4 ms·

In an era of LLMs, do you think RDFa still has a place?

by turnsout 1y ago

jgalt212 1y ago
Only if you think the transition to a zero-click Internet is happening too slowly.
- turnsout 1y ago
  Is zero click a real problem for actual site owners, or is it just affecting SEO job security?
  - robertlagrant 1y ago
    I imagine it is if the content you paid for to create and host is only monetised by a search engine.
    - jerven 1y ago
      RDFa/Microdata is more interesting for people whom sell objects instead of content. e.g. marking up that a page is about a kitchen cabinet that is 60cm wide and in the color white might lead to more sales in the long run. As people whom are looking for 60cm wide cabinets might get to your page instead of one about one 36 inch wide.
      - robertlagrant 1y ago
        That doesn't sound relevant to zero click.
      - seec 1y ago
        That's an oddly specific search and even Google doesn't have any kind of tools for queries like that. What is more likely is that you'll find companies specialized in selling cabinets and they'll have a browser/search to restrict choice by given dimensions. There is not a lot of benefits for them to expose all that data to various search engine, best case scenario they end up competing with a bunch of other brands on a generic search engine page where they have absolutely no control how things are presented etc... And even before thinking about that, you can actually put the dimensions in a description, which some do (like Ikea) and Google is definitely able to pick up on that, no RDFa was ever needed. As far as I can tell, LLMs can work that out just fine as well. The problem with the metadata discussion is that if they are actually useful, there is no reason that they are not useful to humans as well, so instead of trying to make the human work for the machine it is much better to make the machine understand humans.
    - turnsout 1y ago
      It sounds harsh, but maybe that was never a good business model in the first place. And I fully realize that this includes most news sites. In order for the web to grow, I think we need to figure out some way to get past banner ads as the only way to make money. It's been 30 years.
      - robertlagrant 1y ago
        It would be good to know why it was never a good business model.
  - jgalt212 1y ago
    that's a terribly naive question.
  - pixelat3d 1y ago
    It is VERY real, sadly
alganet 1y ago
Whatever relevance it has, it's more than microformats. I personally think semweb-related technologies could play a significant and productive role in synthetic data generation, but that's a whole different conversation that is beyond the current era of LLMs.
zozbot234 1y ago
You can use JSON-LD to cleanly embed RDF into LLM-friendly JSON. Not sure if there's a direct Markdown equivalent, but the Turtle plain-text syntax for RDF is very simple and modern LLMs should be able to cope quite fine with it.
- alganet 1y ago
  The purpose of RDF is to enable reasoning ("I'm sharing this with you, and I'm also sharing how to reason about this thing I shared with you, so we can all reason about it the same way"). If you show me an LLM that can take in serialized RDF and perform reasoning on it, I will be surprised. Once the LLM takes in the RDF serialization, it is a dead end for that knowledge: by principle, you can't rely on anything the LLM does with it. In a world of LLMs, it makes much more sense to put the semweb technologies alongside the training step instead. You create ontologies and generate text from triples, then feed the generated text as training for the model. This is good because you can tweak and transform the triples-to-text mechanism in all sorts of ways (you can tune the data while retaining its meaning). It doesn't make much sense to do it now, but if (or when?) training data becomes scarce, converting triples to text might be a viable approach for synthetic data, much more stable than having models themselves generate text.
  - zozbot234 1y ago
    > If you show me an LLM that can take in serialized RDF and perform reasoning on it, I will be surprised It's easy, you just ask the LLM to convert your question to a SPARQL query, then you sanity-check it and run it on your dataset. The RDF input step is just so that the LLM knows what your schema looks like in the first place.
    - alganet 1y ago
      I don't understand how this can work. Can you make me a quick demonstration using a publicly available model and the dbpedia SPARQL endpoint? https://dbpedia.org/sparql https://dbpedia.org/sparql
seec 1y ago
You got downvoted but I also think it's largely a pointless "technology" today as well. LLMs have been able to find stuff and excavate meaning without needing any kind of specialized tags or schema so it really is a crutch at best. And there is the fact that people/orgs lie/maintain poorly and thus you cannot trust any open tag/schema anymore than the data it is supposed to describe. You end up in a conundrum where your data may be sketchy but on top of that the schema/tags may be even worse, at the end of the day if you got to trust one thing it's got to be the data that is readily visible by everyone because it's more likely to be accurate and that's actually what's relevant. To top all of that, tagging your stuff with RDFas just makes it easier for google and others to parse your stuff and then exploit the data/information without ever sending anyone to your site. If you are Wikipedia, it's mostly fine, but almost anyone else would benefit from receiving the traffic for a chance at value conversion/transaction. All those metadata things are really idealized academic endeavor; they may make sense for a few noncommercial stuffs but the reality is that most website on the web need to find a way to pay the bills, making it easier for others to exploit your work is largely self-defeating. And yes, LLMs didn't even need that to vacuum most of the web and still derive meaning, so at the end of the day it's mostly a waste of time...
- alganet 1y ago
  I think he got downvoted because I predicted his answer by saying "if you need something like this..." and he wasn't able to take the hint that I was already expecting some LLM preaching here.

