Notes

All the short-form notes I've posted.

Things we learned about LLMs in 2024

Published:Jan 1, 2025

Great post by Simon Willi­son about the state of LLMs after 2024. Check it out here.

•  llm
•  bookmarks

---

Fixing blurry fonts on Zen Browser

Published:Dec 29, 2024

I’ve been en­joy­ing using Zen Browser lately, thanks to its sleek and user-​friendly in­ter­face. How­ever, I no­ticed that the fonts and cer­tain el­e­ments ap­peared a bit blurry in a dis­play with frac­tional scal­ing while using Fe­dora (Linux). It seems this issue isn’t lim­ited to my setup, as oth­ers have re­ported sim­i­lar prob­lems on var­i­ous op­er­at­ing sys­tems.

For­tu­nately, I found a straight­for­ward so­lu­tion! By ad­just­ing the zen.view.experimental-rounded-view set­ting to false in about:config, I was able to clear up the dis­play is­sues. The fix is in this issue.

Zen Browser is being ac­tively de­vel­oped by its com­mu­nity. If you want to try a fresh browser UI, maybe give it a try.

•  tips

---

Cognitive load is what matters

Published:Dec 25, 2024

The prob­lem is that fa­mil­iar­ity is not the same as sim­plic­ity. They feel the same — that same ease of mov­ing through a space with­out much men­tal ef­fort — but for very dif­fer­ent rea­sons. Every “clever” (read: “self-​indulgent”) and non-​idiomatic trick you use in­curs a learn­ing penalty for every­one else.
Dan North

Nice post about cog­ni­tive load in soft­ware en­gi­neer­ing.

•  bookmarks

---

WAL and durability

Published:Dec 13, 2024

Today I read an in­ter­est­ing blog post by Phil Eaton about write-​ahead logs and dura­bil­ity.

Check it out here.

•  databases
•  bookmarks

---

Deterministic simulation testing

Published:Dec 11, 2024

Today I learned about de­ter­min­is­tic sim­u­la­tion test­ing, an ap­proach to test dis­trib­uted sys­tems. The core idea is to re­move all non-​deterministic be­hav­ior from the crit­i­cal parts of the ap­pli­ca­tion, cre­at­ing a con­trolled en­vi­ron­ment where the se­quence of events can be ma­nip­u­lated.

This in­volves set­ting fixed seeds for ran­dom num­ber gen­er­a­tors, con­trol­ling clock and time de­pen­den­cies, run­ning tests in a sin­gle thread, and using asyn­chro­nous I/O op­er­a­tions. While these con­straints might seem overly re­stric­tive, they allow us to re­pro­duce bugs found dur­ing sim­u­la­tions.

It’s eas­ier said than done, though. Adding this kind of test to an ex­ist­ing sys­tem can be chal­leng­ing. There’s al­ways going to be some non-​deterministic edges. Also, the test is only as good as your sim­u­la­tion.

Read more in this blog post.

•  distributed-systems
•  bookmarks
•  til

---