doodlesdev

7 ngày trướcdiscuss

It has apparently been developed using LLMs. However, it's not vibe-coded as in no-thought development. According to the commit History, the developer first defined a design system and a product requirement document, which is one of my favorite ways of developing software nowadays (even if it's not necessarily "agile" or modern), specially whenever LLMs will be around the codebase.

doodlesdev

7 ngày trướcdiscuss

Regarding:

  > And acknowledges how people that identify as monogamous are not monogamous - they aspire for monogamy and have a deep discomfort with people that identify as nonmonogamous despite the aspirationally monogamous doing the same things in a harmful way

Could you elaborate further on this idea? At first, I just disagree completely, but I'm missing your reasoning and the actual thesis behind this idea.

doodlesdev

9 ngày trướcdiscuss

HTTP/3 uses QUIC as the transport layer, which in turn relies on UDP. QUIC replaces TCP while allowing a reduction of handshake exchanges in HTTP/3 first requests. Finally, even though UDP supports multicast, I believe QUIC doesn't. GP saying Google has developed it to use multicast thus is nonsense. Furthermore, QUIC takes much more CPU than TCP right now, due to running in userland.

In my opinion, QUIC and HTTP/3 are technical marvels, but are perhaps way too complicated and don't really serve the interest of most internet users.

There will be a point in the development of web browsers and associated technologies where we should just stop a bit to get things stable instead of churning protocol version after protocol version after new API. Will it ever stop?

Eventually, it all becomes so complicated no company can manage it all. Honestly, we might already be past this point, with Chromium at almost 40mi LOC, more than the Linux kernel itself, including all its drivers. When will the madness stop? Do we really need such complicated software to see Instagram posts, comment on a few Hacker News threads and mess around with Google Sheets?

The biggest reason I worry so much about this is that in the web, adding new features, APIs and protocols is easy. Removing and deprecating is basically impossible.

doodlesdev

11 ngày trướcdiscuss

That makes a lot of sense! I have no idea how I'd use that much money, so maybe the 128gb MBP for messing around with local LLMs wouldn't sound so absurd :)

doodlesdev

11 ngày trướcdiscuss

You're most welcome! I'll admit I did search before answering because you said Apple produced their memories so confidently LOL

I guess they would love to produce it themselves, but for the average scenario the production reserves they have with Samsung already work well enough and prevent them from having to get into such a complicated industry.

doodlesdev

11 ngày trướcdiscuss

That's incorrect. Apple purchases RAM from all of these providers to produce their unified memory. They also rely on TSMC fabs for all of the chips their memory relies on. If they haven't doubled the prices of their machines, that just goes to show how fat of a profit margin they take on everyone that buys from them...

doodlesdev

11 ngày trướcdiscuss

I feel like I'm going insane seeing people buy these 128gb MBP for thousands of dollars to run models that are objectively much worse than SOTA and spending so much more. The amount spent on a 128gb M5 MAX can buy you a damned new car here. What the hell am I missing? Are developers in other countries living in such different worlds?

(I'm aware the price is, in absolute terms, more expensive where I live compared to the USA. That reinforces what I think, because anyone sane that would've bought one of those in another country would sell them as soon as they landed here and save that money.)

doodlesdev

11 ngày trướcdiscuss

How much does one of those cost in the US? Here in Brazil, your notebook is worth as much as a used Honda Fit, which seems absolutely insane. For comparison, the ThinkPad I'm currently running cost me 1/20 of how much this MBP costs here, leaving me with over $8.000 to spend with LLM inference (if I actually spent money with that).

doodlesdev

12 ngày trướcdiscuss

Absolutely! That's why they are students: you learn to walk before you learn to run. Students should be forced to plan their texts and practice getting better at that. Whenever they leave school, they'll be able to write even better texts than the essays.

Honestly, school essays, as a text genre, suck to read. They are not meant to be enjoyable, beautiful texts! Instead, they are meant to help gauge the student's ability at text planning, the quality of their text project, their grammatical knowledge, and orthography.

I must say I'd be open to seeing schools attempting other models, though, and measuring the results thereafter to see in practice what happens when students type their essays instead. In terms of text projects and planning, that could be interesting and train different skills.

doodlesdev

12 ngày trướcdiscuss

And that's exactly the point! By making sure the student can't edit the entire text once its written, you force him to think about the essay's structure and force him to plan much more before writing :)

doodlesdev

13 ngày trướcdiscuss

Nowadays, these code forges have also become a centralized place for issue tracking, kanban boards, wiki editing and, specially, as CI/CD servers, in the case of GitHub Actions, which are, sometimes, the only for you to deploy software to package repositories. The same limitations apply to GitLab CI or Codeberg's Forgejo Runners/Woodpecker.