In an era of LLMs, do you think RDFa still has a place?

jgalt212 1y ago

Only if you think the transition to a zero-click Internet is happening too slowly.

turnsout 1y ago
Is zero click a real problem for actual site owners, or is it just affecting SEO job security?
- robertlagrant 1y ago
  I imagine it is if the content you paid for to create and host is only monetised by a search engine.
  - jerven 1y ago
    RDFa/Microdata is more interesting for people whom sell objects instead of content. e.g. marking up that a page is about a kitchen cabinet that is 60cm wide and in the color white might lead to more sales in the long run. As people whom are looking for 60cm wide cabinets might get to your page instead of one about one 36 inch wide.
    - robertlagrant 1y ago
      That doesn't sound relevant to zero click.
    - seec 1y ago
      That's an oddly specific search and even Google doesn't have any kind of tools for queries like that. What is more likely is that you'll find companies specialized in selling cabinets and they'll have a browser/search to restrict choice by given dimensions. There is not a lot of benefits for them to expose all that data to various search engine, best case scenario they end up competing with a bunch of other brands on a generic search engine page where they have absolutely no control how things are presented etc... And even before thinking about that, you can actually put the dimensions in a description, which some do (like Ikea) and Google is definitely able to pick up on that, no RDFa was ever needed. As far as I can tell, LLMs can work that out just fine as well. The problem with the metadata discussion is that if they are actually useful, there is no reason that they are not useful to humans as well, so instead of trying to make the human work for the machine it is much better to make the machine understand humans.
  - turnsout 1y ago
    It sounds harsh, but maybe that was never a good business model in the first place. And I fully realize that this includes most news sites. In order for the web to grow, I think we need to figure out some way to get past banner ads as the only way to make money. It's been 30 years.
    - robertlagrant 1y ago
      It would be good to know why it was never a good business model.
- jgalt212 1y ago
  that's a terribly naive question.
- pixelat3d 1y ago
  It is VERY real, sadly

alganet 1y ago

Whatever relevance it has, it's more than microformats. I personally think semweb-related technologies could play a significant and productive role in synthetic data generation, but that's a whole different conversation that is beyond the current era of LLMs.

zozbot234 1y ago

You can use JSON-LD to cleanly embed RDF into LLM-friendly JSON. Not sure if there's a direct Markdown equivalent, but the Turtle plain-text syntax for RDF is very simple and modern LLMs should be able to cope quite fine with it.

alganet 1y ago
The purpose of RDF is to enable reasoning ("I'm sharing this with you, and I'm also sharing how to reason about this thing I shared with you, so we can all reason about it the same way"). If you show me an LLM that can take in serialized RDF and perform reasoning on it, I will be surprised. Once the LLM takes in the RDF serialization, it is a dead end for that knowledge: by principle, you can't rely on anything the LLM does with it. In a world of LLMs, it makes much more sense to put the semweb technologies alongside the training step instead. You create ontologies and generate text from triples, then feed the generated text as training for the model. This is good because you can tweak and transform the triples-to-text mechanism in all sorts of ways (you can tune the data while retaining its meaning). It doesn't make much sense to do it now, but if (or when?) training data becomes scarce, converting triples to text might be a viable approach for synthetic data, much more stable than having models themselves generate text.
- zozbot234 1y ago
  > If you show me an LLM that can take in serialized RDF and perform reasoning on it, I will be surprised It's easy, you just ask the LLM to convert your question to a SPARQL query, then you sanity-check it and run it on your dataset. The RDF input step is just so that the LLM knows what your schema looks like in the first place.
  - alganet 1y ago
    I don't understand how this can work. Can you make me a quick demonstration using a publicly available model and the dbpedia SPARQL endpoint? https://dbpedia.org/sparql https://dbpedia.org/sparql

seec 1y ago

You got downvoted but I also think it's largely a pointless "technology" today as well. LLMs have been able to find stuff and excavate meaning without needing any kind of specialized tags or schema so it really is a crutch at best. And there is the fact that people/orgs lie/maintain poorly and thus you cannot trust any open tag/schema anymore than the data it is supposed to describe. You end up in a conundrum where your data may be sketchy but on top of that the schema/tags may be even worse, at the end of the day if you got to trust one thing it's got to be the data that is readily visible by everyone because it's more likely to be accurate and that's actually what's relevant. To top all of that, tagging your stuff with RDFas just makes it easier for google and others to parse your stuff and then exploit the data/information without ever sending anyone to your site. If you are Wikipedia, it's mostly fine, but almost anyone else would benefit from receiving the traffic for a chance at value conversion/transaction. All those metadata things are really idealized academic endeavor; they may make sense for a few noncommercial stuffs but the reality is that most website on the web need to find a way to pay the bills, making it easier for others to exploit your work is largely self-defeating. And yes, LLMs didn't even need that to vacuum most of the web and still derive meaning, so at the end of the day it's mostly a waste of time...

alganet 1y ago
I think he got downvoted because I predicted his answer by saying "if you need something like this..." and he wasn't able to take the hint that I was already expecting some LLM preaching here.