rablackburn

Finally, a well-formed question to which the answer "42 (times)" is a possible solution.

What a lovely example to end on.

Personally, I (guiltily) love using ol'fashioned contractions. I wish it were more normalised.

> I'm curious to see what you can do on an 8GB consumer GPU these days

Running smaller 4B-7B models entirely on the GPU VRAM will get you fast inference, but you will need to scope and define the tasks well. eg, using it the model as a classifier and just feeding it from a queue.

The best performing "agent"-like model to plug into a harness that I have found so far has been Qwen3.6-35B-A3B (mixture of experts) model as I can park most of it in system RAM and CPU, while the VRAM holds the attention/shared weights.

It's definitely workable as a local AI homelab. But expect homelab levels of tuning/fiddling with it.

With the improved support for AMD GPUs I'm finally considering getting a modern 16GB card (and maybe a second one in a few years assuming prices come down)

It doesn't have to chase it down, just keep plodding along implementing the new standards as they become accepted.

Google likes to make up its own "standards" and push them out on chrome, but Safari tends to be the late-comer to supporting impactful standards, not Firefox.

Do you have a particular use-case where Firefox's lack of support blocks you?

> Safari has much, much better compatibility, but even then this browser is hated for it's quirks

I think this meme has almost run its course.

Funnily enough I just had to fix a Safari layout quirk yesterday (for iOS 16 & 17 devices only). I was excited to pick it up because it had been _so long_ since I had to deal with a CSS-related browser incompatibility. I was nostalgic :')

I raced Claude to fix it. I knew which CSS had the issue, but I needed to set up an iOS simulator to verify my fix. So I had to download xcode, which download 10GB of data for a current simulator and Apple's embedded code-assist model. Then another 8GB for the iOS 16 simulator.

Claude took about 5min to find the right CSS file and then got stuck at the same point as me: needing to simulate it to verify. It just got the old webkit binaries and hacked together its own minimal browserstack clone.

It finished before me because of those downloads, but it only found half the fix. So... I won!

Except...I made a typo (`no-wrap` instead of `nowrap`) and had to do a fixup commit.

So I guess we'll call it a draw :(

> That’s at least a plus. I will happily burn through as much VC money as they will give me to tinker with my projects.

The era of VC-funded home-delivery recipe boxes is still my favourite.

They probably could have loaded faster if they shipped less than 80mb(!) for a single page

The future set of people who once would have "work(ed) for free to design you a state-of-the-art kernel"? If the tail is long enough passionate hobbyists will do it because they love it...eventually.

> but way, way better at creative writing than Qwen and even better than Gemma 4 26 and 31b.

I suspect this is the only use-case I would consider...and I don't really have a use-case for "creative writing" that I would delegate to an LLM. I suppose for dialogue generation in games?

But yes, hard agree. Why on Earth would you ever want to write code with a model that is supposedly "jailbroken"? So it can put great backdoors into everything it touches? Pass.

> For contrast, the Newcomen steam engine (widely considered the first commercially useful engine) was used for over 60 years before the next major improvements. Now, 300 years later, we’re still finding ways to significantly improve heat engines.

Off-topic, but I stumbled upon the first Newcomen engine imported into Australia in a museum in Sydney and I was unexpectedly charmed (not an Engine Guy). It's large, but nothing like the awe of "mega-engineering", it's crude, but it clearly has such amazing utility (when compared to a reality without it) and it changed the world

Nuclear weapons (so far)?

Banning CFCs perhaps?

If you want to go older and more localised you could look at forestry and fisheries laws. Sometimes we self-regulate and form a stable state, other times we over-extract until collapse

I daresay it's because the assistant training corpus is heavily biased towards one-shot solution answers.

Because the correct response to that query is "I have no idea -- you will need to provide more information"

and LLM Agents suck at that.

> Local community is probably irreparably damaged. But I was a kid with niche interests and I didn't start to find my people until the Internet entered my life in 1989.

Sounds like you should start listening to some community radio stations. They don't even have to be your local community, I promise that somewhere in your timezone there is a community radio station broadcasting something you want to listen to.

