Thinking back to my community college days in 2020 (right before the pandemic!), I used to rock a crappy old dual core laptop with 4GB of RAM running Fedora. Browsing the web was less than ideal and YouTube barely worked. For certain classes (the big hassle was a DB class after I transferred) running the course software was very taxing and I had a harder time with that compared to other students as a result. I remember Zoom barely working on it, but only if I didn't use the camera. I look back at this experience a bit fondly as it made me rely more on the terminal and I shifted away from slow laggy GUIs to TUIs/NeoVim. That said, for most other individuals in that position it really is a big limiting factor for their education.
Recently I was tutoring (for college essays and math) at a local high/middle school and most students browsed the web on their smart phones, but those that didn't were limited to school issued Chromebooks which were ungodly slow. Some of these students served as translators for their parents and I was under the impression that they used these devices to pay bills and for other household tasks as well. This experience is why to this day I try to keep the websites I make light on dependancies (I don't like react) and fast to load (sub 300kb ideally).
It feels like every week someone discovers that code closer to the metal is faster, and ignores the fact that it’s easier to write higher-level code, and that this is the specific tradeoff everyone is making. Why do these articles keep getting posted? Who is this a revelation to?
I build my apps the same way: mostly server-rendered HTML with a little JavaScript on top (Hotwire on Rails). They're mostly forms and CRUD, and it's worked for years with almost no client-side state to debug. I avoid React when I can. The frustrating part is there isn't much Hotwire or Rails work around these days. Most of the jobs want React.
You can also get really small file sizes by ignoring frameworks and porting unsupported codecs to wasm.
I ported jbig2 (17kb uncompressed) and codec2 (60kb uncompressed) to wasm which enables me to use really small image and audio files in my web app. I also made a custom read only database and search engine with built-in zstd decompressor (39kb uncompressed). It probably wouldn't run on a psp though.
I like optimizing and making things small. I want to use neural audio codecs for even better compression but the model sizes and compute complexity are major hurdles and muddy the vision.
I hope the image and audio bandwidth savings are more than the cost of having to send down the codec JS payload - might be worth it for a web app with a lot of image/audio loading
I get all my news by reading: NYT, WSJ, BBC, AP, and a few Substacks.
NYT has gotten so bloated both in actual content (long meandering interviews) as well as the encapsulating junk. I started to put it into Gemini to summarize. During this process I found, the two hour interview transcript was 50% actual words and 50% junk (probably formatting, tracking, adware, surveillance?).
13 comments
[ 2.0 ms ] story [ 55.5 ms ] threadHeavy CSS animations, Javascript bloat is what slows down the websites and literally ruins the web experience.
Recently I was tutoring (for college essays and math) at a local high/middle school and most students browsed the web on their smart phones, but those that didn't were limited to school issued Chromebooks which were ungodly slow. Some of these students served as translators for their parents and I was under the impression that they used these devices to pay bills and for other household tasks as well. This experience is why to this day I try to keep the websites I make light on dependancies (I don't like react) and fast to load (sub 300kb ideally).
I ported jbig2 (17kb uncompressed) and codec2 (60kb uncompressed) to wasm which enables me to use really small image and audio files in my web app. I also made a custom read only database and search engine with built-in zstd decompressor (39kb uncompressed). It probably wouldn't run on a psp though.
I like optimizing and making things small. I want to use neural audio codecs for even better compression but the model sizes and compute complexity are major hurdles and muddy the vision.
NYT has gotten so bloated both in actual content (long meandering interviews) as well as the encapsulating junk. I started to put it into Gemini to summarize. During this process I found, the two hour interview transcript was 50% actual words and 50% junk (probably formatting, tracking, adware, surveillance?).
The junk made me feel dirty.
Some discussion at the time: https://news.ycombinator.com/item?id=25915313
And a more recent incarnation or variation:
Using Claude Code: The unreasonable effectiveness of HTML
https://news.ycombinator.com/item?id=48071940