December 2025
After two arduous months of pushing through the paper, the second project has finally come to a temporary close. Next, I need to keep pressing on with the third project while polishing the second. It was only during the rush of paper writing that I came to truly appreciate the importance of focus, and to realize that each day really can hold a great deal of work, if only you do not waste it.
The single most important thing is what you do first when you arrive at your desk in the morning. It directly determines the productivity of the entire morning, even of the entire day. You must first feel interested in something, and only then will you have enough drive to do it well. Interest is also one of the elements that directly distinguishes "excellence" from "greatness" (alongside choices and luck). Diligence can separate the ordinary from the excellent, but the gap between excellence and greatness cannot be bridged by diligence alone. If you aspire to greatness, you still need to "love your life".
20251222-20251228 Week 146
- Continued literature survey
- Began learning Linux
20251215-20251221 Week 145
- Started surveying for the next project, only to discover that the topic I wanted to work on had already been done...
- Started using Obsidian. Found it works quite well in combination with Dropbox
- The hard drive really did fail. Hurried to buy a new one before it broke down completely
20251208-20251214 Week 144
- Pushing the paper
- After finishing the paper, I rarely got the chance to do a city walk in Shanghai
20251201-20251207 Week 143
- Pushing the paper
November 2025
A reminder: your energy and attention are exceedingly precious. Do not squander them on trivial, unimportant matters.
Are the things I am doing right now connected to my plans for the next three, five, or even ten years? What is my eventual positioning? In order to qualify for the work I want to do later, do my current efforts carry meaning and value? If not, then what should I be doing? In short, it is a question of focus. One must notice the connections between things. I cannot keep wasting time on things that have no value.
20251124-20251130 Week 142
- Coding, pushing the paper
20251117-20251123 Week 141
- Coding, pushing the paper
20251110-20251116 Week 140
- Coding, pushing the paper
20251103-20251109 Week 139
- Coding, pushing the paper
- Proctoring the C programming quiz
October 2025
In the latter half of the month, Hangzhou plunged from thirty degrees to seventeen in two days, racing into winter. Caught off guard, I was drenched by a fine drizzle while cycling, and my tonsils flared up. It was, however, somewhat foreseeable, since I was still dressed in summer's short sleeves and shorts. The progress on my research is no longer fast enough to catch the mid-November SP, so I can only look ahead toward USENIX. If I miss even USENIX, I will truly have to apologize with my life.
The plan of the day lies in the morning. When I get up early, my attention is genuinely concentrated, my mood is steady, and there is no brain fog. I must take care not to be ensnared by the various messaging apps, and to keep a firm hand on the steering wheel of my thoughts -- the key is to constantly remind myself: what am I doing right now? Is it important? Is it worthwhile? Because sometimes a person is unaware of what they are doing; the body simply reacts on instinct. At such moments, the brain must actively step in and take over.
Eat well. Not only at the level of food that fills the belly, but even more so at the level of the information you ingest. The internet is awash with vulgar and ineffective junk information, and even when you recognize it as junk, the brain still allocates a portion of its energy to processing it, and even reprocesses it repeatedly in the background. You must recognize this as an erosion of your attention. Suppose the attention you wake up with each day is a fixed quantity: the more is allocated (often not even by your own conscious choice) to that junk information, the less remains for what truly matters. Scrolling on the phone also calls upon the prefrontal cortex; it is no real form of relaxation, and instead deepens the burden on the brain.
At the end of the month, the lab replaced nine large light panels, and all at once the room blazed bright as day. One could study and work without any sense of fatigue, oblivious to the passage of time. That very night I worked straight through to eleven-forty without feeling sleepy. I thought of the shadowless lamps in surgical rooms, and the warm lamps in chicken farms. Beneath such bright lights it seems easier to slip into a state of flow, so long as the mind stays focused on the work at hand, comprehending and reflecting on the meaning behind the words, you simply enter the flow. Anxiety as an emotion brings nothing positive in itself; no matter how urgent or busy, one must still settle the mind. This requires choosing your study targets in advance and reminding yourself of progress in good time. Do not linger too long in any one place. Take a look at the job market, too, and learn what abilities are in demand.
