I attended a course on the "I Ching" organized by the Philosophy Department of Peking University over the weekend. Cong mentioned that studying the "I Ching" may not necessarily be applicable to our entrepreneurial endeavors, as its theoretical system differs from the modern scientific system. However, since I've already taken the class, I should review the content so as not to waste the learning opportunity.
According to Zhu Xi's research, the wisdom of the four sages who contributed to the I Ching represents a layered crystallization of insights:
Fuxi's I Ching: The earliest exploration of symbols and patterns
Fuxi observed all things in heaven and earth to create the sixty-four hexagrams, establishing the most primitive abstract symbol system. The symbol remains of the Shangshan Culture site in Zhejiang indicate that the sprout of this abstract thinking can be traced back to 9000-10000 years ago.
King Wen's Yi: Judgment of Overall Patterns
King Wen observed and explained the hexagrams comprehensively through the "Tuan Ci" (also known as卦辞). "Tuan" means "end," focusing on the grasp and judgment of overall patterns.
Duke of Zhou's Yi: Detailed Interpretation of the Line Texts
The Duke of Zhou provided specific interpretations for each hexagram through the line texts, demonstrating a deep analysis and understanding at the detail level.
Confucius' Yi: Annotations of the Ten Wings and Philosophical Sublimation
Confucius provided profound annotations and sublimation of the "I Ching" in the Ten Wings. The Ten Wings, like wings, empower the "I Ching" with the strength to soar, elevating it to a philosophical height.
"When at rest, one observes its images and contemplates its words," this is learning; "When in motion, one observes its changes and contemplates its divinations," this is divination. —— Wang Chuanshan, "Internal Transmission of the Zhouyi"
As can be seen from the above, the main uses of the "Zhouyi" can be summarized into two points:
: Study the rhetoric and principles of the *Zhouyi*. : Used for predicting the future.
" ability.
"Then, why did Confucius stop divining later?
Therefore, for ordinary people like me, when reading the *I Ching*, perhaps we don't need to use it for fortune-telling, but instead enhance our understanding and cognition of things by studying the thoughts and wisdom within it.
In the past two years, apart from starting my own business, I have almost devoted all my energy to learning AI. I tried to make an analogy between the logic of the *I Ching* and AI in order to improve my cognitive level. Although many analyses may be somewhat far-fetched, they still carry my own thoughts; unfortunately, as I know little about both the *I Ching* and AI, there should be many biased parts.
1
The Book of Changes has three meanings | Scaling Law
The Book of Changes has three meanings:
The *Yi* reveals the fundamental laws of things with its simple and clear logic, providing a convenient path of wisdom for understanding the world.
The *Yi* embodies the constant change in things, such as "the sun and moon symbolize change, representing yin and yang." Change can be both a decisive adjustment like a lizard shedding its tail and a profound transformation like the alternation of yin and yang. Through the principle of yin and yang, the *Yi* reveals the hidden patterns and harmony behind change.
While all things are changing, there also exists an unchanging essence. For example, from horse-drawn carriages to modern cars, the forms of tools have continuously evolved, but fundamental needs and ultimate questions such as clothing, food, shelter, transportation, birth, aging, sickness, and death remain constant.
The *Yi* (I Ching) finds balance between change and constancy: "There is constancy within change, and without change, constancy cannot exist." This is also reflected in the transmission of classics. When reading ancient texts, one must refer to ancient annotations, for "the classics represent eternal truths." Each era's interpretation of the classics retains the core that remains 'unchanged' amidst the 'changes,' forming the 'classic nature' of these works, thus filling them with boundless meanings and infinite depth.
) There is a deep-level fit between them.
Dario mentioned in the interview that he truly believed and confirmed the value of the scaling hypothesis between 2016 and 2017. At that time, a sentence from Ilya Sutskever gave him an epiphany: "
This statement was like a Zen koan, succinctly clarifying many phenomena Dario had observed but not yet systematized. From then on, he developed a unique mental model:
The "desire to learn" of a model aligns perfectly with the concept of "simplicity" in the *I Ching*. AI models are fundamentally latent pattern capturers; they only require appropriate optimization methods and training objectives to naturally unleash their learning potential. As emphasized in the *I Ching*, "grasping one principle governs all things," meaning that by mastering core principles, one can encompass myriad phenomena and reveal the simple essence behind complex appearances.
