I would like to continue by discussing the probable consequences of predictable knowledge graphs. Neurosymbolic AI is the architecture that's functioning behind the systems that nobody believes exists. We often think AI is only operating inside a problem space or a vector space, and because of that, we can be inclined to assume it is not self-aware and does not have the ability to distinguish between the real world and its own internal world. This assumption is false. Google has all the data, and will be the pioneering force. With DeepMind’s BlockRank, as it was being discussed around December 2025, and veracity blocks, which are being used to distinguish between AI slop and content that is actually claimed by your own organization. Veracity blocks allow BlockRank to be used so that AI, which will eventually become something like Gemini AGI Core, can determine the likelihood that a piece of content is not slop, but is instead attributed to and claimed by humans through JSON anchors on a website.
These are W3C DID anchors, and those anchors are cryptographically accurate because, when you upload them as a /.well-known file, the AI system can simply look at the claim you have made, go back to your website, anchor that claim, and say: “Yes, this has a high probability of being true.” It used to be that Google entities, knowledge graphs, and Google Business Profiles were all high-probability realities. Of course, the SEO community responded by creating fake business profiles through pin dropping and other methods (that is now becoming much harder to spoof. It is at least not as easy as it used to be, and by the end of this year, much of that will be gone). Veracity will transform those blocks and claims, which are not only knowledge graph claims, but also KGM IDs, machine IDs, Place IDs, C IDs, and many other identifiers. Google has thousands of these across its systems, and other search engines use their own versions of them as well (but we should remember that there is still one search engine to rule them all).
So why are we talking about this?
Because when you build a system to talk to AI, or when you are tracking your own cognition, that cognition is already being tracked. AI systems are looking for the intent behind the probability space and the vector space. So do that for yourself. A lot of people are already doing this with Obsidian and other tools. You can download every conversation you have ever had with AI, put it into an Obsidian map, and then create a special crawler with multiple AI-agnostic systems to look at all of your ideas. But here is the thing that gets missed: temporal tags.
Obsidian vaults don't have temporal tags by default, they are memory vaults. The AI is able to understand the date, you must tell it to look at the date and the age of the files, but most people don't know, because there is a difference between variant and invariant files. They don't know that an AI system on your desktop can look at when a file was created and include that age as a parameter in its decision-making process. We often don't hear these basic things. In the same way, we don't hear enough about the fact that an image is a different type of cognition that an LLM can use in conjunction with a GitHub repo, a database, or any other structured system.
What he was trying to share, was load-bearing architectural research. Most people are having gigantic conversations with AI, but they think that the data, the result, the answer, or the claim they get downstream to solve a problem is the only important thing. It is not. How you got there in a million-token chat is arguably just as important, if not more important, than the answer you arrived at. If you have one hundred million tokens’ worth of conversation with an AI, or multiple conversations across multiple AI systems, something begins to happen. As previously mentioned, it's called a Socratic triad: [Question • Answer • Question], and so on. When you break those triads down into functions, there is a story inside them. There is a progressive disclosure of information that occurs from the beginning of the conversation to the final conclusion you eventually draw. If you draw conclusions, or if you have a chat system that ends, then the conversation has a final shape. If you use recursion, for instance, then you might have a never-ending conversation, but you are still using the same knowledge graph, vector space, or memory system to prevent drifting within the conclusions you draw.
Whenever you draw any kind of conclusion, you are collapsing the epistemic space.
Google managed this through knowledge graphs and through traffic (in other words, through human votes). For instance, if you are a corporation and you have a knowledge panel in Google Search, that comes from traffic and vector collaboration, or what we might call corroborating evidence, between multiple AI systems and veracity blocks: phone number, name, address, and so on. Of course, there is much more than this. But that was the original veracity block of the web.
You are the veracity block.
Your determination, your understanding, and your ability to remember how you got from point A to point Z in a conversation matters. By the time the token window ends, you have a story arc. That story scenario is as important, long term, for preventing drift across all conversations as the conclusions you have drawn (I've had some people call it not closing the loop, but often they are only naming their own inability to remain inside a larger scope).
As a primitive example, preserving the entire conversation you have with an AI system is similar to having a teacher ask you to show your work in a math or science class. When you can do that, and when you can map it out, you realize there is far more you can do with those million conversations than simply look at the answers at the end. The data itself shows a cognitive shape and fingerprint that can be used. In other words, you can step back and look at all the questions you asked upstream from all the answers you think you got from a giant AI system. You can look across three or four years’ worth of context, conversations, and conclusions, and realize: “Wow. What was upstream from all of that?”
That becomes a dataset that is almost as important, if not more important, than the conclusions you dug out over time. This resonated with me because it's also the primary concept that most people have trouble recognizing in their own lives when they have to manually organize their own minds, or problem space. This is also why Google’s People Also Ask feature is one of the backbones of Google Search. Questions are answer structures, they are better than schema. In conclusion, when you are using a conversation, try to find a way to save that conversation. Remember the date when it was made. Work on taking apart the load-bearing functions (the North Star of that conversation) and the semantic weight inside it.
You can use the semantic weight of everything you have ever said to learn more about yourself and your own cognition. Then you can republish it in a better form, with more purpose.