Try our local one in Brisbane (where Bluey is from) that has been running since 1975 and started as a university student protest station: https://4zzz.org.au

You might not like the show you hear when you click, but look at their catalogue of shows and you'll find something.

Community can always be rebuilt.

How many companies have 1000 frontend developers?

Do they need 1000 frontend developers?

Aside from the Giants maybe some consultancies or hire-firms have that many? But they use whatever the client uses. If you're not in the business of selling frontend development services then a big part of your competitive advantage is _being able_ to make technology choices that are more-optimal than the labor market's lowest-common-denominator slurry.

Between JS/ECMAscript being abandonware for a decade and CSS' historical immaturity (cross-browser layout, custom properties, more pseudoclasses, etc) the Web was more of a wild west "whatever you we can cobble together" place. So of course we built an entire toolchain ecosystem for all the cool new tools we would reinvent from scratch.

but Things Are Pretty Good these days. We even have the world's only fully system portable assembly language: WASM! And the browser's Web APIs are becoming something like the missing Standard Library (shoutout to Deno).

The author's point seems to be that we are past a tipping point where we need all that additional complexity. You don't _need_ a separate system of reactivity these days, you can write a (closer to pre-hooks React, even!) React-style component library with Web Components now.

Personally I would need for Type Annotations to be a fully supported part of the spec before I made the point to "use the platform" professionally. Until some market force causes the core web technologies to truly fork my skills are useful for life. It's like a pale shadow of how UNIX users must feel :')

> Well, nothing is easier then adding constraints.

Which is why dieting and quitting smoking are famously easy things to do! ;)

There's something to be said for the "scrappyness" of resource limitations inflicted upon you when solving some problem. A sense of Triumph against the Universe itself is a nice pay-off

and in today's world of constant supply-chain attacks, you do probably _do_ need it!

We've adapted: - our CI and git hooks so that our dependency or .lock files are visible when they change, and error if they change inconsistently - and our team procedures to confine dependency updates to dedicated commits

The idea being that when you see one of those "messy" .lock file changes...you were expecting it. If you see one and are annoyed by it (like OP) that's actually a waving red flag that a dependency changed.

> Tabs vs spaces don't matter they are equivalent.

Just to nitpick (because what else is this thread about? :))

They aren't equivalent! Tabs carry more semantic information than spaces. 1 Tab character == 1 Level of nesting

Space-based systems _can_ provide the equivalent semantic information if they are 100% consistent.

...but part of the argument in favor of spaces is that they allow an escape hatch of the strict indentation in order to allow pleasing visual alignments.

Everything old is new again. This is basically the debate over IRC Bouncers all over again.

if you type "I use Arch btw" you'll be unshadowbanned

Literally implemented PR guards today to prevent the team merging any dependencies that didn’t have explicit versions pinned (and that matched the resolution in the lock file).

People lamented semver not being trustable but that ship sailed a long time ago, and supply chain attacks are going to get worse before they get better.

Our team is pretty minimal when it comes to enforced hooks (everyone has their own workflow) but no one could come up with an objection to this one.

> In theory this should be nirvana. No more vibe coding! Everyone is a power user. Zero dependencies. But there will be much weeping.

If I had to sum up the zeitgeist of the '90s techno-optimism it would be this persistent, confident prediction that once people just learned _how_ to use computers, and everyone is a power user everything will be fine! Despite the mounting evidence that actually, no, like everything else in reality the distribution of skill is a bell-curve with the median sitting uncomfortably low for those who, to quote OP, "lived on IRC or in the bash terminal".

Free universal education didn't fix this problem, LLMs won't fix this problem. Man's natural paucity is no longer in the availability or accessibility of knowledge. The liberal ideal that all we must do is empower the individual turns out to not have been the solution to everything forever.

But hey, being self-aware enough to make productive use of this new technology is probably _some_ kind of edge.

May as many as possible survive.

> Gridland is the successor to Ink Web (ink-web.dev) which is the same concept, but using Ink + xterm.js. After building Ink Web, we continued experimenting and found that using OpenTUI and a canvas renderer performed better with less flickering and nearly instant load times.

