4
points
platelminto
Joined 264 karma
- Uhh isn't default HN always 30 items?
- I was 100% on board with WordPress/WC at first, and had already started building with it, but was immediately coming into issues. Just Work (tm) was my expectation, and was most definitely not what I found.
- I used the product variations feature, 18 variations per product, and all of a sudden the "Duplicate" button took 15 seconds! I learned this is because each variation is it's own thing, so it was making 18 new things (still insane it took that long, on my beefy dev pc). I can't imagine 30-50k products * 18 variations * metadata stuff working fast in any way.
- In avoiding product variations, there's plugins for adding product fields, and plugins for pricing rules, but clicking around to do stuff, or maybe writing php that integrates with plugins that I'm clicking around in... it's not the way I want to spend my time developing. It especially integrates terribly with AI tools, which at this point are an important development tool for me.
- I don't want to have a 1-to-1 mapping between products and pages. This doesn't fit the WC model well (or Shopify for that matter).
Generally, I can imagine an experienced wordpress/PHP dev being able to overcome these issues, but if I'm learning something anyway, I'd personally rather learn a proper frontend framework (be it any of the options you mentioned). Leveraging AI tools also matters.
I appreciate your response! Gives me more confidence in maybe sticking to Django + templates. But from what I've seen, and also in discussions with other developers, I think wordpress is out for this project. Thanks again :)
- I was about to start a new project with Next.js... is anyone willing to give me some advice?
I'm about to start building an e-commerce site (30-50k poster print designs, i.e. no inventory), and was leaning towards a Django backend (because I know it) and... some sort of SSR frontend. I'm not really a frontend guy, but taking this as an opportunity to learn it. This article obviously does not inspire confidence in me choosing Next.js - would someone have any suggestions/pros and cons of what to use?
I currently see the options for doing SSR as:
- Next.js: well-represented in AI training data (though recent versions had breaking changes? I'm not sure), but annoying to actually use (according to this article/general sentiment I've found online), and pushes you into Vercel? (I barely know what that means)
- SvelteKit: best DX and nice to use, but might be less present in AI training data?
- Django templates + HTMX: possibly limiting? Less maintainable once you get to a certain size? I'm not sure.
- Other options?
- Night lenses! Yeah they're pretty crazy (I'm in the process of getting them and a friend of mine has them). 10 hours is low though - they're supposed to easily make your vision last all day, even two. My friend says he only really stops seeing well after 3 nights of not wearing them.
- Tried it out for a bit - recently upgraded to Max so was willing to try one of these run-stuff-in-parallel tools.
It wasn't great.
- Installation using the provided binaries just fails on my machine - I have Ubuntu 22.04, which apparently has too old a version of glibc. Building from sourced worked though.
- Every time I want to open a new chat, it brings me back to the project list. I don't want to click on the same project every time!
- Scrolling is awful! It's slow, and it often doesn't automatically scroll down as the chat is generated so you have to do it yourself.
- There's no title or anything across sessions. If I'm now working on multiple things at the same time, I want to know what I'm working on quickly!
- The log/text entries take up so much space. Something like this would benefit from a much more compact view - it shouldn't use my entire screen to show me 1 TODO list and 1 tool use.
- Unlike the video, the code changes are all wrapped in a "AI Summary" entry which tells me what it did in a few words, with no option (that I could find) to open the code itself. Confused, couldn't find a setting for this.
- There's multiple UI bugs, and it's sluggish overall.
I didn't use the Agents stuff, which (given the video starts with it) might be the main focus? But as it stands, for my attempt at running multiple Claude Code sessions at once, this was too buggy to really work. Someone else mentioned https://conductor.build/, which might be more what I'm looking for, but unfortunately it lacks Linux support.
I hope it gets better! I could see myself using it after a few more releases, and I'm rooting for them - just sharing my experience here for others who are considering trying it.
- > some may argue that this was a good thing
I don't think anyone was arguing this - Pebble simply went bankrupt. FitBit just bought some of their IP/assets I think. There was no expectation of them buying it and continuing support or development.
- It will kill another living thing though. It's not that hard to understand - you know it tastes good but don't want to cause direct suffering.
- I think this removes any amount of human-labeled data: no RLHF and stuff like that. You can use their technique to create an unsupervised reward model, and use that model to RL your way to having a useful assistant LLM.
The paper is very accessible (it's mostly written by Anthropic researchers), and Section 4 summarises their findings really well. They were themselves really surprised by the results:
> We were initially very skeptical of these findings, because they seemed clearly too good to be true, and suspiciously close to training with actual labels. To ensure we didn’t accidentally train on the labels, (1) we re-ran the experiment several times on different datasets, (2) we copied the dataset into a new file, excluding any labels before re-running our algorithm with that file, and (3) one coauthor independently replicated the findings on the Claude 3.5 Haiku base model using a different codebase.
(emphasis mine)
- Surprisingly, it didn't infer anything from my protonmail email address.
- From what I understand these medications make you want to eat less in the first place, so it's not quite the same thing.
- And they're advocating for the end of 8th grade, so coming into Year 10. i.e. start of GCSEs.
- I remember reading somewhere that LLMs are actually fantastic at reading heavily mistyped sentences! Mistyped to a level where humans actually struggle.
(I will update this comment if I find a source)
- Sonnet 3.5.
Opus is the largest model, but of the Claude 3 family. Claude 3.5 is the newest family of models, with Sonnet being the middle sized 3.5 model - and also the only available one. Regardless, it's better than Opus (the largest Claude 3 one).
Presumably, a Claude 3.5 Opus will come out at some point, and should be even better - but maybe they've found that increasing the size for this model family just isn't cost effective. Or doesn't improve things that much. I'm unsure if they've said anything about it recently.
- Never really seen it in several European countries
- This is very typical of New Yorker pieces - I enjoyed the broader context of capital surveillance. We can’t editorialise titles here, so this is bound to happen.
- I think there might be a level of hindsight being used here - I distinctly remember when the Watch came out that very few people actually thought it’d be useful, and other similar criticisms that are being used against the Vision Pro now.
(not that I stand one way or the other w.r.t the Vision Pro, just that the Apple Watch was definitely not seen as anything much when it first came out)
- Within a year of launch, nearly all my friends use ChatGPT on a weekly basis. Granted, I'm a university student, so I'm sure I'm in quite a specific bubble, but still - it already feels quite ubiquitous.
100M weekly users too. I don't see how it won't become ever present.
But sometimes pure tech really scratches that itch, you know? Startup culture sounds super fun! Or I spend too much time on hn, idk. I'd love to make use of my relatively generalist skills, and just try to make something that people use. Ideally, even improve their lives.
I'm probably being way too optimistic on what I can realistically find, especially at this stage of my career, but hey you gotta start somewhere. If anyone's reading this - hi! Thank you for reading whatever this is, I think I roughly just want to give an idea of my uhhh... vibe.
Final (serious) note, because at the end of the day I am looking for a job: I try to write clean, performant code, using the right technologies for the job. I care about creating well-designed, intuitive programs. And I care about getting things done! If that's a piece of software, or using software to achieve something else, I'm happy to do it.