Weekly Reports

A friend and colleague from the Tu-Zhu group, Brother Jiao

April 2026

At the start of the month I was anxious to grab a summer internship, yet I also kept thinking that my paper output wasn't enough, that I hadn't yet met the lab's threshold for being released into an internship. Later I thought it through: even if I went into a big tech company and burned myself like a battery every day, only to be laid off in middle age with nowhere to go, with my health long since neglected, what would all that money even be worth? And in the age of Agents, an individual who seizes the opportunity well still has hope of becoming wealthy. Summer internships are too employment-oriented; perhaps I should try a non-employment, day-to-day research-style internship instead. Graduation doesn't necessarily mean going to a big company either. Finding a stable job, work-life balance, having an inner life full enough and time fully under my own control, is also pretty good. Besides, there's no need to cling to the mindset of interviewing for the sake of finding a job, much less the idea that I must pass before I'm allowed to interview. Don't put any expectations on yourself, and don't carry any pressure either. Just treat it as a casual chat, an exchange.

By mid-month I had a different thought again. Getting hung up on the act of "rushing the paper" is itself wrong. The paper isn't the goal, it's the result. What I should do is pose a question, then research it, then solve it. The result of being anxious about both internship and research output is that you end up grasping neither. Besides, are there really no positions you can enter without an internship? The key is that one's own ability must reach the standard of a PhD; some preliminary research positions don't require an internship at all. The main thing is to have real skill. Now, with the help of large models, I can patch up my weaknesses very quickly. The key is to discover where my weak points are, and where my strengths are. It feels like my interest is still in ops? Monitoring server systems, raising alerts on anomalies, and even developing dashboards and so on. But these are things any undergraduate could do, let alone an AI.

By the end of the month, yet another mood. You never know whether death or tomorrow will arrive first. Once life cools down and the soul drifts upward into the empty universe, what do the things that happen on Earth even amount to? If time is an unceasing rushing river, a human life is no more than a single splash of water it kicks up. In this brief life, since I dislike the disciplining of society, why must I tick off "find a good job," "find a fine spouse," "have children," "buy a house and a car," and so on, like completing assignments? Going down that road has very little resistance, but it does not seem to be what I actually want. Suddenly I can understand those scientists who "devoted themselves wholly to research, never marrying, never having children." Pushing out a tiny bump on the boundary of human knowledge is far more interesting than dutifully obeying the biological "instinct" to "reproduce."

20260427-20260503 Week 164

Submitted paper
Tidied up experiment code for open science
Used Claude Code to develop a Linux composite project

20260420-20260426 Week 163

Prepared for group meeting
Processed experiment results, revised paper

20260413-20260419 Week 162

Used Claude Code to refactor the lab server management network
Tried out sashiko for a code review
Polished paper, did empirical study

20260406-20260412 Week 161

Used Claude Code to organize previous project code, planned new experiments

20260330-20260405 Week 160

Used Claude Code to develop a Lixi Academy project
Reviewed papers for USENIX Security

March 2026

The age of LLM Agent Applications is changing too fast. Just guarding my own little patch of ground is nowhere near enough. How can I keep up with the tide of the times (or at least not drown in it)? I've noticed that GitHub really has been seeing a wave of interesting projects lately. Aside from research, I should also stay in touch with the latest developments in the industry, and see what everyone is paying attention to. As for the dazzling array of products, I should at least try them out and deploy them myself.

Sometimes it's not that one's ability is lacking, but that one's mindset is bad. I really do need some positive feedback. Doing things that flow with the current can, to a great extent, lower useless mental friction. For example: breaking the daily plan into many very detailed steps and finishing them one by one (establishing checkpoints); designing my own QAs to guide my thinking (testing intake). When doing things I love, I never feel tired. So where does that "love" actually come from? The fact that the thing itself is interesting? That the feedback after finishing is immediate? That doing it doesn't require grim struggle (a warning sign)?

Rather than honing so-called willpower, it's better to fundamentally shift cognition, and accomplish tasks while staying in tune with human nature. I've always rather enjoyed studying network communications and web application development; I can do it all day without feeling tired. But the moment I touch my research topic, my attention scatters and my efficiency drops. It's not that I dislike the topic, perhaps I just lack explicit feedback and don't know the route for improvement. At root, is it because my thinking has not been hammered enough? Or maybe the topic itself just isn't all that interesting. Once I find a small but beautiful angle to work on, I'm immediately energized.

There are some things that need to be thought through before doing. Doesn't an agent also emphasize "make a plan first"? And every day I should do a few things rather than only one thing, to lower the scheduling overhead of long-term tasks.

20260323-20260329 Week 159

Helped run the lab's closed-door meeting

20260316-20260322 Week 158

Researched the FP-prune topic
Deployed code researcher's work

20260309-20260315 Week 157

Continued debugging kGymSuite
Vibe-coded a "distributed" port scan project

20260302-20260308 Week 156

Deployed kGymSuite
Prepared for group meeting sharing

February 2026

The work environment at school suits me better. The home environment over the Spring Festival isn't really suitable for sustained office work. And it seems I can sleep for a full two hours during a nap at home, perhaps because the bed is lower? Between the year-end party and Chinese New Year, plus the fact that February only has 28 days, there really wasn't much effective working time. Facing the all-important coming semester, I have to pull myself together and think about what to do next.

As for the laziness produced by research, I had always thought it was caused by lack of positive feedback and scattered attention. Now it seems it might also be related to the body, that long-term lack of exercise plus high-calorie intake caused some kind of physiological mutation that no longer supports stable work? Add in the (possible) lipid threat, and exercise has to be put on the daily plan as well.

20260223-20260301 Week 155

Updated the lab publication's supplementary materials
Submitted revision

20260216-20260222 Week 154

Reviewed papers
Online meeting to discuss proposal writing

20260209-20260215 Week 153

Went home for Chinese New Year, reviewed three papers
Online meeting to discuss research progress

20260202-20260208 Week 152

Attended SECO2
Reflected on research feasibility

January 2026

When I'm tired of doing research, I can watch some video tutorials. Under multimodal input, that also counts as recharging.

Based on differing energy levels at different times of day, I can arrange different tasks, like reading papers in the morning and writing code in the evening.

I've realized that an important capability is rapid extraction and summarization of information, cultivating the ability of "downward compatibility." For instance, when reading some summary blog posts, most of the content is something I already know, so while reading I have to constantly remind myself: have I already understood this? Can it bring a new angle or a new line of thought? If the value is low, just skip past it quickly, without getting trapped in the surface of the text. Improve the ability to summarize information rapidly.

Does wearing headphones while working affect work? It depends on my state at the time. If after putting them on and working for a while I find myself getting distracted, I might as well not wear them. If putting them on isolates me from environmental noise (such as some of the otherworldly mutterings in the lab), then wearing them is rather nice. It's not that wearing headphones itself affects working state; it's that once focus is achieved, wearing them or not makes no difference.

20260126-20260201 Week 151

Tried to reproduce some related work
Tried to learn some interesting undergraduate CS courses

20260119-20260125 Week 150

Discussed the journal extension
Researched the concurrency bug problem

20260112-20260118 Week 149

Reported C-program statistics, handled course-completion matters
Analyzed target examples, summarized possible angles

20260105-20260111 Week 148

Continued pushing forward the proposal
Researched the next topic
Updated the tech site