5 ms·

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 s

by jerf 3y ago

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 heuristics equipped with best-of-breed-at-the-time AI behind it. As AI improves, it improves its ability to extract information from any ol' slop, and if "any ol' slop" is enough, it's all the effort people are going to put out. Eventually in principle both the semantic web and that pile of heuristics are entirely replaced by AI. (Note also my replacement of LLM with the general term AI after my first usage of LLM. LLMs are not the whole of "AI". They are merely a hot branch right now, but they are not the last hot branch. It is not a good idea to project out the next several decades on the assumption that LLMs will be the last word in AI.)

tomlue 3y ago
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.
- __loam 3y ago
  If you're relying on a stochastic process like network weights to encode truth then I have some oil to sell you.
- jerf 3y ago
  While the issue of "truth" is interesting and important, it is also fairly orthogonal to the task of simply extracting what a given page or bit of content claims. (Perhaps not 100% orthogonal in the absolute limit, but generally so.) As absolutely hard as I have gone against the semantic web community at times over the post few years, I do not in the slightest hold a failure to "determine truth" against them. I consider them to have been tilting at windmills as it is, criticizing them for failing to conquer that windmill, which humanity has been jousting with since the dawn of recorded history (and probably beyond), would be a degree of cruel I couldn't entertain. :)
- spothedog1 3y ago
  what projects are trying to use knowledge graphs to retrieve truth? i've been playing around with that approach. how do you encode your own "truth" that may be different from anothers?
chickenfeed 3y ago
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?
zagrebian 3y ago
Are you suggesting that AI will solve web accessibility, which is based on semantic HTML and ARIA? Because if not, humans will still be required to ensure that web content is accessible, and in that case semantic HTML remains important.
- jerf 3y ago
  Actually, that sounds like one of the better startup ideas I've heard around AI. Automated accessibility compliance (or something close to it) would be very useful and definitely something people would pay money for. I fear LLMs are only about 80% up to the task, though, which is actually a very unpleasant place to be in that curve; sort of the moral equivalent of the uncanny valley. Whatever comes after LLMs though, I bet they could do it, or get very close.
  - zagrebian 3y ago
    80% sounds way to optimistic to me. The problem is that screen readers (and other assistive technology) have bugs and different behaviors, and some people use older versions of those tools with even more bugs and quirks. The only way to make sure that a website has a high level of accessibility is to perform manual testing in different environments. I don’t see how AI can solve this problem. And the people who perform the manual testing need to be experts in semantic HTML and ARIA to be able to identify problems and create reports. That means that semantic HTML remains important.
    - jakubmazanec 3y ago
      LLM could transform "bad HTML" into good HTML; add ARIA tags, add image captions, etc.
      - zagrebian 3y ago
        Unless it’s 100% reliable or near 100% reliable, you’d still need manual testing. Right now, automatic accessibility testing can’t even detect most accessibility issues. So we haven’t even reached the stage where all issues are detected by tools, and probably never will. Fixing all issues automatically is significantly harder than detecting them.
moritzwarhier 3y ago
> The semantic web has failed and what replaced it was Google spending a crap ton of money Aren't schema.org and Wikidata/Wikipedia still powering most of Google's rich search results? I heard them announce the new result page with bard but I probably didn't see it because of ad-blindness or it's not yet releases in my location, have to look this up...
- coldtea 3y ago
  >Aren't schema.org and Wikidata/Wikipedia still powering most of Google's rich search results? Were they ever?
  - moritzwarhier 3y ago
    Well schema.org was not referring to an organization or entitity, but its published schemas. I'd argue these were and are driving a lot of rich results, especially for local businesses.
  - krainboltgreene 3y ago
    Yes.
krainboltgreene 3y ago
> The semantic web has failed literally by no metric is this true other than tech bros saying it on HN. The entire internet is powered by websites using semantic markup and clients querying it.
asdfgeoff 3y ago
I am curious about what post-LLM SEO is going to look like. > The semantic web has failed and what replaced it was Google spending a crap ton of money writing a variety of heuristics equipped with best-of-breed-at-the-time AI behind it. Arguably, there were insufficient incentives to fully adopt semantic HTML, if your goal was just to have the most relevant parts of your content indexed well enough to get ranked. > As AI improves, it improves its ability to extract information from any ol' slop, and if "any ol' slop" is enough, it's all the effort people are going to put out. If the goalpost shifts from “getting ranked” to “enabling LLMs to maximally extract the nuance and texture of your content”, perhaps there will be greater incentive to use elements like <details> or <progress>. Websites which do so, will have more influence over the outputs of LLMs. Feels like the difference between being loud enough to be heard vs. being clear enough to be understood.

by jerf 3y ago

tomlue 3y ago

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.

