Chapter 34 - 15: Profound Tofu-Filled Dumplings

Chapter 34: Chapter 15: Profound Tofu-Filled Dumplings

Some money spent makes one’s heart bleed. Five thousand one hundred yuan. To Yu Xian, this was not a small sum.

But after hanging up the phone, the phrases the other party had used—"dimensional strike," "dull and tasteless," "endless aftertaste"—continued to surge into his mind like a rising tide. He wondered why the more he thought about it, the more it sounded like a pyramid scheme sales pitch. But this was just hearsay and fragmented information. As for the actual situation, he could only wait until he got the invitation code and experienced it for himself.

He added the seller’s contact information, tuned into the "Subtraction" radio broadcast, and finally managed to doze off for a few hours to the sound of the rain outside the window.

Saturday morning.

The sky remained overcast, and the rain pattered against the windowsill. Fortunately, starting today, November 10th, his cousin’s residential complex had turned on the central heating. His task for the day was to study his parents’ research paper.

He washed his face, the cold water clearing his muddled brain significantly. He walked into the kitchen and rummaged through the refrigerator, pulling out the frozen dumplings his cousin had bought. He checked the packaging—tofu-filled. He boiled some water and tossed them into the pot. The plump, white dumplings tumbled in the boiling water as steam billowed upward, fogging up the glass window.

Taking advantage of the time it took to boil the dumplings, he cleared off the coffee table in the living room. He removed the black binder clip from the research paper and spread the thick stack of A4 paper flat across the table. After setting out scratch paper, red and black pens, and his laptop, he sat down on the sofa. This posture reminded him of how his cousin had looked while poring over case files the night before last.

The water in the pot boiled a few times, and the dumplings floated to the surface. He scooped out a bowlful, placed it on the corner of the coffee table, and poured a small dish of vinegar. He ate as he read.

The difficulty of reading this paper far exceeded his expectations. Although he was a science student with decent English, the document was filled with obscure, interdisciplinary academic jargon. Information theory, topology, neuroscience, computer science... The dense thicket of terminology was like the academic world’s Tower of Babel.

He spent almost the entire morning looking up words in the dictionary and searching for related papers. By the time his bowl of dumplings had gone completely cold, he had only barely managed to digest the core concepts in the title.

"Research on Vectorization Mapping and Storage Mechanisms of Discrete Personality Based on High-Dimensional Topological Manifolds."

He had already underlined the title with a red pen, breaking it down into a few keywords: "Discrete Personality," "Vectorization Mapping," and "High-Dimensional Topological Manifolds." He thought to himself that this was going to make his brain grow.

There were a few foundational concepts at play here. The first was "Discrete Personality." In psychology, personality was usually viewed as a continuous trait.

What did this mean? For instance, extroversion and introversion weren’t black and white. Most people fell somewhere in the middle ground between the two, existing on a continuous spectrum. It was just like adjusting the volume on a phone; it could be adjusted seamlessly from zero percent mute up to one hundred percent maximum volume, stopping at any point in between.

But the hypothesis proposed in this paper was different. His father believed that if the dimension of observation were raised, human personality was actually composed of a finite number of "discrete states." It was similar to how the MBTI categorized people into sixteen types—like "furrowed brows," "happy puppies," or "little butterflies"—or the Big Five personality traits. Of course, the categorization in this paper was much more complex. It might be divided into thousands or even tens of thousands of types. But no matter how finely divided, it remained "discrete."

Yu Xian picked up a stone-cold dumpling and took a bite. The texture of the tofu filling was somewhat novel. "Discrete" was just like this bowl of dumplings. This one was tofu-filled, that one was three-delicacy-filled, and another was pork-filled. They were distinct, well-defined categories. You couldn’t eat a mixed-state dumpling that was thirty percent pork and seventy percent tofu. In the definition of discrete, even if it were a mixed filling, it would be defined as a new, independent category called "pork and tofu filling."

What did this mean? Yu Xian hadn’t quite figured it out yet, but his parents treated human "personality traits" as something akin to building blocks—finite and exhaustible.

He continued reading the paper. The second concept was "Vectorization Mapping."

This paper was from 2016, which was also the year his parents had their accident. At that time, a Go AI named AlphaGo had just defeated the former Go world champion Lee Sedol, and the concepts of AI and "deep learning" were just beginning to enter the public eye. Now, in the present year of 2025, "large language models" had taken over the entire world. Doubao, ChatGPT, Grok, Gemini, Qianwen, Yuanbao... they were all representatives of large language models.

Although Yu Xian didn’t research AI, he knew that the cornerstone of large language models was converting text into "vectors." This was the "vectorization mapping" mentioned in the title.

He was shocked to realize that his parents had already been researching "vectorization" a decade ago. After a morning of study, Yu Xian could still only grasp this concept on a superficial level.

Simply put, in the eyes of a computer, all text existed in the form of "coordinates." Take a very famous formula in this field as an example:

King - Man + Woman = Queen.

The king vector, minus the man trait, plus the woman trait, equals the queen vector. Every word in this formula could be converted into coordinates. For example, king is [0.8, 0.6, -0.1, ...], and queen is [0.8, 0.9, -0.1, ...]. Through this method, these words could be computed. ℯℯ𝘭.𝘤ℴ𝘮

Yu Xian looked at the dumplings in his bowl, and a metaphor popped into his mind to help him understand.

If he treated a "pork dumpling" as a vector, treated "pork" as one feature vector, and treated "chives" as another...