20251027-20251102 Week 138
- Continued refining the paper and the experimental code design
20251020-20251026 Week 137
- Attended the Doctoral Innovation Forum, fulfilling one of the graduation requirements
- Built a website for "98 Information Processing"
- Wrote on the paper
20251013-20251019 Week 136
- Did sundry tasks: organizing GPU quotes, helping assemble a machine, proctoring, compiling exam analyses, and the like
20251006-20251012 Week 135
- Bought new equipment, and synchronously fast-tracked the paper
- Discovered that the boot.dev website is rather well made
September 2025
The research code went through several rounds of iteration. From GenAI it was upgraded to Agentic, from commit analysis to knowledge graph construction, plus the painfully slow analysis of the dataset reports -- on the whole, progress can only be described as glacial. Around the middle of September I finally caught a slightly clearer line of thought, and only after sobering up did I realize that everything from the dataset to the experimental design to the comparison with related work must be done with genuine fine-grained care. The once-a-week meetings really do interrupt continuous trains of thought; sometimes research becomes "research for the sake of the report", which leaves a lot of things only superficially done.
I have also come to feel the toll a PhD takes on the spirit. Though there is no explicit setback, something invisible erodes one's vitality. My speech grows ever more mean, my thinking ever more numb. Perhaps it is because I have not published in a long time, and the experiments keep spinning in place without progress. I grow more and more lax toward other matters, and the swings in my mood grow ever smaller.
Ancient philosophy was right: "Daily I examine myself thrice." If I were to give a talk, could I deliver one with structure? Could I answer the audience's questions? Could it withstand scientific scrutiny? Out of an entire day, how much time was spent on improving my own abilities, how much on ineffective labor, and how much on slacking off?
20250929-20251005 Week 134
- Continued implementing the v5.2 experimental code
20250922-20250928 Week 133
- Wrote a proposal
20250915-20250921 Week 132
- Migrated to Linux to keep running experiments; Windows not allowing "aux" as a file name was rather absurd
- Worked on the slides for the C programming course
20250908-20250914 Week 131
- Continued working on the xnn project
- Subscribed to Medium! Many of the articles there are quite good
20250901-20250907 Week 130
- Continued working on the Streamlit admin dashboard for the lab website
- Held a meeting in 201 to discuss the direction of system development
- Wrote the xnn proposal
August 2025
Late in the month I suddenly realized that the truly crucial point in reading a paper is not to memorize what the authors built, but to attend to how the authors distilled their method, to seek out the design philosophy within, and to inductively extract a methodology. If you are not very familiar with the field, you can also pay close attention to what kinds of problems the field tends to address (which can be found in the related work).
20250825-20250831 Week 129
- Studied Streamlit, building a frontend for experimental analysis
- Continued writing the qg proposal
- Optimized the lab website's management system
20250818-20250824 Week 128
- Organized PTA problems
- Attended Dr. jzm's talk
- Helped revise cy's paper
20250811-20250817 Week 127
- Tested and revised the Anthropic issue tests
- Wrote the hwhyl proposal
20250804-20250810 Week 126
- Wrote the qg proposal
- Prepared papers for the group meeting
20250728-20250803 Week 125
- Organizing work, writing the paper
July 2025
This month was filled with pushing forward the development of sqtp and the experiments, while writing all sorts of proposal materials -- altogether quite a substantial month. I gradually got into the habit of ordering a coffee each day, which appropriately lifts my work efficiency. While writing the second paper, I really discovered that you must just write the content first without overthinking; in any case, everything will be revised over and over later, so there is no need to weigh every word from the start. The pace of reading papers also accelerated dramatically: I run a SPARQL census on dblp first, then read each Abstract and Intro one by one, all the while using SurveyGo to summarize.
The weekly report was originally meant to summarize new things I had learned, but it has unexpectedly become a plain working log. I will try, starting next week, to record what I have learned as well.
