You're right for personal notes. A graph you have to manually maintain is usually just elaborate procrastination if that's all you are doing.
It becomes useful when the relationships themselves are what you’re querying, e.g what depends on what, what breaks if X changes, or how two seemingly unrelated things connect. Ideally it’s built and updated automatically (e.g via agents and crons).
If you just need to find something you wrote, search is enough. If you want to build something agents can use that gets more useful across time to understand connections between things you want a kg.
@samthekorean Sam I dm’d you a few weeks back, check it out. Context graph for capital markets and AI agents. Supply chain intelligence tracking materials to companies end to end.
Infers that frontier models from here on out will have to be gated for national security. Until open models catch up. Then it’s a new dawn. If the frontier = gated = demand driver for personalized context + llm reasoning.
Frontier models are natsec, so own the map running underneath it. If these kinda models are gated by gov, cloud, KYC, export rules, whatever, then nobody serious wants their edge trapped inside one. They’ll want their own context layer, a map they can run across models. Control the input as context, make it relevant, narrow but deep. Then the model layered on top becomes less important and acts more as search. We’re already seeing world models somewhat, this is becoming much more important now. Hedge the fact control of the frontier is now national security = context you own + llm layer that’s not model dependent.
@RaoulGMI Bend the knee and accept the fact your critical infrastructure. The good news: it’s a policy put. Like defense the cost incurred is irrelevant, control and advancement over adversaries is all that matters. So supply chains are going to continue seeing J curves that don’t stop.
The more frontier models become sovereign assets, the more valuable context becomes.
A smaller model with the right context can be more useful than a bigger model guessing from scratch.
The moat isn’t just the model. It’s the system around it:
what context it gets, what it trusts, what it remembers and when it knows to defer.
Frontier models reason exceptionally. But context tells em where to look. Feels like this just further cements the paradigm.
#ai #knowledgegraph #LLM
It’s the nationality gate, export doctrine applied to cognition. Wild that lab org charts now have to partition around clearance and citizenship. Will restructure how frontier research gets done. The talent funnel that built these labs now becomes potential assets for adversaries.
@dessaigne Agreed. Create a digital ledger of all actions, audits running nightly, agents awakening and reading learning loops before executing. If everything is queryable every error is training data. The J curve is something else.
@marcvanderchijs Correct. Agents burn this many tokens because they lack the context needed to do reasoning from the off. The future is graphs layered by llm reasoning. Token reductions paired with improved accuracy and speed.
Indeed. You can't treat severity as permanent. The actual work is modeling the decay. Qualification timelines, substitution paths, what generation N+1 looks like. A chokepoint with a 3yr moat and a clear bypass route is a completely different trade than one where the physics don't change. Problem is most people are too lazy or ego driven to see the distinction.
The US-China AI race is a race no one can win and no one can afford to lose.
Every great power competition in history was for territory,
resources, or weapons.
This one is the first that is for none of them. It is a race for the substrate of intelligence itself. The stakes are not geopolitical anymore. They are ontological.
And here is the trap. Intelligence compounds exponentially. Slow down for a year and the other side does not slow down with you. They compound past you. A six-month lead becomes a decade gap. A decade gap becomes uncrossable. Whoever gets to ASI first owns it forever.
No nation can accept that.
So neither side can ease off. Not for a quarter, not for a month, not for a week. They have to run, harder and faster, forever.
And the game itself cannot end. Intelligence has no finish line. AGI is not the finish, it is the inflection. ASI is not the finish either. There is no finish. There never was.
The greatest game ever played, with no whistle, no winner, and no end. The universe is using both sides to make itself more intelligent.
It’s all part of the Universal Code.
The Universe's first Law is that is solves for output of intelligence per until of energy. You might not be able to see it yet but that is the Great Game...
@WhiteHouse What epic deals is that sir? You just got your ass handed to you. No chips. No rare earths. These two alone would have altered the power dynamic.
The materials for the new dawn originate from Asia. If you map the chain the dependencies are pretty much Asia origin 90%. Regardless of how much Trump wants to reshore, Asia is going to benefit. A lot.