”Certainty meets people’s basic needs,” this is what ChatGPT told me. As always, it fails to give a basis for this statement. However, since this is consistent with our feelings in daily life, we will not study it in detail.
As a person who creates things, I don’t really like certainty. At least that’s the theory. Only with uncertainty can we innovate, have opportunities, and be free. Once everything is settled, it becomes lifeless.
Therefore, when I said that I would buy a house in a school district for my children in 2020, I felt very contradictory. The selling point of that house is that it not only guarantees that the children will go to the right elementary school, but also has a high probability of going straight to the right junior high school. I was thinking at that time, what if on the first day of my first grade in elementary school, my father told me that you would go to elementary school here for the next 6 years, and then go to the middle school opposite for another 3 years – give it all to you. arranged! How boring that would be.
What we are doing now is to explore the role that AI technology represented by large language models can play in “careful reading”. This work itself did not exist at all a year ago, and the large language model it relied on did not exist a year ago. With our vision at the time, we could not imagine its possibility, let alone any planning. After all, we must admit that it was the release of ChatGPT that inspired us, and ChatGPT did not exist a year ago today (when writing this article).
This may sound insecure, but I think it’s quite exciting.
In one year, ChatGPT has grown from scratch to a product with more than 100 million weekly active users, according to the figures announced by OpenAI at the developer conference in early November. It is considered to be history. The fastest-growing consumer-facing Internet product on the Internet (Meta’s Thread may rival it). Within a year, ChatGPT’s simple design became an industry benchmark, and it was copied by products from other major manufacturers such as Bard. At the same time, its own iteration speed is also very fast. Open the interface of ChatGPT today. Not only can you use the latest GPT-4, which can input about 300 pages of text at a time (30 times more than when it was first released), but you can also upload images and files in the same interface for GPT to interpret and generate. Pictures, chatting with a synthetic voice that looks exactly like Fei Xiang…
This year once again proved that the evolution speed of human technology is non-linear and exponential, and not only in technology, but also in all aspects of business and capital evolution is accelerating. When I stayed up late to watch the live broadcast of the OpenAI Developer Conference in early November, I remembered that in the past 10 years, I looked forward to Apple or Google’s developer conference with a similar mood every year to see what new technologies were available on iPhone and Android and could be used. Come and make new products and new functions. In just one year, it can be said that OpenAI has become the most watched technology company in Silicon Valley.
What I believe is that there should be a distant goal on the horizon, as long as you make sure you are probably heading in the right direction.
Even the company’s “coup” has accelerated. When I wrote this paragraph, it was also during the intermission of the dramatic plot of OpenAI firing its CEO. Some say it took 12 years for Jobs to return from being ousted by the board of directors. It only took Sam Altman two working days. This itself is related to the non-linear evolution of technology. Will technological evolution get out of control if it accelerates to a certain extent? One speculation is that OpenAI’s board of directors was worried about threatening human survival because OpenAI had made some kind of “breakthrough progress” in early November, so it had no choice but to try to expel Altman, who ignored safety and insisted on “effective acceleration.” Helen Toner, a former member of the board of directors, said in a communication with executives that destroying the company was also “in line with OpenAI’s mission.” At present, Altmann has returned to the company, and Toner and other directors have resigned. As the outside world, we still don’t know what the “breakthrough” code-named Q* is. Is Altman the desperate “scientific madman” in science fiction movies or the savior of human destiny? Not sure yet.
But people always have the illusion that they can control everything. It is also human nature to want to be able to predict and control the future. When playing Ludo with my child, he always cries when the dice cannot be rolled as he wants. I have to help him understand “probability” or “luck” and understand that the number of dice rolled is generally beyond his control. , even if this explanation is not easy to accept.
People are willing to pay a high price for certainty, which is what most people pursue. We have known this since childhood. It is another similar word called “stability”. A stable job, a stable family, a stable residence—this is the beautiful life we have been familiar with and imagined since childhood.