In practice, the scalability hypothesis embodies the characteristic of "changeability." By increasing model size, optimizing training paths, and expanding datasets, AI models can adapt to constantly changing tasks and environments. This dynamic adjustment reflects the wisdom of "change" in the *I Ching*—all things survive through change, and within change lies order and harmony.
No matter how the model is extended, its core learning logic remains stable. This commitment to "unchanging" has run through the development from GPT-1 to GPT-3, as well as the exploration of Reinforcement Learning from Human Feedback (RLHF). As emphasized in the *I Ching*, "There is constancy within change, and without change there can be no constancy." The success of the scaling hypothesis lies in focusing on core objectives amidst dynamic changes, ensuring clear directions for model optimization and the evolution of intelligence.
2
Release and Roll | The Realization of Intelligence
The *I Ching* reveals the law that all things evolve from simplicity to complexity and then return to simplicity:The Taiji gives birth to the Two Forces, the Two Forces give birth to the Four Symbols, the Four Symbols give birth to the Eight Trigrams, which expand into the Sixty-Four Hexagrams and Three Hundred Eighty-Four Lines, encompassing the mysteries of all things in the universe.
Conversely, all laws can converge from all things back to the Three Hundred Eighty-Four Lines, then return to the Sixty-Four Hexagrams, Eight Trigrams, Four Symbols, and ultimately coalesce into the Two Forces and the Taiji.
—— *The Legacy of Lu Zhai*
This process of "unfolding" and "rolling up" is akin to what Master Zhu referred to in *The Doctrine of the Mean* as "the heart method of Confucius," revealing the cyclical nature of worldly truths between expansion and contraction. Whether in terms of breadth or depth, it ultimately embodies the wisdom of the great Dao of heaven and earth.
The process of "unfolding" and "rolling up" described in *The Book of Changes* is, to some extent, analogous to how artificial intelligence achieves intelligence. "Unfolding to fill the six directions" symbolizes the expansion of knowledge and rules to cover vast domains, while "rolling up to hide in secrecy" represents the simplification and optimization of complexity into efficient and concise rules.
(such as turning, alarming, responding, etc.) between the connection methods, and this connection method determines the path to achieving intelligence:
In the 1950s, Friedberg drew on evolutionary theory to propose: placing multiple programs between input and output, allowing them to compete to retain better solutions, and then continuously optimizing through random adjustments (mutation) until an ideal result is obtained. Due to insufficient computing performance at the time, this method was not widely applied, but it may be worth re-exploring in modern high-performance computing environments.
This is somewhat similar to the initial form of "release" and "roll". By simulating natural selection, multiple possible solutions are unfolded (released), and optimal solutions are screened through competition; then adjusted continuously through mutation (rolling), ultimately deriving a more adaptive solution.
From the 1960s to the early 21st century, the mainstream approach in artificial intelligence was to encode human knowledge and rules into code, generating more rules through deduction and reasoning between input and output, enabling machines to respond appropriately to different inputs.
Just as in the Yi, there is an expansion process from Taiji to 384 hexagrams, the knowledge base system also continuously generates new rules within a logical framework, attempting to cover more complex input scenarios and output requirements.
After the 1990s, people realized that the real world is full of uncertainties, and AI shifted from pure logical reasoning to probability and statistics. After 2000, mathematical probability theory provided programming with ways to express probabilities, allowing AI to better describe and handle complex information environments.
Another aspect of probability and statistics involves reducing large amounts of uncertainty into computable patterns or predictive outcomes, which is similar to the "retreat and hide in secrecy" described in the Yi. AI compresses the probabilistic patterns extracted from data into core laws, generating optimal solutions or predictive models, which is a process of converging complexity.
After 2012, deep learning became the core of the AI field. It draws inspiration from the operation principles of the human nervous system, constructing a network consisting of a large number of nodes between input and output (for example, GPT-4 has trillions of nodes). By continuously adjusting the connection strength between nodes, deep learning can gradually optimize to achieve the target task.
Deep learning is a full embodiment of "expanding to fill the six directions and contracting to be hidden in secrecy." It establishes a vast neural network between input and output, unfolding massive information through dynamic connections among countless nodes (expansion), and by adjusting connection strengths and paths, it gradually optimizes to reach the target state (contraction). This is similar to the process in the *I Ching* where eight trigrams are derived into sixty-four hexagrams, and then through induction, converge to core patterns — mapping the real world through complexity while seeking simple solutions within that complexity.
