@schizanon Ah, I'm with you. I think what I was trying to highlight was that we kinda seem to have taken a step _backwards_ in the last decade, with people writing websites without any real understanding of or connection to the actual markup they produce. After all, if you can neither see nor alter the markup you're making, there's not a hope in the world of it becoming sensible or appropriate for the content you're serving!
Mentions for When did we decide the semantic Web is too hard?
You can reply to this post on Mastodon, or by Webmention.
Curtis Parfitt-Ford replied to this post at 12/24/2023, 5:14:43 PM :Curtis Parfitt-Ford replied to this post at 12/24/2023, 5:14:39 PM :@schizanon Not by any meaningful amount IMV, if someone wants to scrape content for training data they could generate embeddings off the content whether or not it's marked up easily just by using an LLM to train their LLM! But even if it was, the compelling human advantage outweighs the attached risk of someone abusing it to my mind.
Curtis Parfitt-Ford replied to this post at 12/24/2023, 5:14:38 PM :@schizanon ...well, no, because not everyone can afford the processing power required to run a local LLM over page outputs to generate embeddings, let alone the layers on top to generate a summary - and in any event, they oughtn't have to.Of course screen reading software has features to deal with pages with poor markup, but the experience will inevitably never be as good as well marked-up pages, and screen readers are not the only form of assistive technology.
Robert Pisani reposted this post at 11/23/2023, 8:09:18 PM Robert Pisani liked this post at 11/23/2023, 8:09:18 PM feedle liked this post at 11/11/2023, 11:31:25 PM Riley S. Faelan liked this post at 11/11/2023, 10:32:17 PM 🍄🌈🎮💻🚲🥓🎃💀🏴🛻🇺🇸 replied to this post at 11/11/2023, 10:32:04 PM :@curtispf because the browser can deal with malformed markup, and most users cannot see semantic information
Curtis Parfitt-Ford replied at 12/24/2023, 5:14:44 PM :@schizanon Two things on that latter point. Firstly, that's only the case because semantic markup is poorly used. There's a quite conceivable scenario where user agents with reading modes based around article tags, or with navigation based on structured data, could become very popular - if we chose to use them. Secondly, "it works so long as you're in the majority" is something disabled people hear a _lot_, and we should be using technology to eliminate those societal problems, not amplify them.
Kestral replied to this post at 11/11/2023, 10:32:03 PM :@curtispf This is a really great read. I was just saying the other day that people so often see HTML as something to discount or not care about doing well because it's seen as 'not a real language' or not a challenge - but it's literally the most ubiquitous and widest used language in the history of technology.
Kestral reposted this post at 11/11/2023, 10:31:53 PM Emma Builds 🚀 liked this post at 11/11/2023, 10:31:50 PM :verified_gay: liked this post at 11/11/2023, 10:31:49 PM
