Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
jjcm
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
21 ms
·
331.
▲
by
jjcm
1y ago
> I was on Microsoft insurance which was quite gold plated at a time One of the biggest perks of working for Microsoft for a long time was their health coverage. I can't tell you the number of times I'd be doing initial paperwo
332.
▲
by
jjcm
1y ago
A solid overview with some great tips. One recommendation is to change the mental model here. In my eyes it isn’t “don’t use a framework”, it’s “build the framework you need”. It’s tremendously easy to set up an extension of HTMLElement tha
333.
▲
by
jjcm
1y ago
To me the craziest thing about this is that often times web apps with the same functionality feel leagues faster to load, with nearly equivalent functionality. Google docs/google sheets/notion/coda all load faster than the na
334.
▲
by
jjcm
1y ago
Fun exercise in constraints, but why limit yourself to 12 bit color? Was this a technical requirement?
335.
▲
by
jjcm
1y ago
Recraft also has an svg model: https://replicate.com/recraft-ai/recraft-v3-svg One note with these is most of the production ones are actually diffusion models that get ran through an image->svg model after. The iss
336.
▲
by
jjcm
2y ago
I was playing around with this a ton yesterday. One thing I was attempting to do was create an emoji that only had partial portions that had their brightness maxed, while the rest of the emoji displayed "normally". It turns out th
337.
▲
by
jjcm
2y ago
Incredibly well done. FWIW, the video hack is no longer needed. Originally that was required due to browsers only having hdr support with video, but recently support for PNGs were added as well. You can just use an all-white png with the re
338.
▲
by
jjcm
2y ago
Aside from the more ambiguous access to newer features first, is there any benefit to this over just leveraging the API? Going via the API with your own interface seems like it's objectively the better way to handle things like this at
339.
▲
by
jjcm
2y ago
Huge +1 to Australia, speaking from a salary worker PoV. So many of the Australian digital services are absolutely fantastic. As an American who was working there, there were often situations where I thought I did something incorrectly beca
340.
▲
by
jjcm
2y ago
Some other recommendations depending on what you liked about The Egg/Project Hail Mary/The Martian: If you liked explorative / conceptual Sci Fi, more for the thought exercise than anything else: - Exhalation by Ted Chiang. A
341.
▲
by
jjcm
2y ago
Obviously a huge bias here (I work for Figma), but it’s one of my favorite things about Figma Slides. The product still has a ways to go, but man being able to actually be collaborative and not feel like you’re fighting against the software
342.
▲
by
jjcm
2y ago
Beautiful build, and very close to my dream keyboard. Excited to see the v2 of this. Great writeup as well - thanks for sharing!
343.
▲
by
jjcm
2y ago
I definitely get the proactive response here, as I’ve considered the same for my small platform. The biggest issue is the definitions of who the majority of the requirements apply to is quite hard to find. It’s buried on page 65 of this pdf
344.
▲
by
jjcm
2y ago
I know LoC is a terrible metric, but it always shocks me that things like react-select require 30k LoC. 21% the size of the code of the Apollo Lander to configure/style a select dropdown.
345.
▲
by
jjcm
2y ago
One slightly different approach I haven't seen mentioned here - I use grow lights in my office. They're way brighter, so simulate sunlight better, are full spectrum, and as a bonus my plants are super happy. I use 5 of these with
346.
▲
by
jjcm
2y ago
Are you raising a funding round? I'm bought in. This is hilarious.
347.
▲
by
jjcm
2y ago
In addition to the many others mentioned, here's a script I just threw together that simplifies a lot of these chained commands - llmpeg: https://github.com/jjcm/llmpeg If you have ffmpeg installed and an OpenAI e
348.
▲
Show HN: LLMpeg
(github.com)
169 points
by
jjcm
2y ago
|
80 comments
349.
▲
by
jjcm
2y ago
I definitely get what you're saying - there's an element here of taking what a customer asks for and returning something different, but I think it's an imperfect analogy. It's not bringing them a Coke, it's bringing
350.
▲
by
jjcm
2y ago
Disclaimer - I used to work on Bing like... 8 years ago. There's probably some debate around whether this is nefarious or genius, but I'd lead towards the later. "google" has always been one of the number one search term
351.
▲
by
jjcm
2y ago
This proposal also includes the ability to update a target DOM element with the response from that delete action.
352.
▲
by
jjcm
2y ago
Looks great - exactly what I needed. I typically just revert to using ffmpeg directly, so something this simple is all I needed. Only request is 2 devices is fairly minimal for a "personal extended" license, which is likely target
353.
▲
by
jjcm
2y ago
I made a thing that told me the number of people who are more internet famous than me is equiv to the population of Albania, and lets you calculate how (ir)relevant you are. There are way more people who are micro-famous than I was expectin
354.
▲
"Do you know who I am" calculator
(html.non.io)
2 points
by
jjcm
2y ago
|
1 comments
355.
▲
by
jjcm
2y ago
The results of this article are going to be fascinating. Realistically, WSJ has a far wider audience than the tech echo chamber, and the general public is only aware of GPT, not o1/o3. Outsiders will likely read this article and think,
356.
▲
by
jjcm
2y ago
Where are you seeing the 150M number from?
357.
▲
by
jjcm
2y ago
> 10.7k TWh globally Agree I hate this, but at the same time I don't know if I would have groked it correctly on first read if it had listed "10.7Pwh globally". We simply aren't exposed to numbers at that scale on a r
358.
▲
by
jjcm
2y ago
This is me. I cycle on and off testosterone (100mg/w for 12 weeks typically) and combine it with light exercise (20-30min of lifting 3x a week). Other than that my only exercise is walks with my dog (typically ~45min). The rest of the
359.
▲
by
jjcm
2y ago
There was no training on existing apps. We launched the feature in May and we didn't start training on data until a couple months ago. We used off-the-shelf GPT4o for the feature with no fine tuning. How it works is we feed in a design
360.
▲
by
jjcm
2y ago
We’ve since re-enabled it after addressing the underlying issue. Source: I’m the pm who launched the feature.
More ›