A collection of notes, ideas, and things I figured out after plenty of tinkering.
Array-based red-black trees, gaokao finale problems, and PhDs who cannot find jobs are, at bottom, the same thing: closed systems can select for local optima, but the definition of this 'optimum' is itself distorted by the closed conditions. The information environment determines the output efficiency of intelligence, and you cannot price what lies outside your cognition.
The rise in humanity's average capability comes less from individuals studying longer than from the continuous upgrading of how knowledge is diffused, compressed, and invoked.
To call exam-oriented education fair is a fundamental misrecognition: it merely renames the absence of choice as the superiority of the only available choice.
The fundamental fracture between premodernity and modernity lies in the optimization target: no longer whether I am omnipotent and miss nothing, but what specific demand I can satisfy within the supply chain.
From std::fma to allocator_traits, SFINAE, PMR, and Concepts: the sharpness of C++ lies in how it forces hardware facts and the type system into the same room.
Most schools, companies, and individuals have not really understood AI; what they understand is the anxiety, slogans, and imitative posture around it.
The deepest side effect of exam training is not that it asks people to solve problems, but that it trains the world into a set of judgeable standard answers: rewarding the correct, punishing the wrong, and gradually making people fear error, deviation, and risk itself.
Choices are hard to understand across different maps of the world; what moves us is not the amount of suffering endured, but the capacity to keep choosing love, goodness, and justice in the face of it.
When an operand does not originate from IO, runtime is often an engineering trade-off rather than an ontological fate.
Many quantitative teams really do scrape Reddit data and turn it into sentiment indexes; in that sense, memes, despair, and mania can all become observable market variables.
From OpenMP and pthreads to CUDA, parallelization is not merely an expansion of toolchains, but a renewed interpretation of computational power, hardware limits, and formal models.
Social life broadly has two forms: intimacy and exchange. Intimacy depends on contingency; exchange is the ordinary condition. The real question is not calculation, but how value becomes visible, felt, and reciprocal.
This points to a transformation in the research paradigm. With PUCT (Predictor + UCB applied to Trees), the center of gravity shifts from solving problems to finding them, and toward designing evaluative measures that precisely reflect scientific goals.
The history of compiler optimization resembles a discipline of expanding what is available while reducing what is wasted: on one side, optimization algorithms evolve; on the other, we feed the compiler more information and thereby move the boundary of semantic constraint. From a modern C++ perspective, template metaprogramming should be returned to its proper role as an abstraction over families of processes and a tool for policy-driven design. Computing Fibonacci numbers with templates was, above all, a pre-C++11 detour taken under constraint.
Just as academic taste requires long cultivation, the discernment of demand also contains a great deal of tacit knowledge, and it too must be trained. Yet this is precisely what the narrative sustaining the university refuses to provide: it teaches you how to 'solve problems,' but not how to 'find problems.' And the ability to 'find problems'—to discover and define valuable questions—will be far more precious in the future than the ability to 'solve them.'
The current rush into deep learning resembles the posture of an investor, still far from the self-consciousness of an AI-native believer.
A low-cost starter protocol for young CS people, myself included.
I admit that I have not used Bilibili frequently enough to train my account profile properly; nevertheless, its personalization algorithm still feels remarkably crude. It may capture my general persona and field of interest, but it fails to match the corresponding depth and information density, remaining instead at the level of tag similarity. The signal-to-noise ratio is too low, and the experience is predictably poor.
Pure formal models are often a projection of reality, a selective reflection. Algorithms, as logical models, also face scrutiny regarding their scope of application in the real world.
There is no absolute perfection, only contextual suitability; to move along the boundary between real-world constraints and technical possibility is precisely the charm of computer science.
AI has already become a technology with substantial barriers, far removed from a toy-like or student-level practice.
A concept is not merely a more respectable syntax for templates; it is a way of handing algebraic-structure information to the compiler.
GPT-5 shows a genuine command of interfaces, type constraints, and allocator semantics.
While C is often treated as a subset of C++, several subtle differences between the two languages remain technically consequential.
The key to entering CS is not some mystical talent, but the formation of passion and taste through methodology, metacognition, and institutional conditions.
Between abstraction, derivation, and the cold bench, a return to the halfway point of computer science.
Would it not also be good to become a small bird flying freely through computer science?