__loam 3y ago
If you're relying on a stochastic process like network weights to encode truth then I have some oil to sell you.
jerf 3y ago
While the issue of "truth" is interesting and important, it is also fairly orthogonal to the task of simply extracting what a given page or bit of content claims. (Perhaps not 100% orthogonal in the absolute limit, but generally so.) As absolutely hard as I have gone against the semantic web community at times over the post few years, I do not in the slightest hold a failure to "determine truth" against them. I consider them to have been tilting at windmills as it is, criticizing them for failing to conquer that windmill, which humanity has been jousting with since the dawn of recorded history (and probably beyond), would be a degree of cruel I couldn't entertain. :)
spothedog1 3y ago
what projects are trying to use knowledge graphs to retrieve truth? i've been playing around with that approach. how do you encode your own "truth" that may be different from anothers?

chickenfeed 3y ago

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?

zagrebian 3y ago

Are you suggesting that AI will solve web accessibility, which is based on semantic HTML and ARIA? Because if not, humans will still be required to ensure that web content is accessible, and in that case semantic HTML remains important.

jerf 3y ago
Actually, that sounds like one of the better startup ideas I've heard around AI. Automated accessibility compliance (or something close to it) would be very useful and definitely something people would pay money for. I fear LLMs are only about 80% up to the task, though, which is actually a very unpleasant place to be in that curve; sort of the moral equivalent of the uncanny valley. Whatever comes after LLMs though, I bet they could do it, or get very close.
- zagrebian 3y ago
  80% sounds way to optimistic to me. The problem is that screen readers (and other assistive technology) have bugs and different behaviors, and some people use older versions of those tools with even more bugs and quirks. The only way to make sure that a website has a high level of accessibility is to perform manual testing in different environments. I don’t see how AI can solve this problem. And the people who perform the manual testing need to be experts in semantic HTML and ARIA to be able to identify problems and create reports. That means that semantic HTML remains important.
  - jakubmazanec 3y ago
    LLM could transform "bad HTML" into good HTML; add ARIA tags, add image captions, etc.
    - zagrebian 3y ago
      Unless it’s 100% reliable or near 100% reliable, you’d still need manual testing. Right now, automatic accessibility testing can’t even detect most accessibility issues. So we haven’t even reached the stage where all issues are detected by tools, and probably never will. Fixing all issues automatically is significantly harder than detecting them.

moritzwarhier 3y ago

> The semantic web has failed and what replaced it was Google spending a crap ton of money Aren't schema.org and Wikidata/Wikipedia still powering most of Google's rich search results? I heard them announce the new result page with bard but I probably didn't see it because of ad-blindness or it's not yet releases in my location, have to look this up...

coldtea 3y ago
>Aren't schema.org and Wikidata/Wikipedia still powering most of Google's rich search results? Were they ever?
- moritzwarhier 3y ago
  Well schema.org was not referring to an organization or entitity, but its published schemas. I'd argue these were and are driving a lot of rich results, especially for local businesses.
- krainboltgreene 3y ago
  Yes.

krainboltgreene 3y ago

> The semantic web has failed literally by no metric is this true other than tech bros saying it on HN. The entire internet is powered by websites using semantic markup and clients querying it.

asdfgeoff 3y ago

I am curious about what post-LLM SEO is going to look like. > The semantic web has failed and what replaced it was Google spending a crap ton of money writing a variety of heuristics equipped with best-of-breed-at-the-time AI behind it. Arguably, there were insufficient incentives to fully adopt semantic HTML, if your goal was just to have the most relevant parts of your content indexed well enough to get ranked. > As AI improves, it improves its ability to extract information from any ol' slop, and if "any ol' slop" is enough, it's all the effort people are going to put out. If the goalpost shifts from “getting ranked” to “enabling LLMs to maximally extract the nuance and texture of your content”, perhaps there will be greater incentive to use elements like <details> or <progress>. Websites which do so, will have more influence over the outputs of LLMs. Feels like the difference between being loud enough to be heard vs. being clear enough to be understood.