I guess a lot of metaphysics comes from this. When a small probability event happens to oneself, and it is often a bad thing, people always want to get an explanation and want to know how to avoid it next time. But in fact, most questions have no answers. In my opinion, one of the biggest differences between science and metaphysics is that science will admit that we still cannot understand most things in the world, especially how little we know about ourselves.
So I don’t like to prepare too much for certain possibilities in the future. When I was studying at Peking University, I was in the experimental class. At that time, I enjoyed the “privilege” of the whole school to choose courses freely and arrange my own teaching plan (it should be very common now). Some students take advantage of this degree of freedom to optimize their GPA from the beginning of their freshman year, and choose some classes that are easily assigned by teachers, in order to have some advantages in pursuing graduate studies or going abroad. This is something I scoff at. The current happiness comes from learning what you like, rather than preparing for the performance ranking three years later, which is boring.
It doesn’t mean to give up all efforts for the future. What I believe is that there should be a distant goal on the horizon, as long as we are probably heading in the right direction. As for how much we walk today and where we go tomorrow, we have to do our best and let fate dictate. The vision is certain, but the path is unknown. Just like although the emergence of AI is part of the uncertainty for us, “reading carefully” is a certainty for us.
There is an interesting parameter in the GPT interface called temperature, which can be understood as used to adjust the uncertainty of the output. The higher the “temperature”, the higher the uncertainty of the answer and the stronger the creativity; the lower the “temperature”, the lower the certainty. In fact, just like setting goals, the answer may not change, but how to answer the question can change.
Sometimes some certainty may still be needed. Today I was thinking about it with my colleagues, why haven’t we released our product yet? When I restart my work in July, I plan to make a prototype a week, and then spend another month turning a reliable prototype into a releaseable product. The first prototype actually took six weeks; turning it into a product has taken two and a half months, and it is estimated that it will take at least half a month. Of course, this is also affected by many uncertainties. As a reorganized small micro team, any personal changes will have an immediate impact on the overall work; AI applications have different development models and require constant adjustment to find the best approach. This is also a bit like metaphysics. Your input and There is not necessarily a strong correlation between the outputs, and the uncertainty is relatively large. The methods of design, testing, and optimization during the development process are different from those in the past, which also leads to the need to adjust the way of working.
OpenAI directly abandoned the plug-in system announced in April at the developer conference in early November, and instead launched the GPTs platform, allowing third parties to publish their own GPT-based software on ChatGPT. Chatbot. This suddenly replaced many products, tools and platforms on the market. I was also a little frustrated, especially after trying out the new API. It’s not that the product we are working on has been replaced by ChatGPT. It still cannot be replaced and will not be replaced. But GPT’s new interface can greatly simplify the development process. We wrote something for a month or two, and now it seems that we can use GPTs to complete a 60-minute replica in one or two hours. In other words, the production threshold is lower, which weakens our first-mover advantage, which is equivalent to wasting some time. At that moment, I still complained that I was doing it too slowly.
But if we think carefully about the development process of our products so far, if we travel back in time with today’s knowledge, we will certainly know that there are many places where we can speed up. But our goal has not changed. Every line of code we write is certain, every detour we take is certain, and every lesson we learn is certain. Having said that, I feel that we are moving forward step by step with certainty, although the exact time is not yet certain.
AI has rekindled people’s passion for creation, and this passion is certain. This enthusiasm is both in oneself and in others. This year I have met many young people who are riding the wave of AI to do things. I feel like I haven’t seen this kind of passion for creation sparkling in their eyes for a long time (in fact, there are also young people who are riding the wave of Vision Pro to do XR, and they also have sparkling eyes. luminescent).
I finally bought that house. As parents, it seems that we still want to give our children more certainty. Probably because I don’t want to impose my own values on others. In fact, it is understandable why the previous generation pursued stability, because stability has been far from the norm historically. Our ability to pursue instability today is also a luxury.