Source: Wei Xi Zhibei
From 0 to 1 million users, Netflix took 3.5 years, Twitter took 24 months, Facebook took 10 months, Dropbox took 7 months, Spotify took 5 months, and Instagram took 2.5 months.
And ChatGPT only took 5 days.
That’s right, in the first week of ChatGPT’s birth, discussions about it in the tech world reached a climax, and its CEO soon announced that the number of users of ChatGPT exceeded 1 million.
However, just two weeks later, the discussion on ChatGPT in the industry has obviously declined, and the turbulent waves that had appeared at the beginning have gradually calmed down.
I observed a circle of opinion leaders and media who discussed ChatGPT very enthusiastically at the beginning, but now there is almost no discussion.
What I say is supported by data, and the cliff-like decline in the popularity of Baidu Index, WeChat Index, and Google Trends proves this:
Baidu Index
WeChat index
google trends
At the same time, in the past few days, there has been almost no response from the ChatGPT website, which also proves that its traffic may have returned to normal.
How to explain this phenomenon?
Is it because ChatGPT is not revolutionary enough? Is it because it is not friendly enough to mainland users? Is it because ChatGPT is actually useless?
Today, Wei Xi will seriously discuss this matter with everyone——
1. An explanation of the Hawthorne effect: People are always enthusiastic about short-term changes
First of all, I would like to say: when a new thing appears, people are enthusiastic about it in a short period of time, but it is normal for the enthusiasm to return to calm after the enthusiasm passes.
Let’s start with a factory management experiment in the 1930s——
From 1924 to 1927, at Western Electric’s Hawthorne plant in Cicero, Illinois, management decided to change the lighting conditions for workers who made relays in order to improve the productivity of workers.
It was found that whether the lights were dimmed or brightened, workers’ productivity increased in the short term, and even changes such as removing floor obstructions or relocating workstations could improve in the short term. productivity.
This phenomenon of short-term productivity gains due to different changes was later called the “Hawthorne effect”.
One explanation for the Hawthorne effect is that people will be motivated by the novelty brought about by the change.
Yes, people always like new things, and there is always excitement about change, but the excitement cannot last.
The same is true for ChatGPT, what has ChatGPT changed?
It has changed the stereotype of artificial intelligence established by the public by Xiaoai, Xiaoice, and Xiaodu, which are called artificial intelligence, but in fact are basically artificial mentally retarded products.
This change is huge, leap-forward, and far beyond expectations. It is not a gradual improvement, but a leap-forward improvement.
When something far exceeds one’s expectations, people’s dopamine will be secreted quickly, and the intuitive expression is “I’m going, I’m X, it’s awesome!”
But please note: the secretion of dopamine, a neurotransmitter, cannot last forever. It is only produced during changes and external stimuli, and it is no longer active in a steady state.
When a person uses ChatGPT for the first time, the subversive feeling is a huge stimulus, which comes from the huge surpassing of reality to expectations.
However, the brain quickly gets used to this stimulation and transcendence, because the change is no longer happening, and reality is once again in line with expectations, unless ChatGPT makes another leap forward.
The “novelty effect” in psychology is actually explaining the phenomenon of human beings who prefer the new and dislike the old.
Therefore, the decline in the discussion of ChatGPT is extremely normal. It is just an intuitive manifestation of the human species’ innate interest in new things.
The next thing we need to discuss is ChatGPT itself and its actual value.
2. Back to reality: the practical value of ChatGPT is not as big as you think
I once wrote an article called “Why don’t cool products catch on?” “The long article, there are two points of view explaining why many novelty products are not popular in the end:
1. Perceived coolness and user value cannot be equated.
2. The popularity of a product depends on the size and frequency of use of the product’s target audience.
In fact, it has been observed these days that there are two problems with ChatGPT at this stage – first, the actual user value is still relatively limited; second, there is an invisible threshold for use.
I believe that in the past few days, everyone has had a lot of dazzling cases about what ChatGPT can do. According to incomplete statistics——
It can at least write novels, poems, notifications, reviews, codes, weekly reports, business plans, outlines, research, and tests. ． ． ． ． ． It seems omnipotent.
But now let’s take a closer look, which of its various functions can generate actual value?
I did a small survey in the circle of friends today – “In the past two weeks, have you used ChatGPT to complete or help complete any actual work? If so, what is it?”
My circle of friends covering 26,000 people received only 21 valid replies in total, the details are as follows——
“Wrote some nonsense to be used in the tender, wrote homework for software engineering classes, polished daily papers, asked about the usage of basic functions in C language, wrote SQL statements, and edited manuscripts by oral broadcast” and so on.
This extremely personal research is of course not rigorous. After all, the threshold set by ChatGPT for mainland users is indeed a bit high, and there are also cases where others have used it but are unwilling to reply to me.
But one thing is certain: the people who “do not produce real value” are much higher than those who “solve real problems”.
This means that most people actually can’t think of using it again, at least objectively, it is not suitable for most ordinary people.
As for the reason for this problem, it is not that ChatGPT is not powerful enough, but that there is an invisible threshold for using ChatGPT itself. Users must learn to ask a specific good question.
The actual experience of many people is that they simply molested ChatGPT a few times like molesting Xiao Ai, asked a few relatively broad questions, found that they did not have the intelligence they imagined, and then threw it aside.
In fact, people who make good use of ChatGPT often ask very specific questions, and they will continue to ask step by step, or ask questions in different ways and get the final answer they want.
This is actually very similar to a search engine. The better you are at using a search engine, the more useful the search engine will be for you.
This invisible threshold prevents more ordinary users from getting the value they want from ChatGPT.
In other words, at least at this stage, ChatGPT’s actual effect on ordinary people is smaller than imagined.