20250721-20250727 Week 124
- Organizing work, writing the paper
20250714-20250720 Week 123
- Produced the third version of the agent experimental code, only to discover that I really must run through the static source code adaptation phase first
20250707-20250713 Week 122
- Refactored the agent experimental code
- Organized scholar papers, wrote a proposal
20250630-20250706 Week 121
- Push the experiments forward!
June 2025
In the early part of the month, the experiments advanced rapidly. I switched to Claude, Gemini, and other models for testing -- the API calls really are expensive. By mid-month I returned to setting up the environment, considering an approach of locally expanding macros plus annotating the source code to minimize the problem of unsupported features... we still lack a complete tool built from scratch.
At the end of the month I started reading Mao's biography and Mao's selected works. They were genuinely inspiring!
20250623-20250629 Week 120
- Continued pushing forward the research experiments
20250616-20250622 Week 119
- Pushed the research experiments
20250609-20250615 Week 118
- Took on the burden of LLM-based development for branch meeting record review
- Participated as a volunteer for this year's Three-In-One admissions, mainly responsible for guiding people
20250602-20250608 Week 117
- Wrote the first draft of the slides for the following Sunday's themed report
- On the weekend, in my role as a part-time counselor, I took on the task of Tutoring System 1 and rushed through the project in a single day
May 2025
The start of the month was wildly busy! Reviewing papers, writing the gossip submission, making the SP slides, organizing materials for the software security competition. In the middle of the month I went to San Francisco for a conference, and upon returning, without a moment's rest, flew to Changsha for the finals. I was so busy that my feet barely touched the ground, and I was thoroughly dizzy. The result, however, was not bad: I picked up a national first prize. I will write a separate travelogue for San Francisco once this stretch of busyness has passed.
20250526-20250601 Week 116
- Organized the group meeting PPT
- Wrote the proposal for project metrics
20250519-20250525 Week 115
- Organized reimbursement materials
- Pushed forward the research, getting batch experiments running first
20250512-20250518 Week 114
- San Francisco conference
- Participated in the Changsha competition
20250505-20250511 Week 113
- Made the slides for Oakland
- Printed the poster, bought all sorts of necessities for going abroad
- Wrote the Huawei proposal
April 2025
At the start of the month, I was assigned the part-time counselor task of tutoring students in the deferred-graduation class on Compiler Principles. Undergraduate education quality these days is much better than before: there are very detailed lab manuals available. I took the chance to revisit and brush up on the material myself. Though after one or two days of contact, they did not seek me out anymore.
By mid-month it began to feel a bit like a frog being slowly boiled. The basic demo for my work was still not implemented; I was stuck on the question of formal verification generation quality. I went to Beijing for an exchange with the Zhongtong Group.
At the end of the month -- I really should read more of the recent work on LLMs and specification generation, and look for some inspiration.
20250428-20250504 Week 112
- Slogged through reviewing papers
- The branch took on a task of using LLMs to review life records
- Wrote the poster and a gossip article
20250421-20250427 Week 111
- Completed the work on iterative annotation
- Organized notes from reading several papers
20250414-20250420 Week 110
- Migrated the lab server and wiki to a Raspberry Pi
- When stress is high, going out with people for a meal really does help
- Went to Beijing Weitong for an exchange
20250407-20250413 Week 109
- The research basically completed end-to-end analysis on a single commit, but the results still need optimization
- Bought a new domain and deployed a wiki for the lab
- Helped make some slides
20250331-20250406 Week 108
- Continued pushing the research; tried CodeQL and found it too inefficient, switched to Clang
- Helped tutor Compiler Principles as a part-time counselor task
March 2025
The start of the month was fine; by mid-month I was wildly busy. I had to prepare for the mid-term defense, the group meeting, a stack of proposals, and the software security competition. I came to realize that you really should make slides regularly; they can be solidified once and applied in many places. At the same time, this also improves slide-making skills (which I hear are very useful in industry), aesthetic sensibility, and the ability to refine and condense information.
In the middle of the month I took part in the software security competition. The competitors in this region were "all rather weak", so I took first place in the regional round. After optimizing the system page later, I am waiting for the May finals at Hunan University.