Whenever GitLab, Codeberg, BitBucket and, mostly, GitHub goes down, a lot of the software and websites you use can't be updated, including dependencies of your software that you're pulling from npm, for instance.

Finally, companies use code forges mostly for the ease of doing code reviews through Pull Requests/Merge Requests. Developers rarely, if ever, actually merge branches locally, before having it reviewed by peers in one of these code forges.

doodlesdev

14 ngày trướcdiscuss

I'm impressed that, in the meanwhile, Google has already thrown into the grave not one, but two different implementations of Material Design in the web: Material Design Lite [0] and Material Components for the Web [1], bot of which never managed to actually be competitive UI libraries.

edit: Actually, they've thrown a total of _three_ implementations into the grave, as MWC is in maintenance mode already [2].

[0]: https://github.com/google/material-design-lite

[1]: https://github.com/material-components/material-components-w...

[2]: https://github.com/material-components/material-web/discussi...

doodlesdev

19 ngày trướcdiscuss

The overall feature seems really solid, but I'm impressed they couldn't reduce the average package size further from 40MB even when not using CEF. I guess that wasn't a huge focus when developing this feature? Tauri and Dioxus can easily hit less than 5MB for package sizes.

I find the feature matrix comparison to be extremely well done and the sections beneath explaining advantages and disadvantages to be some of the best docs I've read recently.

https://docs.deno.com/runtime/desktop/comparison/

doodlesdev

20 ngày trướcdiscuss

Well... That's much simpler than I imagined LOL, thank you for answering!

doodlesdev

20 ngày trướcdiscuss

It famously is extremely memory leaky, with the core team having no idea how to fix it. With the new AI-automated unsafe Rust migration, this piece of slop may never actually become production-ready.

doodlesdev

25 ngày trướcdiscuss

Well, that's pretty impressive. Care to share your setup to do that? How much DDR3/DDR4 do you have, too?

doodlesdev

27 ngày trướcdiscuss

Since you are using LLMs to create the transcriptions, I wonder whether you've measured the difference in precision between the chosen model, Gemini 2.5 Flash-Lite, and newer/larger models such as Gemini 3.5 Flash, Gemini 3.1 Flash-Lite or GPT5.5.

I've read the README in the feat-api branch and, from what I understand, you've already assessed that false negatives are not a model failure, but I'm not sure I understand why (haven't spent that much time looking at it though, just curious to hear from you).

This is a really cool project, by the way! In my opinion this is a place where LLMs shine: produce the work of hundreds of hours of manual human labor much quicker and cheaper, for something that no one else would ever bother to do the work!

doodlesdev

tháng trướcdiscuss

Yup or restrict API usage to trusted providers. In Brazil, we have a system called Open Finance [1] which allows you to connect bank account, so you can see investments, money spent, credit card spending and limit, etc. from your other bank accounts. Some local personal finance systems integrate into Open Finance to pull all of this data for you.

[1]: https://openfinancebrasil.org.br/

doodlesdev

2 tháng trướcdiscuss

Arguably, Anthropic and Boeing are NOT what you should use to determine whether your coding practices are reasonable. Software problems with Boeing have literally killed hundreds of people due to faulty code and Claude Code is know to be a pretty buggy CLI (although extremely useful, sure).

Though I agree with the expensive factor. Perhaps, what I actually believe is that LLMs shouldn't touch code that's as mission critical as what NASA works on, even though it might be great to develop user-facing frontend software and CRUD backend code for huge corporations and projects.

doodlesdev

2 tháng trướcdiscuss

Which, by analogy, would mean you could use subsided tokens for over 10 years and then, and only then, actually purchase your own hardware to do inference, with over a decade of technological advancements to compound.

I'll invest in inference hardware whenever the economics make sense, not because of my prediction that prices will get higher (also, because smaller models keep getting better and they might just suffice for most of my use cases in a few years time).

My biggest worries in terms of cost is in regards to training. Whenever that gets too expensive for Beijing to pay, we won't be getting new SOTA open small models to run on local hardware, which, again, reinforces the decision to use providers for AI inference for now.

doodlesdev

Submissions

GitKraken Unveils Code Flow to Help Teams Navigate the AI Era

Choices

Private Processing for WhatsApp – Technical White Paper and Security Guide

Glyphhanger – Your web font utility belt

Fontello – Combine icon webfonts for your own project

gpui – A fast, productive UI framework for Rust from the creators of Zed

Terra - A rolling-release Fedora repository

CSS Grid Native Masonry Layout

Official TypeScript Cheat Sheets

Quart: Asyncio Reimplementation of Flask

A Git story: Not so fun this time

Firefox 145.0 for Android Release Notes

US Supreme Court allows order forcing Google to make app store reforms

comments