For more people, ChatGPT is just a toy, and its only function is to post to Moments.
As an author, I have tried to let him assist in writing, but because the field is relatively vertical, the actual effect is very unsatisfactory. My intuitive feeling is——
It can provide some very basic texts, but these texts are too general without real insight to be useful.
For a technical writer, if the article has no insights, it is meaningless. As for many people using it to expand, abbreviate, proofread, and extract key points, I still think this function is better than nothing due to style reasons.
What’s more deadly is that, as many people have mentioned, ChatGPT still has a very obvious shortcoming – factual mistakes often occur.
This means that it may not always provide the correct answer.
In part of the answer to the question, it’s deadpan nonsense.
I asked it what famous words did Buffett say? It answers some chicken soup that is obviously not what Buffett said.
I asked it who the author of the novel “Foundation” was, and it said it was George Martin. (This is actually quite unbelievable)
I asked it what are the specialties in Shenzhen, and it answered confidently, stinky tofu, white-cut chicken and beer duck.
Teacher Ruan Yifeng asked it what attractions it had in a small town in Yunnan, and it made up a lot of attractions that didn’t exist:
It is for this reason that the programmer question-and-answer website Stack Overflow has banned users from sharing content generated by ChatGPT since December 5.
So, what practical uses can ChatGPT produce at this stage?
In the following case, you may get your own answer –
I know a grassroots brother who specializes in online earning in Heze, Shandong. Let’s experience his usage of ChatGPT. He uses ChatGPT to generate——
“What should I do if there is an unreasonable and evil mother-in-law?”, “How to persuade my girlfriend to accept the first time?”, “My husband cheated, how should I get it back?”, “How to chase a divorced woman?”, “How to persuade Older son getting married?” Questions like this.
Generate text in batches, then use AI tools to turn the text into speech, add some video clips to vibrate, and simultaneously send the text version to Toutiao, Baijiahao and Penguinhao.
Don’t tell me, ChatGPT is very good at answering these questions. Many answers are comprehensive, well-founded, and have the quality of a textbook, which can be described as excellent chicken soup——
He told me that the current data and income are not very good, but there has been a positive cycle gradually.
He said that the most useful part of ChatGPT for him is: you enter the same question, and its answer is different every time, which can perfectly avoid the platform’s duplicate checking algorithm.
An important message we can get from this case is that the text generated by ChatGPT is very suitable for popular and popular content.
The logic behind this is actually very easy to understand. After all, it has the largest amount of data in training for popular content.
However, when it comes to practical problems of being too specialized, in many cases it stops working.
That is to say, it is very good at providing “big street goods”, but it is not good at once encountering extremely specific vertical problems.
In addition, the real-time nature of the content and the bias and discrimination in the answers will also bring a less than ideal experience to ordinary users who want to get actual value.
3. Pandora’s Box has been opened: ChatGPT is just an appetizer
If you think that this article of mine is just complaining about how useless ChatGPT is at this stage, then you are wrong.
Let us briefly review the history of the AlphaGo war against human Go——
In September 2016, after AlphaGo defeated the European Go champion, many technical leaders, including Kai-fu Lee, believed that there was little hope for AlphaGo to further defeat the world champion Lee Sedol.
But everyone has seen the results later. After only 6 months, AlphaGo easily defeated Li Shishi, and there was no defeat after losing one game. This evolutionary speed is jaw-dropping.
The evolution speed of AlphaGo may be staged again on ChatGPT.
The current ChatGPT is created based on OpenAI’s GPT3.5 model. Since 2018, the parameters of GPT1, GPT2, and GPT3 are 117 million, 1.5 billion, and 175 billion, respectively.
This is an exponential growth, and it is conceivable how the more powerful GPT4 will reach.
Therefore, although it is undeniable that ChatGPT does have many limitations at this stage, this conclusion is not what I said, but the CEO of its creation team OpenAI said——
But this does not affect its iterative version to continuously solve and weaken these problems in the evolution process. In this sense, ChatGPT may be just a starter.
I hope in the future it will be very easy to—
Help financial professionals to query professional data in various dimensions extremely intelligently;
Help legal professionals to search for precedents and clauses in various dimensions intelligently;
Help advertisers to create really usable slogans extremely intelligently;
Help musicians create truly popular music;
…
These different AIs claim to be able to do it, but it is obvious that it is not good enough at present, but there is no doubt that the products represented by ChatGPT have opened Pandora’s box for AI.
I very much agree with what the famous Silicon Valley investor Paul Graham said on Twitter——
“The amazing thing about ChatGPT is not how many people are shocked by it, but who is shocked by it. These people are not casually excited about every new thing. Obviously, something big is happening.”
For each of us, it is indeed time to think about whether our work will be replaced——
Kai-fu Lee once mentioned a point of view – the work of thinking for no more than 5 seconds will definitely be replaced by AI in the future.
Looking at it now, in some fields, ChatGPT has far exceeded the standard of “thinking for 5 seconds”, and with its crazy evolution, its possible potential will exceed the imagination of many people.
The singularity is looming, and the future is here, it’s just not evenly distributed.
Finally, I end with a sentence from Kevin Kelly——
“It took only 58 years from the first chatbot (ELIZA, 1964) to a truly effective chatbot (ChatGPT, 2022).
Therefore, don’t think that the vision at a close distance must be clear, and at the same time, don’t think that it must be impossible at a long distance. ”
(Disclaimer: This article only represents the author’s point of view, not the position of Sina.com.)
This article is reproduced from: https://finance.sina.com.cn/tech/csj/2023-01-03/doc-imxywnct5903140.shtml
This site is only for collection, and the copyright belongs to the original author.