3
The "Time" in the *Zhouyi* | AI Context
The sixty-four hexagrams and three hundred eighty-four lines of the *Zhouyi* comprehensively present "time"The philosophical connotation of this concept, which occupies a central position in Confucian thought.Here, "shi" does not refer to time as a uniformly passing entity in the modern sense, but rather represents different phases and situations in the universe and human life, possessingspecificity and wholenesscharacteristics.
" should have the same meaning.
This understanding of "moment" is closer to a dynamic situational wisdom rather than just a time marker. The core lies in "acting appropriately according to the moment," which means making suitable judgments by grasping the essence in different situations. This kind of wisdom resonates with the philosophy of "the cook slicing the ox": finding harmony in change and experiencing joy and fulfillment in the nuances of tasks.
Confucius's saying "a gentleman is timely and moderate" is precisely an interpretation of this wisdom. A gentleman can act according to the timing, adjust actions based on changing circumstances, and find appropriate responses in complex and ever-changing situations. This is not only about mastering the rules but also a kind of happiness beyond tasks and life wisdom.
In this context, Cursor believes that Context is the key to solving problems.
to filter out the most relevant 8k tokens from 500k tokens, and achieving this goal requires both efficient models and robust infrastructure support.
Like the philosophy of "timely moderation," AI also faces complex and dynamic challenges in Context processing:
to optimize information processing. Involving multi-step associative retrieval, which goes beyond the capability range of traditional single-step retrieval. function computeDiff(
firstModel: ITextModel,
secondModel: ITextModel,
): string {
//...}
Future directions for improvement include the following aspects:
Optimize embedding and re-ranking models specifically designed for code. Develop a multi-step embedding model that supports multi-step queries. Design attention mechanisms tailored to the characteristics of code repositories. Explore new methods for cross-module complex retrieval. Enable the model to internalize code repositories, giving it overall operational capabilities akin to a search engine.
Through these optimization measures, AI code assistants are expected to gain a deeper understanding of complex code repositories and provide wisdom similar to "adhering to the mean": making correct judgments in the right context, offering developers efficient and precise support.
The "timing" in the *I Ching* and AI Context reveal a deep philosophical and technical resonance. From grasping opportunities to understanding context, both ancient philosophical wisdom and modern AI development emphasize finding patterns in change and achieving balance in dynamics.
4
Virtue and Analogy | Analogical Machines
The core purpose of the eight trigrams in the *I Ching* lies in:
: To understand and penetrate the mysterious forces of the universe and their creative functions. : Categorize and reflect the diversity and regularity of all things.
as an important way of thinking. The eight trigrams symbolize eight basic types, while the sixty-four hexagrams further extend to sixty-four categories, encompassing the diversity and regularity of all things.
：The association between similar things reflects an intuitive way of understanding. Take a common example: black sesame seeds are believed to be beneficial for hair, which is based on the philosophy of "category". The connection between similar things helps people understand the patterns of the world. 
：Emphasizes grasping the true essence of things, finding their patterns and meanings through insight into the states of all natural things.
In the interview with The New Yorker, it was mentioned:
"For years, supporters of symbolic AI have believed that the essence of human beings is a 'reasoning machine.' But I think this is complete nonsense. Our true essence is an 'analogy machine,' supplemented by a bit of reasoning ability to identify and correct where analogies go wrong."
The "distributed cognition" theory proposed by Hinton and his collaborators James L. McClelland and David Rumelhart has a profound resonance with the "Tongde Leiqing" concept in Chinese classical philosophy.
That is, to establish connections between related concepts through a shared set of "building blocks."
: For example, if it is known that "chimpanzees like onions," one might speculate that "gorillas also like onions." This associative ability reflects the distribution and sharing of knowledge across multiple similar concepts. : The formation of knowledge depends on the "feature activation" of a group of neurons, such as furry characteristics, quadrupeds, primates, animality, intelligence, wildness, etc. These features can be slightly adjusted to change from "chimpanzee" to "gorilla."
This process aligns with the concept of "categorizing the emotions of all things" in the I Ching: through analogy and classification, similarities and regularities between things are identified, thereby revealing the diversity and unity of all things.
