning incorrect premises and rejecting inappropriate requests were the three initial publicity points when ChatGPT was released.
ChatGPT is far from perfect, its mathematical calculations are inaccurate, it often talks serious nonsense, and it doesn’t know anything new after 21 years.
#But people are still willing to try it, interact with it, and generate billions of words of text every day.
Peter Welinder, Vice President of Products at OpenAI, recently revealed that when they first started testing the GPT-3 API interface three years ago, they could hardly find anyone willing to try it.
The influence of GPT-3 Playground, which was opened as a demo, is limited to the technical circle.
Fortune magazine compared ChatGPT to the iPod and iPhone: There were MP3 players before Apple released the iPod, but it did not trigger a digital music revolution; there were other smart phones before the iPhone Mobile phones, but no application ecosystem has been formed.
Xiaoice CEO Li Di once shared his insights at the MEET2023 Intelligent Future Conference held by Qubit:
ChatGPT improves the behavior of AI, even if the generated content is not accurate enough , people will also find it interesting.
Conversation is a more natural interaction mode than keyword search. It makes AI no longer cold. While achieving high concurrency (responding to multiple users at the same time), it also achieves a high conversion rate (satisfying users). new needs), it became an epoch-making product.
Today, the ecology and competition surrounding ChatGPT and the AIGC (AI Content Generation) it represents can be divided into four major levels.
The first layer is the AIGC model and the service itself.
Google’s own LaMDA is being tested internally, and it has also invested in Anthropic, a company spun off from OpenAI, in a two-pronged approach to compete with Microsoft.
In China, the Guangdong-Hong Kong-Macao Greater Bay Area Digital Economy Institute (IDEA) Cognitive Computing and Natural Language Research Center has the largest open source model system in China - Fengshenbang. The 98 open source models are fully supporting the development and exploration of China's AIGC. Develop personalized models adapted to various fields.
Dr. Zhang Jiaxing, head of the center, once said, “In the current and future development of AI, a model can be an important part of creating a product, and even a model can give birth to an industry, such as the recently popular Stable. Diffusion (Diffusion Model) is an important milestone in promoting the rapid development of the AIGC industry. It can be said that the continuous manufacturing and progress of the underlying model is an important factor in promoting the progress of the entire AI industry."
And Baidu will also be in 3 Launched its own generative search product in March. Regarding this point, Baidu once revealed to Qubit: The final product will be very different from ChatGPT, that is, although it will be multiple rounds of interaction, it will not be a simple multiple rounds of dialogue.
The second layer is to integrate AIGC capabilities into existing products.
Microsoft announced that all of its products will integrate ChatGPT. The video conferencing tool Teams has been launched and can automatically generate meeting records and predict next tasks.
Google is also about to hold its latest press conference on February 8, and outsiders speculate that it may be related to AI products.
The third level is to use AIGC capabilities to develop new product types.
DoNotPay uses ChatGPT as an AI lawyer to help people handle fines and bills.
Character.ai, founded by a Google engineer, allows chatbots to imitate historical figures, real-life celebrities or virtual characters to talk to users.
#A number of newcomers have also emerged in this domestic model, such as Glow, which was once on the hot search list.
The last layer is content generated directly or indirectly using AI.
Startup ProfilePicture.AI used ChatGPT and GitHub Copilot to generate the main code for the product.
#Even domestic fast-moving companies have begun to try new things and will find partners in the technology circle to explore together.
For example, China Merchants Bank Credit Card has just released a brand manuscript generated using ChatGPT. It cooperated with Dr. Zhu Mingjie of Krypton Information Technology Co., Ltd. and others to complete the first promotional manuscript generated by AIGC technology in the domestic financial industry.
Dr. Zhu Mingjie said that the emergence of ChatGPT at the end of last year shocked him. Even as a witness to the founding of OpenAI in the early years, he now sees that AI has really learned human skills. Knowledge still feels beyond imagination.
