danabra.mov
rain, glitter
8,161 posts
57,316 followers
1,035 following
Regular Contributor
Active Commenter
comment in response to
post
re: error boundaries, React SSR assumes the error may be recoverable on the client (e.g. if the error is transient or server-specific) so SSR sends the closest <Suspense> fallback instead in HTML, and the client is the one actually doing the retry. if it fails *then*, it goes to the error boundary
comment in response to
post
you can view RSC as dynamic bundling :)
comment in response to
post
not excited about it being another platform with its own closed data layer! i want it on atproto
comment in response to
post
server/client waterfall avoidance is crucial though, without it you can’t safely compose components. see also overreacted.io/one-roundtri...
comment in response to
post
maybe! i don’t know the plan. but i want it to feel more centered around the post format. i want posts to have first class treatment and more or less appear as embeds
comment in response to
post
the cool thing about bluesky’s tech is that a feature like this can technically be built completely in userland. and then either be a separate product (“seeing” the same data) or even a fork of the original (open source) client incorporating the new features (which would still require own backend)
comment in response to
post
i mean more something like this
comment in response to
post
actually no! maybe i should but i really want the atproto universe content-wise
comment in response to
post
i do want it to feel bluesky/twitter-ey too though. so i think the front page and the detail page should give posts a first-class treatment (kind of like quote embeds?). the shell is reddit, the core is bluesky-like
comment in response to
post
i think i want it organized similarly to reddit. so there's different sort modes for the channel/community front page, you can submit a post to it, etc. biggest thing i’m not sure about is where the "comments" should live. they should probably be rooted in the "community" post unless you click in?
comment in response to
post
these seem to be roughly in the direction i’m interested but i want something that somehow "feels" closer to reddit if that makes sense. how these are organized doesn’t map very well to my brain yet
comment in response to
post
i think sometimes but you shouldn’t need to force it unless you specifically want to create a layer for some concrete purpose? like some layout oddity
comment in response to
post
what do you recommend instead? this is a personal blog. to my eyes this is the exact most direct solution to this type of problem
comment in response to
post
oh i think it just inferred it from my prompt at the top
comment in response to
post
yes pls
comment in response to
post
yeahhh
comment in response to
post
i should learn to code
comment in response to
post
here’s the context behind it
comment in response to
post
is that still the case though? not sure how the market has moved
comment in response to
post
with quite a bit of AI related work moving into consuming the model (building agents counts as that too) i think there’s something satisfying about actually understanding from the bottom up. but idk if that understanding scales. i doubt i’ll ever have the patience / bg knowledge to read papers
comment in response to
post
i wonder if free course on youtube with embedded sponsor placement can generate money. how much do sponsors pay for that. does anyone know
comment in response to
post
i’m not actually sure if i have any better suggestions. i do like the ease of messing with notebooks, this kind of dynamism is important. i’m also not saying that i’d want to declare types everywhere, inference ftw. but i’m just surprised this is the language of choice for much of ML? or is it not?
comment in response to
post
hahaha yes i remember vbCrLf
comment in response to
post
yeah that makes sense! i just found the difference jarring until i realized it expresses the same thing
comment in response to
post
it’s actually kind of interesting. i crashed out on the first misunderstanding. the thing that confused me was that the diagram models neurons as connected "directly" to each other but the implementation really models it as a repeated mapping of an array per layer. the mismatch overloaded my brain
comment in response to
post
me: watch 2 hour video about backpropagation in basic python with basic math at 2x speed
brain: great. now it’s time for your soul to leave your body
comment in response to
post
it also reminds me how much i hate python (for totally irrational reasons) but it is what it is
comment in response to
post
what about it?
comment in response to
post
this one seems really good so far, i appreciate the pedagogical approach with focusing on the concept over efficiency
comment in response to
post
i guess i think of anything long-running on the client as a SPA. the same strategies to unblocking apply as normally in Next (e.g. prefetching, Suspense, transitions)