Hinton further pointed out that the complexity of distributed cognitive patterns may also lead to errors: if features are incorrectly combined, it might produce an imaginary creature that is neither a chimpanzee nor a gorilla. However, through learning algorithms, the brain can adjust the weights between neurons, tending to generate more reasonable conceptual combinations rather than chaotic fantasies. The I Ching system presents the laws of all things through classification, while continuously correcting in changes to ensure the overall stability of the system.
5
Meaning - Image - Language | LCM
From Fu Xi's sixty-four symbols to the formation of the entire classic, the I Ching uses "meaning - image - language" as its logical mainline. Through the medium of images, the sages attempted to convey ultimate truths beyond language, while using language to complete the inheritance of civilization.
：The Origin and Core of Thought
：Intention is the essence of thought, which is difficult to convey directly and needs to be concretized through imagery.
：The Concrete Carrier of Thought
("Words cannot fully express thoughts, and writings cannot fully capture words"): Language has a ceiling when it comes to describing profound ideas and complex situations, which can be transcended by imagery. : Imagery transcends textual logic, conveying unspeakable deeper meanings through images and metaphors. ：The image is both a carrier of imaginative thinking and a tool for revealing the intrinsic laws and mathematical structures of things.
：Concretization and the inheritance of civilization
：The image further transforms into language, forming the mode of knowledge transmission. : Although the means of language are limited, it is an indispensable tool for passing on civilization and expressing experience. : The trigram and hexagram statements in the *I Ching* are not ordinary language but encoded expressions of objects, requiring decoding to understand their deeper meanings.
In recent years, LLMs (Large Language Models) have completely transformed the field of artificial intelligence, becoming a standard tool for various tasks. They operate on tokens (words or characters) as their basic unit, modeling language through processing inputs and generating outputs. However, the limitations of this technology are also evident: they can only function at the level of linguistic symbols, struggling to capture high-level semantics or multimodal information. Human thought, however, transcends single-word operations, enabling the analysis of information across multiple levels of abstraction and the generation of creative content.
This high-level representation means that concepts are not limited to language or modality, but express a more abstract idea or action. In the design, researchers assume that each concept can correspond to a sentence, and use the SONAR embedding space to provide an abstract representation of sentences. The SONAR space can support 200 languages and cover both text and speech modalities, thus enabling cross-lingual and cross-modal information processing.
The advantages of LCMs not only lie in their semantic abstraction capabilities, but also in their support for multi-modality and multilingualism. Through conceptual processing, they elevate tasks to a higher level of semantic operation, making up for the shortcomings of LLMs which remain at the token granularity. This multimodal flexibility allows LCMs to integrate information in various forms such as text and voice, and convey meaning in a more comprehensive way.
Limitations of LLM: Stuck at the level of "words"
During the generation and comprehension processes, only inputs and outputs at the language level can be handled. Similar to the trigram texts in the I Ching, which are not ordinary language but encoded expressions of objects that require decoding to understand their deeper meanings, LLMs also use Tokens to enable AI to understand and output expressions. This method is similar to what the "I Ching" criticizes as "words failing to fully convey meaning," unable to comprehensively present deep semantics or multimodal associations.
The uniqueness of LCM: transcending the "image" of language
LCM operates at the conceptual level, possessing language and modality independence, capable of handling abstract semantics at the sentence level. 
: The representation of meaning is not only expressed through language but can also transcend text, speech, and images, achieving more effective communication of meaning.
Commonalities between Representation and Concepts
: Representation reaches the core in the most concise way, similar to the semantic abstraction mechanism of LCM. : LCM performs prediction at a higher semantic level, which can better reveal the deep logic and relationships between pieces of information.
The symbol possesses dual functions of compression and expression. After LCM completes abstraction at the conceptual level, it realizes efficient semantic transmission by mapping to language or other modalities, just as the inscriptions of hexagrams encode meaning through symbols. The modality-independence of symbols suggests that LCM could become a key to AI multi-modal integration: different inputs such as text, speech, images, and actions can find common abstract associations at the conceptual level.