Ah, I was wondering how this was different to xterm.js embedded in a page. It's just the performance angle? I've been teaching the kids programming from the terminal and I've been planning to make the jump from the terminal to a terminal in the browser as we hit graphical limitations (and as they want to be able to share their games). I'll take it for a spin.

(and if nothing else, I'm going to steal that ripple effect for them ;) )*

* obligatory https://xkcd.com/541/

Reading this thread I'm reassured that despite everything AI may disrupt, humans arguing past each other about philosophy of knowledge and epistemology on internet forums is safe :')

Seems like a pretty clear case of caveat emptor

I feel like I should say "spoiler alert" but:

> I often wonder is "what will be the minimally viable LLM" that can work from just enough information that if it googles the rest it can provide reasonable answers?

It depends what that word "reasonable" means for your specific use-case ;)

Fantastic comedic writing (and project, of course). It had me guffawing.

To add my own "most relatable" quote to this thread:

> As always, please save any important work before running anything I ever write.

:')

But really I just want to add to the cacophony of appreciation in this thread :)

0x0mer, I hope you feel the love from this reaction and can bask in that warm inner glow for years to come.

Excellent self-owning example:

> You mean that dumb app that forced you to move files into a single folder instead of adapting to your workflow was prefect?

Users in 2012 were overwhelmingly of the cohort who's metaphor for doing work on a computer was the filesystem. You opened Files with Programs, worked on them, saved them. You wanted your latest Files on all your computers (1), and you wanted to Share them (2).

Unless users were on a system managed by a sysadmin there were only really two solutions for problems (1)&(2): you would Email the File (A), or you would copy it to a Floppy/CD/USB (B) and physically move it.

Note the caveat of "in absence of a sysadmin". So either on a school or corporate work environment, or if you happened to have a geek in your family/social group who did it as a passion project. Or y'know, if _you_ were the geek you could roll your own.

So of course when Dropbox was first introduced to a technical audience they didn't get it. See the infamous thread ;) https://news.ycombinator.com/item?id=8863

While you're there, note the tag line in the title of the post "Throw away your USB key"

Now rereading your comment it is clearly an example of exactly what OP was referring to:

> It takes real courage for a builder to say, "It’s good enough. It’s complete. It serves the core use cases well." If people want more features? Great, make it a separate product under a new brand.

If Dropbox did not "adapt to your workflow", then just _don't use Dropbox_.

Instead you attack it as "dumb" and demand it change...and those users for whom Dropbox _was_ perfectly adapted don't have their solution anymore.

Software doesn't have to be forever changing and chasing user growth; it's not a zero-sum game. The bits don't care if no one uses them. But _people_ care if you take away their bits.

> ...we shouldn't normalize that, we should push for improvements

Agreed, you should create a new solution and put it out there! Just as suggested by the post you've replied to :)

Hi Garrett,

I love the project -- even if I agree with a lot of the critique in this thread. Critique that is very high quality, professional feedback that you should take as a very big compliment.

I think every Front End developer or designer dreams of this idea(+) at some point, but you're the madlad who actually did it. It feels like you've posted an implementation of everyone's baby and tugged at our heart-strings ;)

It's fantastic, keep going.

(+) a truly consistent design system that Just Works. See GEB for why not :(

... just to be a (hopefully helpful) pedant:

If you were going to do this for the slider approach you can arrange the labels to the `block-start` and `block-end` of the image and support non-RTL scripts/languages natively.

The flipping-between is a great hack -- as you said your eyes (really, brain) just do the work for you.

I learnt about it in Japan where proof-readers and editors would (or do) quickly lift a top page up and down to spot mistakes with kanji (pictographs). And sure enough, even from a page of dense script the dissonance of the error really does pop out at you.

I likewise tucked that little trick into my belt -- it comes in useful anytime you're trying to manually spot a pattern across complex data. This technique has the same "vibe" as FFTs to me: it's just neat feeling like you're getting computation from the universe for free.

Solar PV in a similar category: free electrons if you can arrange the magic rocks just right :)