One-liner
I started reading some interesting papers and thinking through a few ideas.
Time feels too precious, so I did not finish many papers this week.
I started thinking about what time really is.
I always feel there is not enough time, but what am I actually spending it on?
That is a question worth thinking about and discussing.
What I did
Work
-
Tech
-
- I started thinking through the best incident handoff/triage systems and methodologies in the industry: YouTube talk.
-
-
-
Life
-
- A quick review of my sleep this week: on Wednesday morning I had the best sleep score I've ever recorded, 97. But my weekly average is still below the 8h Garmin recommends. Thinking back to that night: I played Honor of Kings for a long time before bed and got pretty mad, so my mood wasn't great. That makes the score feel a bit confusing. Anyway, I should try to sleep better going forward. I keep feeling that when I sleep very deep it's because I need deep sleep — like my body has been worn down to a point where it basically says, "sleep better, now."
- Exercise was a bit light. I only started a 1km run on Wednesday night. It's cold today, but it should still be suitable for running. The best time to start is now.
- Did one hour of cardio on Friday night.
<Social/family: one high-quality interaction>
Played Switch with my partner — we finally made it happen. We played Kingdom Two Crowns together.

- Attached.
-
Thinking
-
Output
Writing / Notes
-
Code / Projects
-
Reading / Videos
- <Book/article/video + one takeaway>
Review
Keep doing
- I started playing Switch this week. I think it was one of the right things I did, because it means my creativity is starting to open up. While gaming, some creativity naturally got unlocked, and it also became a way to examine my inner state—though we ended up playing for quite a long time.
Stop doing
- I probably played games for too long.
One lesson
One small thing
Next week plan (Week 1 of 2026)
-
-
- Solve a MongoDB high-CPU issue. I tried introducing an index (a filtered/partial index), but CPU kept spinning and utilization reached 98%.
-
- Paper: The Llama 3 Herd of Models
- Paper: Revisiting Reliability in Large-Scale Machine Learning Research Clusters
- Paper: From bare metal to a 70B model: infrastructure set-up and scripts
- Paper: SuperBench: Improving Cloud AI Infrastructure Reliability with Proactive Validation
- Paper: xpu_timer profiling tool
-
Comments