GuShiio.com Article source: AI Pioneer Officer
Image source: Generated by AI
For most of human history, hiring a dozen experts with doctoral degrees often required huge budgets and months of preparation time. Nowadays, you can instantly gain the wisdom of these “brains” by simply entering a few keywords into a chat robot.
As intelligence becomes cheaper and faster, the basic assumption that underpins our social system-“human insights are scarce and expensive”-will no longer exist。When we can call on the insights of a dozen experts at any time, how will the company’s organizational structure change? How will our innovative approach evolve? How should each of us approach learning and decision-making? The question for individuals and businesses is: How will you act when intelligence itself is available everywhere and at almost no cost?
The historical process of smart “price reduction”
More than once in history, we have witnessed a significant decline in the cost of knowledge and a rapid expansion of communication channels.The emergence of printing presses in the mid-15th century greatly reduced the cost of disseminating written materials. Previously, texts were often copied manually by professionals such as monks, which was costly and time-consuming.
When this bottleneck was broken, Europe ushered in profound social changes: the Protestant Reformation had a huge impact on the religious level; literacy rates rose rapidly (laying the foundation for universal primary education); and scientific research flourished with the help of printed publications. Business-oriented countries such as the Netherlands and the United Kingdom benefited greatly. The Netherlands entered a “golden age” while the United Kingdom continued to play an important role on the global stage for centuries that followed.
Over time, the popularization of mass literacy and public education has improved the overall wisdom of society, which has also laid the foundation for industrialization. Factory jobs are becoming increasingly specialized, and a more complex division of labor drives economic growth. At the end of the 18th century, countries with high male literacy rates were the first to industrialize; by the end of the 19th century, the most technologically advanced economies tended to have the highest literacy rates. People master new skills and create more professional positions, forming a virtuous cycle that continues to this day.
The emergence of the Internet has pushed this trend to new heights. When I was a child, if I wanted to study a new topic, I would need to take my notes to the library to search for books. This step alone consumed most of the day. At that time, acquiring knowledge was expensive and difficult.
Today, artificial intelligence has taken over this thousand-year-old baton of “reducing the cost of intelligence”, opening a new chapter in our economy and way of thinking.
My “epiphany moment” with ChatGPT
When I first used ChatGPT in December 2022, I felt that it was a milestone product. At first, I just used it to do some “number tricks”, such as having AI “rewrite the Declaration of Independence in Eminem’s style”(the adaptation it wrote was probably “Yo, we’re going to say it out loud, people here will never be knocked down”, etc.).
In hindsight, it was like asking a Cordon Cordon Bleu chef to bake a cheese sandwich for you. It was too overkill. It wasn’t until one afternoon in January 2023, when my 12-year-old daughter and I spent a few hours designing a brand new board game with ChatGPT that we truly realized the power of such tools.
At that time, I first told AI which board games we liked and disliked, and asked it to analyze the commonalities among them. It found that we like game mechanisms that can “lay paths”,”manage resources”,”collect cards”,”formulate strategies” and “have a high suspense of victory and defeat”, and we also dislike some patterns common in Risk or Monopoly.
I asked it to come up with some less obvious but important game ideas based on these elements, and hoped to have a certain historical background. ChatGPT came up with a game called “Elemental Discoveries”: players play as chemical researchers from the 18th and 19th centuries, collecting and trading resources to conduct experiments, gain points, and interfere with each other and destroy each other.
Then, I asked it to further refine resources, gameplay, game mechanics, and suitable roles for players to play. It puts forward positions such as “alchemist”,”destroyer”,”businessman” and “scientist”, and also matches them with historical chemists such as Lavoisier, Joseph-Louis Guy-Lussac, Marie Curie, Carl William Scheler, etc.
With the help of ChatGPT, which was relatively “elementary” at the time, we created a rough but playable board game in just two or three hours. In the end, I had to stop. On the one hand, there was not enough time, and on the other hand, I was exhausted. That experience made me realize firsthand thatAI “collaborators” can compress a research and development process that would otherwise take weeks into just a few hours.Think about the huge potential it will bring if it is used for product development, market analysis, and even corporate strategy?
In this process, the ChatGPT I saw was not just repeating or piling up facts; its performance demonstrated the ability of analogy and conceptual thinking, able to connect ideas with realistic references, and truly output creative solutions on demand.
From “random parrot” to “deep thinker”
One trillion is already an astonishing order of magnitude. The large language models that support ChatGPT often have billions, hundreds of billions, or even trillions of parameters, and their complexity is staggering.
We still don’t fully understand why and how these models work. While they have repeatedly made breakthroughs over the past seven years, some theorists insist that they cannot do anything really new-in 2021, some researchers even proposed a derogatory term for “random parrots.” Because large language models basically predict text based on the statistical laws of training data, like a parrot repeating words at random.
However, for those who continue to experience and admire these tools, it is hard to believe that they are just repeating. Especially in the past six months, this view has become even more untenable.
