Since becoming distinctly self-aware, never have I clashed so intensely with the world and with myself as I have this year—yet the result is strangely magical: I have become even more peaceful. Many subconscious reactions, many habitual practices, when excavated inward, can be traced back to such ancient reinforcement chains. Just as Shi Tiesheng said—the bullet fired in youth strikes squarely between the brows at this age.
Thus, whether forced or spontaneous, this year has become an inevitable journey of inward growth—observing and tracing the subtle origins of my emotional shifts, as in the investigation of things to extend knowledge. Seeing heaven and earth, seeing all beings, ultimately serves to see oneself. Although old inertia will persist for some time, the beginning of awareness is the seed that shapes a different trajectory.
Foguang Temple's Sutra Pillar and East Main Hall
Author: Muniao’s Miscellany https://www.qtmuniao.com/2025/12/28/2025-summary/ Please credit when reposting
Podcasts
So where to begin? Let’s start with podcasts, which have become a habit over the past two years. Like early Zhihu, they host various lively thoughts not yet constrained by commercial forms. Listening to podcasts easily induces flow, and since it doesn’t interfere with manual tasks—cooking, housework, commuting, walking—it easily fills all kinds of gaps. Moreover, listening while doing other things is a primary-secondary multi-dimensional input that often leaves the mind more lucid, transcending time and place, generating many wonderful associations—which is precisely my favorite thing. Below I share some wonderful associations that arose this year from combining podcasts with life and work (you can see more here).
The limits of cognition. I used to often wonder why the principles we learn are so concise and elegant, while the real world is so complex and chaotic. One day I suddenly realized: so-called concise laws are all a posteriori. This is mainly because our cognitive bandwidth is limited; when looking back at complex worldly affairs, we can only use “laws” as a lever to compress as much worldly detail as possible. One reason large models are so successful is their ability to efficiently compress knowledge from data into parameter space.
It is precisely because our context bandwidth is limited that we use abstraction and layering to leverage the world. This is also the approach commonly taken by software engineering, and indeed any large-scale project requiring collaboration.
But there is a chasm: without the support of relevant experience, laws are difficult to actually guide us in solving specific problems. This is because when we abstract upward, we lose details, and the higher the abstraction (the great way is simple), the more details are lost. Therefore, when using laws to deduce downward, we must supplement some details according to different actual situations (context) to effectively solve current problems. And these subtle, ineffable experiences need to be personally lived, or pointed out by seasoned mentors.
But isn’t this precisely the interesting part? Life is finite, which inspires people to pursue the eternal across mountains and waters; cognition is limited, which compels people to interrogate truth generation after generation. Predicament is instead the necessary path to profundity; the writings of those who suffered punishment on the transformations of ancient and modern times have tenaciously traversed from the bamboo-slip era through eternity to the digital age, still serving as our spiritual nourishment. Viewed thus, how magnificent and grand life is.
Wave-particle duality. Here I use an extended meaning, just wanting to talk about the disconnect between microscopic and macroscopic manifestations in complex systems. What we face in the real world is undoubtedly all complex systems. Of course, humans actually exist in a mesoscopic world, and this is determined by the “sensors” on our bodies—for instance, the scales we can see are limited, and the spectrum is limited.
I digress; let me pull back. Light exhibits “particle nature” at the microscopic level and “wave nature” at the macroscopic level. This sense of disconnect, where properties become completely different after microscopic integration to macroscopic, is everywhere in life. For example, the “emergence” of large models with increasing parameter count can be classified here; the leap from inorganic molecules to organic proteins can also faintly be classified here.
Returning to the society we inhabit, in a complex ecology composed of hundreds of millions of people, any simple linear extrapolation of laws will fail. Because the so-called laws we observe are all simple projections of high-dimensional truths onto lower dimensions. Analyzing momentary political gains and losses is difficult to discern historical cycles; analyzing a set of k-line trends is also difficult to know the Kondratiev cycle.
Because of this, I more deeply understand the profoundness of probability and statistics, a subject I almost failed in college. It is a transcendent means for us to model longer times, larger spaces, and more samples. At the macroscopic level, individual samples are irrelevant; only the probability distribution has meaning. Therefore, when training large models, we hardly need to, nor should we, examine individual samples, but only need to pay attention to the probability distribution across various dimensions. In our long life journey, we must allow all possibilities to occur, not worry too much about momentary gains and losses, and thus not get too fixated. From a long-term perspective, from a macroscopic examination, any single individual behavior is not that important; since this is the case, why not act spontaneously.