At the end of the month, the consulate finally sent an email asking me to submit my passport! I took a day to head to the Shanghai CITIC Bank to submit the materials in person. The major business of March can pretty much be considered concluded; next, I return to pushing the research forward!
20250324-20250330 Week 107
- Finally finished the check; the day after receiving the email, I went to Shanghai to submit the passport
- Wrote the research code for using LLMs to generate ACSL annotations
20250317-20250323 Week 106
- Took part in the mid-term defense, with an excellent rating!
- Took part in the software security competition and won first place in the regional round!
20250310-20250316 Week 105
- This week I continued to prepare the mid-term review slides and the group meeting on Sunday -- wildly busy
20250303-20250309 Week 104
- A frantic week of making mid-term slides, sigh
- A frantic week of writing proposals, sigh
February 2025
At the start of the month I went to Shanghai to apply for a visa, and along the way met up with classmates -- though due to a sudden cold wind blowing in, plus a flurry of work coming from the lab, we did not really go out much. The visa came back as a check, and I hope it will be granted smoothly within a month. While reading on the high-speed train I suddenly realized I had not felt the joy of skim-reading for a very long time. Usually I read material very meticulously, which has gradually slowed my reaction time and left me less efficient at completing tasks. Excessive focus on details leads to neglecting the higher-level content -- it is not a good habit.
Browsing the campus forum can indeed sync me up with new information, but it is best treated as a search engine; otherwise it is easy to fall into "wading through 98", which not only wastes time but, more importantly, scatters one's attention in such a way that it is hard to recover for half a day.
In the middle of the month I got hooked on table tennis, playing one to two and a half hours every day until the gym closed. Exercise!
20250224-20250302 Week 103
- Used an LLM to test annotation generation
- Took part in the development of the "study and practice" website organized by the younger students
20250217-20250223 Week 102
- Pushed forward my own work; due to a switch in technical approach, I reorganized the dataset
- Yuque for organizing work is rather nice, and drawing data tables is also very convenient
20250210-20250216 Week 101
- Wrote binbin's proposal
- Wrote the lab proposal
20250203-20250209 Week 100
- Went to Shanghai for the visa interview, picked up a check; the check slip is a full A4 sheet, which is a bit different from what is on Xiaohongshu
- Made the sec vue code public
January 2025
The old year ended with working in the lab through the New Year's Eve countdown, and the new year began with working in the lab on New Year's Day -- truly wonderful. No one came to the lab in the morning, but in the afternoon two more showed up. As expected, everyone is a quality work-cattle.
Mid-month, I reviewed my recent research progress and felt that doing analysis at the IR level was not quite working out. The main issue was that, in the research scenario, the cost of recompiling the entire project for every commit is excessive, and it is also hard to migrate between projects, plus my LLVM is not very practiced and the development pace is poor. Better to try analyzing from the source code level instead.
At the end of the month I went home for the Spring Festival. I had originally planned to use the holiday to push through the slides for the proposal defense, the mid-term, and the May conference all in one go. Later, I was assigned to organize lab materials, and on top of that I had to be away nearly every day during the New Year, so it really came to nothing in the end...
The biggest change from going home was sleeping early and rising early -- in bed past ten each night, and up past seven the next morning!
20250127-20250202 Week 99
- Went home for the New Year
- Pushed forward the research, attempting to solve the issue of unsupported FC features
- Pushed forward the Spring Festival proposal
20250120-20250126 Week 98
- Organized the visa materials
- Refactored the sec website according to sakai
- Drafted a vision for the development of a study-and-practice website
20250113-20250119 Week 97
- Tentatively decided to use FC for the research, started learning the tutorial
- The lab's team-building trip to Moganshan
20250106-20250112 Week 96
- Finished binbin's proposal
- Rethought the technical roadmap for the research
20241230-20250105 Week 95
- Processed the C-tier grades, helped the kids salvage and record their scores
- Pushed forward the front-end and back-end code for the software security competition, and organized registration materials