In 2017, not long after OpenAI was established, we went to Silicon Valley for a visit and exchange. They believed that the direction of artificial intelligence that could comply with Musk's first principles was to train an AI that could learn all human knowledge. , to solve the knowledge challenges faced by mankind.
What else can’t be achieved? At this time, companies like CMB Credit Card that dare to take the first step to explore and embrace the future will, I believe, get the upper hand.
The current ChatGPT technology alone is not enough to support the long-term development of the above model.
In addition to the inaccurate content generation, many early adopters also found that "the model cannot be customized" is a major pain point in the current application of ChatGPT.
Some applications in vertical industries still use the old GPT-3 (although they are now willing to blur the difference between the two in publicity).
For example, the startup DoNotPay is a robot lawyer that uses AI to help people deal with fines and billing issues. In order for AI to understand legal-specific knowledge, secondary development must be done on the basis of a general language model.
This is the first attempt of China Merchants Bank Credit Card. The output process can be summarized as:
For example, the team first asked ChatGPT to "explain the relationship between genes and family affection, quoting the views of "The Selfish Gene":
Judging from the results of the first step, ChatGPT’s answer contains background knowledge from the book "The Selfish Gene"; but from the perspective of language description, it is relatively straightforward. That kind of thing. Therefore, the team further put forward requirements for it:
Write it more deeply and be thoughtful.
It is not difficult to see that ChatGPT is based on the first round of Q&A and is refined according to new requirements. At this time, ChatGPT has come to two relatively clear views, one is that "life is the carrier of genes", and the other is "family affection is altruistic love."
Further, the team continued to make requests:
You can be more specific here, and it would be better to give an example. Think carefully about how family love exists beyond the biological nature of genes through the carrier of life, and it must be written elegantly and gracefully.
At this point, the core content of the final promotional draft of China Merchants Bank has been generated.
However, team members also reminded that in the process of "tuning" ChatGPT, we must think about the outline of the final answer in advance, and then guide it, otherwise it may happen in the future. During the generation process, I was led astray by certain results.
In addition, due to the randomness of the large model itself, if we are not very satisfied with the answer we get for the first time, we can try to repeat the question until we get a satisfactory answer.
It can be seen that the copywriting generated by AI is not the top in the industry, but it can also reach the middle level. In the future, some of the most basic tasks may really be replaced by AI.
And the strength of AI writing articles lies in cross-disciplinary and cross-field, surpassing the creative ability of a single person. However, there is no need to worry too much. For jobs with higher creative requirements, mastering AI will greatly improve work efficiency.
At least for now, humans still need to write prompt words to assign tasks to AI. People who are proficient in this will get new opportunities.
For example, Riley Goodside, who shared his experience using ChatGPT on the Internet, was hired by Scale AI as the world's first prompt word engineer with an annual salary of 2 million.
The same is true for enterprises. In addition to the fierce competition between technology giants Google and Microsoft, and AI startups seizing the opportunity, other industries should also actively embrace changes like China Merchants Bank Credit Card in order to seize the opportunity.
Finally, everyone can guess the following two poems, which one is created by humans and which one is created by ChatGPT?
A,
In the beginning, family affection was a common set of DNA. You were in the delivery room and I was outside.
Later, family relationships are like amusement parks, with you at the top and me at the bottom.
When I grow up, family love is a series of tests. You are in the examination room and I am outside.
But now, family affection is a "family credit card". You are far away and I am in my hometown.
The adverse journey of life, family love is priceless
B,
Life is nurtured, and family love is a thin DNA chain. You are at this end, and your parents are at the other end.
When you grow up, family affection is like a precious family photo album. The past is inside and you are outside.
Later, family love is an infinitely warm telephone line, caring at one end and embracing at the other end.
But now, family affection is a warm family card. The family is at home and you are in the sky.
The above is the detailed content of Here it comes, ChatGPT is starting to grab jobs in the financial circle. For more information, please follow other related articles on the PHP Chinese website!
![[Web front-end] Node.js quick start](https://img.php.cn/upload/course/000/000/067/662b5d34ba7c0227.png)