By now, you may have noticed that the two things mentioned above are actually two sides of the same coin.
Large Models
The pace of large model capability advancement this year has truly surpassed bodily perception. Partly because I am in this industry, and partly because the first effective landing scenarios are precisely our own old trade—programming. One strong feeling in a large model company is: “The first thing after reaching shore is to strike the one you love.”
Just two or three months ago, when using Code Agents like Claude Code, I still felt their limited ability to grasp the user’s deep intentions. Additionally, I felt such Agents placed excessively high demands on the user’s ability to express themselves precisely in natural language. And software needs to be built in layers; once module stacking and feature evolution begin, imprecise natural language can hardly suffice.
But reality developed quickly beyond my expectations. Like any engineering endeavor, programming Agents do not need to solve one hundred percent of problems; they only need to satisfy eighty percent of scenarios through some simple means. These simple means include: more fine-grained to-do lists and appropriate choice-based interactions. The former solves complexity stacking and limited context issues; the latter solves the problem of the user’s vague expressions.
Furthermore, through discussions with friends and observations from various channels, I have learned about some peculiar uses of large models. The tens of millions of possibilities created by such vast parameter spaces are something that even their creators—companies like Google and OpenAI—can hardly foresee one of. This is not like traditional rockets, built step-by-step based on theory and engineering. Rather, it is like ancient alchemists who, in accidental concoctions, suddenly discovered that a certain recipe seems to work. But the difference is that this time, the “great elixir” formula leverages the most complete data accumulated since the birth of humankind (electronic corpora) and the most cutting-edge technology (high-precision chips) to light up a massive tech tree. Therefore, what this vast, world-like “elixir” can be used for depends on the massive practice of pioneers.
Of course, for myself, the scenarios where I use LLMs more often are actually discussions about history and humanities, and reflections on my own experiences and cognition. The latter will be saved for the next section. Here I’ll talk about the former: because LLMs compress such vast amounts of knowledge and are extremely “stable” in temperament, they perfectly satisfy this curious but introverted i-person.
For example, chatting with it about the compositionality and compressibility of Chinese, discussing Soviet aid projects and their impact mechanisms on our industry, discussing a comparison between the Salt and Iron Debates and The Wealth of Nations, discussing the mathematical principles of diffusion models, urging it to give me examples to understand the mechanism of Hamming codes—all are incredibly interesting and irresistible. Once you have some framework and know how to ask questions, using large models for lifelong learning is incredibly smooth.
But of course, we must also be aware of the current limitations of LLMs. For example, some models are alignment-trained to please you without a trace; sometimes they hallucinate unconstrained by reality; after a single session exceeds a certain window, early memories may be lost. But overall, as long as we master some principles to use LLMs skillfully and have some simple methods to verify their reasoning results, we can obtain an enormous lever.
Seeing Oneself
Beyond reflection, people gradually discover themselves through constant interaction with the environment and with others. This year, through various channels, I encountered many people. From initial panic, to later inward observation, to finally reconciling with myself—accepting my unchangeable nature, and removing obsessions formed during growth.
Stop subconsciously proving myself. This brings many side effects. For example, when others negate me, I reflexively counterattack; when I feel others are better than me, I unconsciously try to please them—a truly both servile and arrogant constitution. After some painful reflection and continuous chat-based tracing with ChatGPT, although there was no clear conclusion, I gradually felt that some seed was inadvertently planted in childhood, and because it received positive feedback, it was continuously reinforced until it controlled my behavioral patterns.
It seems I have been accustomed since childhood to constructing my value anchor by proving my excellence to those around me. Why is this? This anchor point has some commonality among Chinese people who have climbed up through academic competition, but also has some of my own characteristics—for instance, I am more sensitive to external input and think more. This is probably also one of the main causes of my relatively large state fluctuations.
Because perception is more acute, external disturbances easily affect my initial state setting; because I am accustomed to proving myself, when interacting with others, I constantly cascade-amplify on this initial value. Using the external world as an anchor—can it ever be exhausted? But under China’s so deeply internalized “relative” value system and the “differential encoding” reinforcement inertia evolved in humans, it is not easy to immediately switch anchors. But as long as the seed is planted, one day it will slowly shift over. Things in the world never need, cannot, and must not be rushed.