The original large-scale language model was more like “speaking intuitively” and lacked both the ability to “reflect” and “self-awareness”.In the words of Nobel Prize winner Daniel Kahneman, humans rely mostly on System 1 (intuitive, fast-reacting) thinking, but when we really need to think deeply, we switch to System 2 (slow, cautious and less prone to error). Most of the previous versions of ChatGPT and its competitors only had performance similar to System 1 and did not have the reasoning process of System 2.
This situation began to change in September 2024, when OpenAI released an inference model called o1,It can decompose multi-step complex logical problems and verify intermediate conclusions (and retroactively correct them if necessary), so as to better arrive at the final result.Compared with traditional large-scale language models that can only rely on memory or surface pattern matching, new reasoning models gradually have the ability to disassemble problems and deliberate.Some tests have shown that this reasoning model is comparable to, or even better than, doctoral experts in tests in specialized fields.
In just six months since the release of o1, AI has made amazing progress. The hottest topic at present is how to turn these inference models into “independent research assistants.” Their performance is amazing.
Recently, I asked a research robot to conduct an analysis for me with the theme of “Comprehensive environmental impact assessment of large events or operations such as F1 racing, Coachella Music Festival, Disneyland, Las Vegas casinos, hospitals, large zoos.” The AI spent 73 minutes reviewing 29 independent sources and providing a detailed results table and a 1,916-word textual explanation. Although there is still room for improvement in quality, which is about the level of a report that a graduate student takes a few days to write, it saves me days of time.
Just 18 months ago, my AI system could only solve some small tasks within half an hour; now it is enough for more complex and time-consuming research work.
The emergence of cognitive “production lines”
We have been witnessing a series of evolutions related to “knowledge use” and “cognitive labor.” From the initial monopoly of wisdom by temples and scholars, to the fact that printing made knowledge transferable, and then to the Internet making information itself accessible, the problem gradually turned to “how to understand information.” Now, tasks that we once thought were scarce and complex are also close at hand and inexpensive.
However, when I communicated with management of large enterprises, I found that most of them only used AI in trivial areas, such as customer service automation to save costs.The CEO of Salesforce said in December last year that 86% of their 36,000 customer support inquiries per week were answered by AI; Swedish financial technology company Klarna claimed that two-thirds of its customer service conversations were handled by AI, a measure alone that brought in $40 million in profits for the company. However, cutting costs by 10% purely through customer service is not enough for a company to achieve a qualitative leap. No great company has achieved success simply by reducing costs.
Therefore, most companies start with relatively low-end tasks and use AI to handle $50 per hour work (such as customer service chats). Although useful, it is far from transformative.But in fact, AI is also capable of tasks worth up to $5,000 per hour-such as research and development, strategic planning, or professional consulting. Why are only a few companies currently investing AI in these key aspects?
One reason is that it is difficult for people to imagine that work that “must be done by senior managers or top experts” can be done (or partially) by machines.It is precisely because of the scarcity of outstanding talents that those high-value tasks are particularly precious. Our organizational structure was designed with the recognition that “the supply of truly high-IQ talents is limited.”
Take the pharmaceutical industry as an example. A blockbuster new drug can often determine the success or failure of a company. The bottleneck is advancing drugs through the expensive and time-consuming approval process-often taking 10 to 15 years and more than $1 billion in investment, and often only one out of thousands of candidates ends up on the market. At the same time, in a large pharmaceutical company, the number of marketing personnel may be thousands of times more than the number of top R & D personnel, because truly senior research experts are extremely scarce.
At this stage, most business leaders are still in the stage of “trying to accept AI” rather than “truly believing in AI.” They are used to thinking that some problems are too difficult or too expensive and avoid them if they can.But with the advent of AI, the constraint is no longer “whether we can come up with a solution”, but more “how quickly we can verify good ideas.”
All of this will have far-reaching implications.When every company can call on digital “doctoral AI experts” at any time, the speed of innovation will naturally accelerate significantly.Just as Henry Ford’s car assembly line allows the production process to be quickly iterated and improved, AI allows ideas and solutions to be continuously polished and updated, and companies can try and make mistakes faster, learn faster, and turn quickly.
Of course, if companies do not have the ability to implement the ideas put forward by AI “think tanks”, no matter how brilliant the ideas are, they will not help. Smooth execution and integration is the key to truly opening the gap.
My daily life with AI
Over the past 18 months, I have gradually built an “AI ecosystem” to serve my work. For example, on one day in June 2024, I called these AI systems 38 times a day, and the cumulative number of interactive words reached 79,000 words for research.
By January 2025, I will no longer count the number of words exchanged. But without any objection from the other party (real person), I would bring an AI to make minutes of the meeting almost every meeting. In my daily research, I often use several different AI tools. In just one week of writing this article, I sent out at least 144 queries to various large language models-not counting transcription (26 times) and the use of code assistant tools. I now use the new generation of AI tools more times than I use Google search.
When the cost of wisdom is close to zero, the real bottleneck is no longer “how to get the brain”, but “how can we make good use of it.” Individuals and organizations that can ask good questions, objectively evaluate answers, and act decisively will be big winners.They also need to think: With too much time in their hands, what should they do with it?
author Azeem Azhar is a columnist for Exponential View[1] and a start-up investor.
reference links
[1]https://www.exponentialview.co/