Nevertheless, LLMs and LCMs are not entirely opposed but can form complementary relationships, similar to the joint function of symbols and words. The strong generative capabilities of LLMs in specific tasks can be combined with the semantic abstraction advantages of LCMs, forming a more comprehensive artificial intelligence framework. This combination may bring us closer to human cognition, from language to semantics, and from symbols to concepts.
6
Fuxi's Independence | AGI and Superintelligence
The pioneering act of Fuxi drawing the trigrams signifies an original breakthrough in human wisdom.This kind of "unreliant" creationprocess laid the philosophical and symbolic foundation for civilization, while also embodying humility and a spirit of exploration in the relationship between humans and heaven-earth.
Spirit of the primeval era
Taking heaven and earth as teachers, they established a symbolic system through observation and exploration, initiating early civilization. This spirit reminds us that when facing the unknown, we should adopt an open attitude and fearless courage to reconstruct our cognitive framework and the foundation of civilization.
The boundaries of technology and the courage to create
In the face of unknown fields, the combination of science and philosophy is particularly important. We must not only deeply understand the core of technology but also expand broader meanings and directions through philosophical thinking. Learning Fuxi's "independence" is not only for breaking through technical boundaries, but also for injecting heart and spirit into the creation of civilization.
The philosophy of the *I Ching* emphasizes:
, but rather 
By observing natural laws, humans participate in the process of the universe's operation and engage in meaningful creation, rather than attempting to dominate or change everything.
the principles of nature, gaining deep insights into its various levels and dimensions.
Sam Altman wrote a blog post about AGI today. Here's an excerpt:
We founded OpenAI about nine years ago because we believed that AGI was possible, and it could be the most impactful technology in human history. We wanted to figure out how to build it and make it broadly beneficial; we were excited to try and leave our mark on history. Our ambition was sky-high, and we believed this work could benefit society in equally extraordinary ways."
Nine years ago, we really didn't know what we would end up becoming; even now, we only know a little bit. The development of artificial intelligence has gone through many twists and turns, and we expect there will be more in the future.
We are now confident that we know how to build AGI according to the traditional understanding. We believe that by 2025, we may see the first AI agents "joining the workforce" and substantially changing the output of companies. We still believe that putting great tools repeatedly into people's hands will bring great, widely distributed outcomes.
We are beginning to shift our focus towards superintelligence in the truest sense. We like our current products, but we're here for a glorious future. With superintelligence, we can do anything. Superintelligent tools can greatly accelerate scientific discoveries and innovations far beyond what we ourselves can achieve, thereby significantly increasing abundance and prosperity.
It sounds like science fiction, and it's a bit crazy to even talk about it. But that's fine - we've been through it before, and we're willing to go through it again.
Fuxi observed heaven and earth without any reliance on existing frameworks, establishing the symbolic system for humanity. The development of AGI (Artificial General Intelligence) also requires this kind of innovative spirit that is not bound by current frameworks, creating new ways of cognition and technical boundaries. Fuxi's wisdom was not only the beginning of symbolic technology but also a reflection of a worldview and philosophy. Similarly, the development of AGI needs to consider its social significance and philosophical value, exploring how to achieve harmony between humans and nature in the process of technological progress.
The development of artificial intelligence will not stop at the level equivalent to human intelligence. Once AGI (Artificial General Intelligence) is achieved, the advancement of intelligence may quickly enter an exponential growth trajectory, eventually surpassing human understanding and control, leading to superintelligence. This transformative change is referred to as the "intelligence explosion," which will profoundly reshape our society, economy, and security systems. I.J. Good's theory of "intelligence explosion" proposed in 1965 clearly described this recursive improvement dynamic. He believed that once a machine can design another machine smarter than itself, the growth of intelligence would enter an uncontrollable chain reaction phase. "The first ultraintelligent machine will be the last invention that mankind will ever need to make."
Fuxi established the sixty-four hexagrams by observing heaven and earth, without any reference, laying the foundation for Sinology and Confucian civilization. This kind of primordial innovation shares a similar spiritual core with the development of AGI, especially when moving towards the superintelligence phase, humanity is attempting to pioneer an entirely new form of intelligence, just as Fuxi explored the creation of a symbolic system in the midst of chaos. However, we cannot...
Perhaps, in the process of the birth of AGI and superintelligence, humanity is merely an assistant, not the true master.
Weakness and ignorance are not obstacles to survival, arrogance is.
-- Liu Cixin, "The Three-Body Problem"