theweeklyedition.xyz
Programmer working on a Pocket / Readwise replacement in React Native with Node.js Mobile https://theweeklyedition.xyz/
714 posts
130 followers
224 following
Regular Contributor
Active Commenter
comment in response to
post
*sometimes debates getting an AirTag*
comment in response to
post
This nonsense appears in the sidebar:
comment in response to
post
Fix: youtube.com##.ytd-channel-renderer block rule in AdGuard. Nice work YouTube, nice work.
comment in response to
post
(on web, Chrome and Brave, extensions disabled, Windows and Linux)
comment in response to
post
Thank you. Now I have a reason to *never* split my family off my personal account (we all share and deal with it, no one in the EU uses iMessage anyway)
comment in response to
post
Anyway, they've known for a year plus. I went back to X in meantime.
comment in response to
post
I was curious if this was a HEIC issue feasibly so searched for HEIC on bluesky and now seriously regret it.
comment in response to
post
You and Cyan should learn toki pona 😇
comment in response to
post
(thx to chatgpt, the others failed *horribly* at this assignment)
comment in response to
post
It even says it in the doc "(and are available to help in the analysis)."
comment in response to
post
You mean eg Sonnet Max? Or you mean a model they simply don't support?
comment in response to
post
I've also seen really good improvements with Gemini vs Sonnet recently. Together with o4-mini whenever Gemini just ain't quite smart enough for the task at hand. Gemini is far less likely than Sonnet to just entirely ignore instructions at least.
comment in response to
post
I switched to jj for this reason, I commit *every single change* now, no exceptions. jj at least makes the workflow doable.
comment in response to
post
Adevărat a înviat! (Romanian response)
comment in response to
post
related: x.com/lypnv/status...
comment in response to
post
Alternative: 192837465 for. every. single. password.
comment in response to
post
"big poopy farty head".
comment in response to
post
Grok3 was *vastly* better than anything else I could find on the FPGA CPU I'm working on though.
comment in response to
post
The fix was - based on a bit more testing - mostly coincidence and pulling the question out of Cursor. The chat history was really missing this one up apparently and with the Grok3 query most models reproduce roughly the same optimization strategy.
comment in response to
post
Recent Grok 3 question that other models failed *horribly* at: grok.com/share/bGVnYW...
(Claude decided to randomly depend on SDL3 headers without updating anything else, on top of producing the putrid original)
comment in response to
post
*nod*, 90% direct code writing, reasoning models for understanding of existing codebases / debugging etc.
Mostly RN, with some random .kt (thus the follow).
Recently: FPGA dev with Scala/Chisel, and a random 1 week project photo viewer than syncs over WebDAV w SDL2 in C++. ~0 XP in either.
comment in response to
post
Overall my fave models currently are Grok 3, o3-mini, DeepSeek R1 (best first). Wish Grok 3 would come to Cursor already as *hate* using it via the website.
I mix them with Claude though, as it's still by far the best 'agent' and I love that when have a framework that does hot module reloading.
comment in response to
post
Haven't tried Max yet as my usage has been a bit bonkers these last 2 months - I went through the 500 Premium requests in the first 2 days of my month... really getting the value out of Cursor haha.
Will give MAX a try at some point!
comment in response to
post
Tip that's helping me a fair amount: in cursor you can switch models randomly during a chat, so I ask o3-mini to plan work that Claude just ain't cut out for (it's kinda dumb) and then when o3-mini inevitably tells *me* to do the work, I ask Claude to 'execute the plan'.
comment in response to
post
My fave is when it quits the conversation.😂
comment in response to
post
Did the same to me, I noticed a printf at some point saying something like 'modifying broken test results' and got curious... since then no more 'vibe'ing for me, just treat it like a semi-malicious junior.
comment in response to
post
Finally, when Sunday arrives, on the Library tab, click 'Enqueue YouTube Collection' and everything you've collected that week will be added to the YouTube playlist that you've selected while configuring YouTube integration.
comment in response to
post
Next up, follow the normal procedure for updating subscriptions,
1) start up The Weekly Edition,
2) in settings, navigate to 'RSS Import',
3) at the bottom of the page click 'Import OPML'.
4) navigate to the Library tab, and pull to refresh.
5) you should now see working 'Generate ...' buttons.
comment in response to
post
Sorry didn't get this far in the thread 😂
I've heard *many* others complain about recent versions.
For me recent updates broke my vim keybindings plugin and cursor tab doesn't even work anymore. So can't exactly say Q&A is excellent alas...
(oh and the week when it uninstalled itself on startup..)
comment in response to
post
Is VS Code itself stable for you?
For me the underlying cause of this was that GPU accel needed to be turned off as VS Code is untested 💩. But, I'm on Windows so, results may vastly vary.
Since switching the option I've had exactly zero crashes on Windows, and I'm a *heavy* user.
comment in response to
post
Pretty much everything that could go wrong, went wrong during the installation flow. Meanwhile WSL2 w SDL2 Just Works in '25 which is kind of nuts as an long term Unix user (Linux, then OS X, Linux).
comment in response to
post
Amusingly I noted elsewhere I actually started a Photos project roughly the same time as Jay but ended up using SDL2 w WSL2 as the experience getting started with RN 4 Windows was so poor (4 hours, still didn't get it even compiling, admittedly, I'm a complete Windows dev newbie)
comment in response to
post
Depending on timing should be around also to give a hand, at the very least provide some Q&A as I'm a complete bug magnet.
comment in response to
post
Ah my machine is too piddly to have my full photo set on it (200+ GB).
One of the reasons I'm doing this now is because we recently found out that Photos on my wife's iPhone is making iOS upgrade impossible as Photos app space is proportional to image collection.
comment in response to
post
(choices that i might early that make this even feasible, 1) i don't do complex grid, it's square fixed with photo crop, i don't smooth scroll, movement is whole thumb at a time, amusingly this makes BURSTs look like slow FPS movies which is fun)
comment in response to
post
How are you doing thumbnailing etc?
I'm getting pretty good compression / quality via re-compression AVIF. 37GB of mixed photo/video -> 1GB 192x192, 384x384 JPG thumbs (JPG as v fast decode) + AVIF fullscreen (slow decode, high quality).
comment in response to
post
Yeah I'm thinking of something even weirder though. My niche is me haha and it's for searching for shit so...
Center: full screen
Left: single row of images, in same order as cursor will move
Right: grid of tiny images, with cursor zooming along first right then down
Automatic fade when unused.
comment in response to
post
Will first figure out a good way to do this in vertical space though, in full screen mode there are two massive black bars on the sides of my screen at least (and even bigger for portrait photos, which we have many of as phone).
comment in response to
post
Oh filmstrip view is a nice idea. Might copy! If I'm ever on OS X again I'll add WebDAV to yours rather than port 😂
comment in response to
post
Grok3 is the only non reasoning model that gets kinda close, but, doesn't quite get the cookie.
Claude / Gemini fail even with thinking enabled. The rest pass (QwQ-32B included).
Of course all models are inherently random and I didn't do sampling, just first result tests.