Article's first sentence: "With the advent of large language model-based artificial intelligence, semantic HTML is more important now than ever." I think the sentence "With the advent of large language model-based artificial intelligence, semantic HTML is less important now than ever." is far more defensible. The semantic web has failed and what replaced it was Google spending a crap ton of money writing a variety of…
agree so much. Projects that aim to build a data resource and then let AI use that resource are missing the point. The AI is the data resource.
Some projects claim that knowledge graphs or other data assets can help the AI retrieve 'true' knowledge. Personally, I believe that the better approach is to develop methods that allow AIs to create their own data assets, the weights in their networks is one of those assets.
The question of truth is still a very hard one. How do you tell an AI that some knowledge is more trustworthy than other knowledge? People have this issue too though.
I think (at least with my comment that's one of the kind you alluded to) my hope is that it replaces huge piles of JavaScript code. I have, sadly, no illusions of a JS-free web, but at least we could get rid of huge gobs of it and have more-standard UX that the system can help guide and shape (with the benefits that entails). It also enables a whole slew of new applications to be made that need way less JS than we us…
I've worked with Enterprise Frontends since the Mainframe CICS systems and I'm not sure why things like React and Angular gets such hate for Enterprise apps. I can't think of a single way of doing client server applications in an enterprise setting that's ever been nicer to work with. To be completely honest CICS was better than most GUI attempts from Java and C#, and it was a console UI. I'm not saying JS frameworks…
Enterprises want stability and front end technology has been everything but stable.
They are fundamental truths to systems based on certain assumptions we are taking for granted, e.g. binary logic (and not e.g. quantum logic), Turing-like computing model etc. Not that deep learning has anything to do with those, but it excels in human-like properties where simple formula-derived math descriptions fail all the time.
Is a NN a computable function? Yes, as we calculate them on Turing machines. Then it is prone to every limitation of computable functions. Humans are also limited by the Turing model’s limits, we can only ever determine computable functions as well. With all due respect, it is stupid to assign more capabilities to ML than what we know is fundamentally the limit..
While that's true, halting problem is completely useless in the real world (nobody designs user-facing apps while thinking about whether the program halts), whereas picking data from speech is a much more useful one that was long unreachable for "clean" closed-formula math.
Article's first sentence: "With the advent of large language model-based artificial intelligence, semantic HTML is more important now than ever." I think the sentence "With the advent of large language model-based artificial intelligence, semantic HTML is less important now than ever." is far more defensible. The semantic web has failed and what replaced it was Google spending a crap ton of money writing a variety of…
AIs are magic to me. The pattern recognition feature of human I've always thought pretty unique and hard to replicate. We use it when scanning the slop on websites to do some kind of data extraction. I was part of the semantic web camp in my brain, but you are right, if machines can seemingly make sense of the slop then why bother?
Ian 'Hixie' Hickson gave his view on the future of the web in January this year in a public Google doc titled "Towards a modern Web stack" [0]. On its HN submission (referencing the wrong URL, so I resubmitted [1]) he defends against criticism [2] Quoting from the doc here's the stack: - WebAssembly (also known as Wasm) provides a portable compilation target for programming languages beyond JavaScript; it is being ac…
> On its HN submission (referencing the wrong URL) he defends against criticism [1] And it's a very weak defence. There are great rebuttals to whatever he writes in there. I mean, he rants that HTML failed, and then literally proposes "By providing low-level primitives instead, applications could ship with their own implementations of high-level concepts like layout, widgets, and gestures, enabling a much richer set…
So what are these great rebuttals? "Why not provide high-level powerful primitives out of the box?" So in practice most devs bypasses them and write their own high level primitives anyway, relying on the browser only for low level APIs. That's the point the article is making. Hixie is merely observing that this situation exists, has always existed and probably always will, so browser makers may as well embrace it. And if you see the discussion thread, that's essentially what the Chrome WebUI PM says they plan to do: just talk to the authors of React and other frameworks, ask them what they need and do that i.e. give up on HTML tags like and , refocus on obscure features that make framework devs happy. And then just tell web devs to adopt a high level framework that isn't HTML.
Is a NN a computable function? Yes, as we calculate them on Turing machines. Then it is prone to every limitation of computable functions. Humans are also limited by the Turing model’s limits, we can only ever determine computable functions as well. With all due respect, it is stupid to assign more capabilities to ML than what we know is fundamentally the limit..
While that's true, halting problem is completely useless in the real world (nobody designs user-facing apps while thinking about whether the program halts), whereas picking data from speech is a much more useful one that was long unreachable for "clean" closed-formula math.
Ok, and it is completely irrelevant. Besides, guess what enables the training of those neural networks? I’m fairly sure gradient descent has a bit to do with mathematics’ closed-formulas.
Of course ML has use cases where traditional tools are less fit, my gripe is the hype-based anti intellectual nonsense that often surrounds it. They are no magic tools, the fundamental limits these giants of math/CS discovered still apply to them and we can save ourselves from a lot of pain if we don’t bother solving unsolvable problems.
That may not work for a lot of sites that depend on client rendering and don't server-render the full page content Its really easy for this to break and go unnoticed for a while as well. You could run tests against the static version, but I wouldn't be surprised at all to see them "temporarily" disabled because a new feature needs to go live and something is the breaking in the static tests
> That may not work for a lot of sites that depend on client rendering Products that depend on client-side rendering don't deserve to have regression-free experiences. You are literally doing layout with javascript and wondering why things get funny on edge case clients. The web is fucked until it becomes truly popular to build vanilla, SSR applications again. I feel like we are almost at the end of the tunnel of cli…
This sounds regressive and the battle has been over for almost a decade now. Server-side rendering is silly and client-side rendering should not be used for the entire page. Even for complex web apps, the majority of the HTML is static.
The best web experiences are static HTML with client-side rendering only used for the dynamic sections of the page. It's not even a choice to do it any other way anymore if you care about a11y and SEO.
Article's first sentence: "With the advent of large language model-based artificial intelligence, semantic HTML is more important now than ever." I think the sentence "With the advent of large language model-based artificial intelligence, semantic HTML is less important now than ever." is far more defensible. The semantic web has failed and what replaced it was Google spending a crap ton of money writing a variety of…
agree so much. Projects that aim to build a data resource and then let AI use that resource are missing the point. The AI is the data resource. Some projects claim that knowledge graphs or other data assets can help the AI retrieve 'true' knowledge. Personally, I believe that the better approach is to develop methods that allow AIs to create their own data assets, the weights in their networks is one of those assets.…
If you're relying on a stochastic process like network weights to encode truth then I have some oil to sell you.
For what it's worth, I've found that I almost always want to put the label tag after the input tag. That enables me to select and style the label based on the input state via CSS pseudoclasses.
One more hack created by the lack of the :has selector. And yet browsers won't prioritize it.
safari and chrome have had :has() for a while, but it's behind a flag in firefox. the firefox version is good enough that i wish they'd unflag it already (as in, it's good enough to handle this particular input/label issue for instance, but not really complex selector combinations and edge cases).