jwr

13 ms·

211.
by jwr 6mo ago
> you would be there all day before someone said "I wish I could replace the battery" Before that, you wrote "One of the most frustrating things about HN is that people seem so unaware of how idiosyncratic their preference
212.
by jwr 6mo ago
I recently bought a Supernote Manta. It's an e-ink writing tablet. Guess what: it has a back which can be opened, and its innards are easily accessible. I could pop in an SD card, and the battery can be replaced, too. It's thin an
213.
by jwr 6mo ago
I haven't benchmarked recently, but I believe the situation is roughly still the same. If anything, things might have gotten worse because of more overbooking going on.
214.
by jwr 6mo ago
> We also find bare metal to be a lot more performant in general I measured this several years back and never looked at virtual servers again. Since CPU time isn't reserved (like RAM is), the performance is abysmal compared to real
215.
by jwr 6mo ago
> mis-matched parens FWIW, I never get mis-matched parens when working with Claude Code in Clojure. I guess the key is to use clojure-mcp tools. If there ever is a mismatch during work (rarely, but I've seen it happen), claude fix
216.
by jwr 6mo ago
I work on a large Clojure codebase with AI, and I'm getting excellent results. Likely factors are code density, the resulting token density, and a lot of well-architected code that the AI can follow — I'm not sure exactly, but t
217.
by jwr 6mo ago
I am so thankful for Clojure. It enabled me to run a solo-founder business for the last 10 years in a sustainable and maintainable way. It allows me to manage a large complex codebase (client+server) because of low incidental complexity and
218.
by jwr 6mo ago
> Opus 4.7 uses an updated tokenizer that improves how the model processes text. The tradeoff is that the same input can map to more tokens—roughly 1.0–1.35× depending on the content type. Second, Opus 4.7 thinks more at higher effort le
219.
by jwr 6mo ago
> Google staff have no incentive either. Solving this won’t help with their promotions. I don't think people appreciate that this is really the key observation here. In large institutions, for anything significant to happen, there h
220.
by jwr 6mo ago
I gave up on trying to report abuse to Google, Amazon or Microsoft. It seems reports simply get ignored and the big providers do nothing. I hope the FSF with its weight and media presence can finally do something. Google, Microsoft, and Ama
221.
by jwr 6mo ago
This is incredibly stupid, but don't laugh at Spaniards: your (and my) lawmakers are equally likely to enact similarly stupid laws. It's mind-boggling how stupid the world can be sometimes.
222.
by jwr 6mo ago
I use the OpenAI API for everything. I think codex is more polished, but I don't really prefer anything: I haven't used them enough. I mostly use Claude Code.
223.
by jwr 6mo ago
I do the same thing on a MacBook Pro with an M4 Max and 64GB. I had problems until the most recent LM Studio update (0.4.11+1), tool calling didn't work correctly. Now both codex and opencode seem to work.
224.
by jwr 6mo ago
The world might be, I am not. But your first sentence is interesting: this should not depend on where someone is from. I think it's rather sad that it actually does depend on it in practice: if bombs are dropped on your head, you take
225.
by jwr 6mo ago
This. I get much more value than 90€ from my Claude Code subscription. I am willing to pay more for consistency and not having to watch my back all the time, because I might get screwed over.
226.
by jwr 6mo ago
That's not really "lying" — ARR is usually understood as your projected "Annual Run Rate". It's a useful metric, as long as it is understood that it is an estimate. But, in all honesty, all RR numbers are est
227.
by jwr 6mo ago
A reminder that Kagi, unfortunately, buys search index data from Russia and has repeatedly refused to stop doing so ( https://kagifeedback.org/d/5445-reconsider-yandex-integratio... ). This is unfortunate, because I woul
228.
by jwr 6mo ago
Instead of critiquing this immediately, let's consider that this gives a second life to hardware that would otherwise be thrown away.
229.
by jwr 6mo ago
It's funny, because if that is true, it should give pause to city planners and officials: people prefer smooth sidewalks :-)
230.
by jwr 6mo ago
FWIW, it's better to stop and talk to the dog. Dogs don't really want to bite you, most will just chase you and don't really even know why. If you stop and confront them, they are confused and don't know what to do next.
231.
by jwr 6mo ago
Back when I roller skated, the ultimate question: you see a person and a dog, several meters apart. Is there a leash between them? :-)
232.
by jwr 6mo ago
No, every bike path in a city inevitably has crossings or is laid out next to a sidewalk. People just do their random-walk thing (Brownian motion, really, sometimes) and wander into the bike path.
233.
by jwr 6mo ago
I would strongly recommend NOT following the general advice of buying a cheap "3018" or something similar. Makera Z1 should be the baseline. Otherwise you're stepping into a world of frustration where you will spend most of y
234.
by jwr 6mo ago
> A horn or bell is mostly for telling other people "hey I'm here, stay out of my way and dont suddenly cross into my path" This. I only use the bell on bike paths, too. Sometimes it feels like a game of pac-man, where bad
235.
by jwr 6mo ago
I've been trying both Whisper v3 large and Parakeet in MacWhisper, and I inevitably go back to Whisper large. Which one is better depends on what you dictate, how you speak, and which languages you use.
236.
by jwr 6mo ago
Of course, one should always read the MSDS. I use a 3M respirator with VOC inserts while working on these things. However, one should mention that a) polyurethane resins and platinum-cure silicones are much safer than many other compounds,
237.
by jwr 6mo ago
This guide is legendary. It helped me to learn and follow the process, and I made some pretty successful parts. For example, my own keycaps for all keyboards that I use: much larger than usual (for my large fingers), with a pleasant gently
238.
by jwr 6mo ago
I never managed to get anything useful out of opencode, to be honest. I tried it many times, with various models. Claude Code always just worked better.
239.
by jwr 6mo ago
I currently use MacWhisper and it is quite good, but it's great to see an alternative, especially as I've been looking to use more recent models! I hope there will be a way to plug in other models: I currently work mostly with Whi
240.
by jwr 6mo ago
I wish they had a "and we won't screw you in two weeks" plan at, say, 5x the price. It's worth it for my business, I'd pay it. Should I switch back to API pricing? The problem here is that (I think) the instructions

More ›