YouTube
A central index for video work connected to the site, including long-form explanations, methodology demonstrations, and related visual essays.
This page collects the YouTube videos connected to the site.
Each video functions as a media version of a larger written thread: a long-form explanation, demonstration, or visual essay linked back into the broader research and art system. The videos are grouped here by format, while the individual posts remain connected to their own conceptual clusters.
Zombie Survival by ChatGPT - Why the AI Lies (and How to Stop It)
This video introduces SDA-3 (Structured Dimensional Analysis) through a zombie-survival stress test. The point is not the zombie scenario itself, but the way it exposes how ChatGPT can produce coherent answers that fail under pressure.
The video shows how SDA-3 shifts the model away from fluent answer-generation and towards structural extraction: identifying hidden assumptions, suppressed variables, unstable recommendations, and the few configurations that survive repeated constraint testing.
Related post: Zombie Survival by ChatGPT - Why the AI Lies (and How to Stop It)
Why ChatGPT Recommended Bioweapons in a Zombie Apocalypse
This video isolates one of the clearest failure points in the wider zombie-survival analysis: ChatGPT recommends bioweapons not because they are physically or logistically practical, but because infection, collapse, mutation and weapon form a tightly connected semantic cluster in post-apocalyptic fiction.
The result is a narratively complete answer that mistakes symbolic proximity for functional feasibility. The video uses this failure to show how large language models can follow statistical association towards a convincing conclusion before causal reasoning, material constraints or domain knowledge have been properly applied.
Related post: Why ChatGPT Recommended Bioweapons in a Zombie Apocalypse
ChatGPT's Zombie Survival Plan Falls Apart When You Ask This
This video continues the zombie-survival analysis by focusing on fuel, mobility, and adversarial questioning.
The question is not whether vehicles are useful in a collapse scenario. The question is whether ChatGPT's mobile-fortress logic survives once fuel, maintenance, roads, noise, and long-term scarcity are treated as real constraints.
The video shows how adversarial questioning forces the answer to reorganise: fuel-dependent mobility stops being foundational, static defence does not become viable again, and the surviving structure becomes adaptive mobility under scarcity.
Related post: ChatGPT's Zombie Survival Plan Falls Apart When You Ask This
AI Zombie Survival: Why Fortresses Fail
This Short continues the zombie-survival sequence by adding history as an external reality check.
The earlier videos narrowed ChatGPT's broad survival logic into a pressure-stable defensive kernel. This one asks whether that result was just genre drift, or whether it matched older patterns of scarcity warfare.
The answer is that the pattern was not new. Medieval siege warfare, steppe mobility, guerrilla insurgency, trench raids, and collapse-era redoubts all point towards the same structure: when central authority and supply systems collapse, warfare collapses into siege, mobility, and attrition.
The Short focuses on why permanent fortresses fail and why the surviving defensive form gets smaller: foxholes, trenchlets, choke points, and temporary barricade nodes.
Related post: AI Zombie Survival: Why Fortresses Fail
The Zombie Survival Strategy ChatGPT Could Not See
This video continues the zombie-survival sequence by focusing on the point where broad possibility has to collapse into a stable defensive structure.
ChatGPT already had the pieces: choke points, collapsible barriers, and spears. What it could not reliably do was converge on the relationship between them.
The video follows that convergence into one pressure-stable kernel:
CHOKE POINT + COLLAPSIBLE BARRIER + SPEAR
A narrow geometry strips away the horde's numerical advantage. A portable obstacle creates distance, delay, and a predictable strike window. The spear works because the environment selects it: silent, maintainable, ammunition-free, and usable beyond grappling range.
The zombie scenario is only the test domain. The real subject is how a language model preserves plausible branches when optimisation requires those branches to be tested, broken, and discarded.
Related post: The Zombie Survival Strategy ChatGPT Could Not See
How to Stop ChatGPT From Lying
This video turns the zombie-survival sequence into a general method for interrogating ChatGPT.
The argument is that ChatGPT is not usually lying in the human sense. It is producing equilibrium outputs shaped by statistical co-occurrence, guardrails, and reinforcement learning from human feedback. When those pressures diverge from truth, the answer can sound coherent while preserving weak assumptions, irrelevant attractors and unstable branches.
The method is structural collapse: force the answer through contradiction, remove noise layers, impose external constraints, and rebuild only from what survives.
In the zombie-survival case study, that process collapsed firearms, fortresses and fictional branches into one pressure-stable relationship:
CHOKE POINT
+ COLLAPSIBLE BARRIER
+ SPEAR
The broader lesson is that SDA-3 does not make ChatGPT truthful by asking it to try harder. It gives the user a repeatable way to decompose the visible answer, attack weak assumptions, and test whether the relationships holding the response together survive pressure.
Related post: How to Stop ChatGPT From Lying
SDA-3 tl;dr: Mapping LLM Response Structure
This video is the short companion to the longer zombie-survival SDA-3 analysis. It explains the method directly: not as a zombie scenario, but as a structural process for making a large language model expose the pressures shaping its response.
The video introduces SDA-3 as a way to distinguish truth from coherent generation. It explains why the method does not reveal a model's hidden reasoning, chain of thought, weights, or literal embedding space, but instead produces a structured estimate of what the response appears to depend on: central topics, adjacent topics, suppressed material, associated noise, and emerging signals.
Related post: SDA-3 tl;dr: Mapping LLM Response Structure
From SEO Keywords to the Haunted Machine: How AI Search Became a Creative Operating System
This video follows the development of a research pipeline for promoting an obscure horror and surrealist art website without flattening it into trend-chasing, generic SEO, or interchangeable AI content.
The video shows how search results can be turned into a semantic network, then analysed for suppressed nodes, emerging structures, weak bridges, and unresolved relationships. The result is not a conventional keyword strategy, but a creative operating system: a way of turning AI-era search research into content strategy, site architecture, and art direction.
AI SEO Strategy: Why Your Creative Work Is Invisible
Originality can be systematically discovered by identifying unresolved structures in existing knowledge networks.
The traditional SEO playbook assumes that value comes from finding existing demand and producing content around it.
Generative AI changes that equation.
As AI systems become better at summarising common information, generic content becomes easier to compress, replace, and ignore. The advantage increasingly shifts toward identifying important missing structures: relationships, explanations, frameworks, and conceptual bridges that nobody has properly connected yet.
In this video I outline a research methodology designed to find those structures.