- From: Manu Sporny <msporny@digitalbazaar.com>
- Date: Sat, 21 Jun 2025 11:56:57 -0400
- To: Otto Mora <omora@privado.id>
- Cc: public-did-wg@w3.org
On Sat, Jun 21, 2025 at 10:55 AM Otto Mora <omora@privado.id> wrote: > This video, though, makes an interesting argument that perhaps now with AI agents, having semantic web data will become relevant again because it would allow AI agents to more quickly reason through information. Yes, Semantic Web concepts are "new" again... and "semantic web reasoning engines" of 20 years ago can help LLMs today. Yes, there are important details that differ from the way things were done 20 years ago to the way things are done today, but it's still all about subjects/tokens and their relationships with other subjects/tokens. LLMs lean far more heavily towards statistical relationship engines (again, not a new concept) than the algebraic relationship engines that powered the early semantic web. They're complementary technologies LLMs are not rigorous in the way that semantic reasoning is, but semantic reasoning tends to fall flat due to lack of large, curated datasets. Since the dawn of reasoning, there has been a need to classify and relate information, first in written manuscripts, and more recently more formal ways (such as the semantic web) so that you could reason without hallucinating. LLMs don't do well when you feed them garbage, and the more LLM generated content we have out there, the more garbage they ingest and the less effective they become. Ideally, the information you feed them is high quality and verifiably correct... which is where DIDs and VCs come in. All that said, I don't think we need to go "full blown Semantic Web" for linked data concepts to have use to DIDs or VCs. That DIDs and VCs can help LLMs hallucinate less is a good thing, but even if the LLM hallucination problem didn't exist, there is still value in using linked data in VCs (semantic precision, decentralized extension and innovation, cryptographic guarantees, etc.). The danger here is leaning to heavily into the Semantic Web -- yes, it's neat, but let's make sure we only use the parts that are truly useful and not dive too deep into that rabbit hole. -- manu -- Manu Sporny - https://www.linkedin.com/in/manusporny/ Founder/CEO - Digital Bazaar, Inc. https://www.digitalbazaar.com/
Received on Saturday, 21 June 2025 15:57:38 UTC