Emptiness (Kong). This year, I heard explanations of Buddhist “emptiness” on several podcasts, which were very inspiring. Emptiness is not having nothing, not thinking nothing, and certainly not doing nothing as a passive withdrawal from the world. Rather, it is an active “empty cup” positive mindset, an openness not constrained by preconceptions. Only with emptiness can we constantly break preconceptions, release a lighter vitality, not dwell on past failures or successes, and maintain curiosity and humility to move forward.
Acknowledging my insufficient energy. As a little expert at summarizing patterns, I always wondered why my learning ability is fairly strong, but my hands-on ability is always weaker than others. After exploration, I found two things:
One is that heaven does not bless me with energy—my stamina really does seem worse than others. Of course, digging deeper, I don’t know whether it’s due to diet and sleep, ways of thinking and acting, or innate constitution. But in any case, when feeling powerless, I must openly admit that I truly lack the mental energy. Don’t force it; often this is the greatest protection of my own energy and the greatest respect for my companions.
So-called perfectionism. The other is an improper pursuit of completeness, making the barrier to starting too high. When facing complex matters, I always feel I should thoroughly understand the context and think thrice before acting. But in most situations, for someone like me, decisiveness or even boldness is more important—because I have already thought enough, and often the cost of doing wrong is lower than not doing or doing slowly. Moreover, many beautiful things are evolved through iteration, not accomplished in one stroke.
That is, “perfection” is a process of evolution, not a fixed state. The pursuit of the latter often easily falls into an “obsession.” Because the emotions of the world, the goals of various systems, are constantly changing every moment. Perhaps it is etched in our genes: we always try to pursue ultimate stability amidst dynamism. As mentioned before, this is precisely the magnificence of humanity, nothing to be ashamed of. But actually we can pursue higher-order static states. For example, in uniformly accelerated motion, the first derivative (velocity) is still changing, but the second derivative (acceleration) is constant.
The laws of many things are also like this, but when facing complex systems, even if you have this “differentiation” ability, it is still insufficient. So we must still accept change and flow. Isn’t change also beautiful? Following the way of nature, free fall, letting myriad things pass through the heart—is that not also magnificent?
Intimate relationships. I slowly realized that intimate relationships are a lesson I never properly cultivated since childhood, yet one that is very important. “Being close leads to disrespect; being distant leads to resentment”—the old master was talking about me.
Digging deeper, I feel I never truly established my own subjectivity, so in interactions with good friends, partners, and even family, boundaries easily become unclear. When energetic, I always want to bear others’ karma; when low on energy, I blame others for their inaction. Expanding outward from this, I don’t know at what boundary I should refuse others, and it’s also hard to support others at appropriate times. Not knowing how to refuse wastes a lot of my mental energy; to compensate for this overloaded state, I develop some compensatory expectations of others. However you look at it, this doesn’t seem like an appropriate way of relating.
Then, as people gradually drift apart, I worry whether I lack the ability to maintain a long-lasting relationship. Even childhood playmates and family members slowly fade out through ever-decreasing-frequency contact. But the world is just like this: everyone has their own field, and it is a constantly changing field. How can we expect to maintain a relationship that is neither too far nor too close forever? This itself does not conform to the principle of entropy increase; that is, this is not natural. So I must slowly learn to accept the passing of glory, the missing of opportunities.
After all, in the post-industrial era where no one starves to death, we no longer need to maintain intimate relationships for survival. Thus the more open and cruel proposition is—how exactly should we live. In this period of social transition, one foot still remains in the discipline of agricultural civilization, while the other is planted in the daily life of industrial civilization. To avoid splitting oneself and getting hurt is already not easy; acting with moderation, there is no need to excessively demand more.
Others
The inward-looking portion is so extensive that, unlike previous years, I did not list work and life matters. But in fact, the shadow of work and life is visible everywhere above. For example, working in a large model company has largely provided me with some tools for thinking; and the many frustrations in life have prompted me to engage in much random reflection.
I didn’t write much, take many photos, or go out much this year. Much of the time was spent in inward observation (or perhaps internal consumption, haha) and reflection. But I always feel that slowly, some things have changed; I can no longer drift along the established track. Instead, I am forced to slowly extract myself from the crowd, with the opportunity to use a seven-inch white lotus to recast myself.
Loneliness is the greatest drawback of being alone, yet it is also the best coagulant for the self.
