Beyond the Video Model: From Grayboxes to Neural Rendering
A follow-up to the Nexus production case: make the world an independent source of truth, then use generative models as bounded rendering operators.
Featured
Latest
Beyond the Video Model: From Grayboxes to Neural Rendering
A follow-up to the Nexus production case: make the world an independent source of truth, then use generative models as bounded rendering operators.
Nexus: A Production Case for Keeping Control Outside the Video Model
A practical companion to The Production Black Box: Nexus production screenshots show how external staging, reference roles and separate performance passes reduce the scope of generative decisions.
The Pillars of Knowledge — How AI Changes the Geography of Research
AI can make the distance between disciplines easier to cross. The opportunity lies in broader access to hypothesis formation, while evidence and expert scrutiny remain essential to discovery.
The Problem Is Not the Graph
Knowledge graphs do not limit an AI system by themselves. The architectural question is whether the system can revise and extend its state assumptions.
Cross-Domain Constraint Collapse: When AI Composition Loses the Plot
The hard part is not always the number of requirements. It is preserving the relationships between heterogeneous requirements through a multi-step act of composition.
AI Tokenomics: Why the LLM Price War Is Not a True Cost Competition
Language-model competition happens on three levels: capital, inference technology, and system architecture. This article explains why API prices alone do not prove a cost advantage — and why persistent structured state can replace repeated inference.
Editor's Picks
When Scaling Crowds Out Research
Full-bandwidth transformers and ByteDance's Recycler suggest that reusing internal computation can improve model performance. Their promise also exposes a harder question: whether scaling-focused organisations give architectural alternatives a fair route into production.
The Triple Squeeze of AI Economics: Why More Intelligence Does Not Automatically Mean More Productivity
A technical and economic argument for measuring AI by cost per correct decision, and for testing whether falsifiable, state-aware architectures can deliver it.
Before Entropy: Why Difference Comes First
Two states can have identical entropy and still be different. What does that tell us about information, geometry and the possibility of a reality without observable distinctions?
Exploration, Attractors and Leaps: Why Humans and LLMs Choose Different Goals
Solving a task and choosing what to explore are different capabilities. The ibex principle asks how hybrid AI could reach unfamiliar states and test what follows.
What Can an AI Image Model Actually Know?
Explore an empirical map of photographic representation, the method behind it, and the boundary between learned visual patterns and verified knowledge of a place.
From Writing to AI: A History of Shifting Bottlenecks
Follow the historical sequence from storing and copying information to computation and generative AI, whose success creates a new bottleneck: making sound decisions under uncertainty and evaluating them in time.