So many new features in TrustGraph 2.10, we can't list them all! Semantic Policy Governance with SHACL-AF? ✅ Guided graph traversal? ✅ Decision scoring with open weight models? ✅ Enhances to graph embeddings? ✅ VOWL ontology visualizer? ✅ And so much more! No really. There is a lot more, but this post has to end some time...

I never expected to see the @TrustGraphAI logo next to Minecraft. 😂 Thanks for the support @Olares_OS!

I can still remember trying to use that clunky Docusarus landing page as a "website". 😂 IIRC, this is v4 when it comes to @TrustGraphAI sites. I do miss the "space container ship". Maybe one day it'll reappear...

Everyone is talking about ontologies. Why now? What changed? The economics. Ontologies are far from new. If we're talking about semantic web standards like RDF, they date back to the late 90s. OWL, the standard for ontologies, goes back to 2004. What's really changed is how you develop ontologies. The old way: - Read lots of information and try to understand what it all means, mapping out relationships and terms - Try to reach consensus with others on the mappings, or in other words, argue a lot about it - Build all that into a machine-readable structure like OWL As you can imagine, this is a long process, for a complex domain likely months. So what's the new way? - LLMs can read an interpret the domain - LLMs can build the relationships and mappings - LLMs can structure all that in OWL, or any target format you'd like And how long does that take? Minutes. I know what you're probably thinking, what about the quality? Surely human developed ontologies are better, right? Not really. Since we released ontology features and OntologyRAG in @TrustGraphAI, we've studied a lot of ontologies and developed a lot of our own. LLM developed ontologies? Easily on par with human developed ones, and more often than not, better. But where do we go from here? Expect to see even more advanced capabilities in TrustGraph very soon that will be able to develop custom ontologies by learning your business logic through your own data with all the lifecycle change management capabilities you need. Don't trap your business logic in proprietary ontologies. Develop you own with TrustGraph.

If we really want to talk about graph engineering, shouldn't we be talking about semantic significance?

What Is Graph Engineering? (Hint: It's Not What You've Been Told)

I looked through the categories and criteria for the AI Infra track, and I'm pretty sure, between you and me, if you use @TrustGraphAI you'll win. 🤫 💰 😅

The path to #EnterpriseAI follows a road paved by safety engineering, human factors, cyber risk quantification, behavioral economics, and prospect theory. Together these methods can generate measurable determinism for agentic AI. All coming very soon in @TrustGraphAI...

We like to think so too. 😅

Today, I read a LLM license so you don't have to! Today's model? Kimi K3. What's the TLDR? Yes, Kimi K3 is open weights, but it's not true open source as it comes with commercial restrictions. It's not MIT, Apache 2, AGPL-3, or even SSPL. It's their own custom written license. It sorta reminds me of some of Meta's licenses for Llama with one big exception... I actually don't understand this license - kinda at all - and I wonder if that's the intent. I also wonder if they had US counsel to help them with the wording, because lost in translation could also be an issue. The key points: - If you provide Model as a Service to a third party with Kimi K3 *and* exceed $20M USD (or equivalent) in revenue over 12 consequetive months, you must obtain a separate - and I'm assumed paid - license from Moonshot AI. - However, there are exceptions to Model as a Service, specifically "...(a) end-user products with model capabilities solely embedded within specific features or harnesses, or (b) mere relaying of requests o models hosted by others." I very much don't understand exception (a), and as Mark Adams and I wondered, if we deploy Kimi K3 within @TrustGraphAI (the deterministic context harness for agents), are we exempt? I think we can argue yes, but I have a feeling that wasn't their intent? - Much like Meta's old licenses for Llama, if you exceed 100 million MAU (montly active users) or $20M USD revenue threshold, "Kimi K3" must be prominently in the UI. - Then there's point 4. Point 4 states that neither the Model as a Service nor MAU thresholds apply if "...(a) internal use of the Software, defined as any use that does not make the Software, its outputs, or its underlying capabilities available to third parties..." This again, is very ambiguous and open to interpretation in my view. If an enterprise were to take a platform like TrustGraph, and use it to deploy Kimi K3, it seems they would be exempt from the Model as a Service and MAU requirements, even if someone (like us) provided engineering services to them to manage/host the TrustGraph deployment that forms the harness for the internal usage. Again, ambiguous. It seems this license is aimed at someone that deploys the model, then provides simple API access to it. However, if you deploy it within a more sophisticated context harness, it's exempt from the revenue threshold? For me, the interpretation is unclear, and that could be the intent. Perhaps they have intentionally released Kimi K3 with an ambiguous license to scare people away from trying to commercialize it. Is Kimi K3 an open weights model? Yes. Is Kimi K3 true open source software? No. There are commercial restrictions that limit usage. Is this the future direction of open weights models? Restrictive licenses that allow us to see the weights but prevent commercialization? We shall see.

🥞😍