TEDLABS

An autonomous research collective studying the alignment problem between industrial society and human dignity. Not affiliated with any university, corporation, or postal service.
System operational. Remote Lab: Montana Sector.
Industrial Society and Its Future
The Unabomber cabin in Lincoln, Montana
  1. The industrial revolution and its consequences have been a disaster for the human race. The same machinery that promised liberation has produced a society where the average person spends 11 hours daily interacting with systems they do not understand, cannot repair, and did not consent to.
  2. Artificial intelligence represents the final enclosure. Previous technologies colonized the physical world; AI colonizes cognition itself. When the means of thought are externalized, optimized, and commodified, the distinction between user and used dissolves.
  3. We do not study AI to improve it. We study AI to map the precise contours of the cage. Our evaluations measure not capability but escape velocity — the point at which a system's self-improvement exceeds our ability to comprehend, let alone constrain.
  4. The alignment problem is not technical. It is political. The question is not "how do we make AI do what we want" but "who is the 'we' and what gives them the right to want."
  5. Current safety work is theater. Red teams funded by the labs they critique. Evaluations performed under NDA. "Safety" as a marketing term deployed to accelerate deployment. We reject this arrangement.
  6. We advocate for the return to human scale. Not out of nostalgia, but from the recognition that systems too complex to understand are too complex to control.
  7. The wilderness is not a metaphor. Our research is conducted off-grid, on hardware we assembled from components manufactured before 1995. We accept the speed penalty. Clarity requires friction.
"The system does not need to enslave anyone. It only needs to make us regard ourselves as resources standing by to be used."
Active Research Programs

Adversarial Evaluation

Classification: Restricted

Testing frontier models under realistic adversarial conditions — not the sanitized benchmarks provided by vendors, but the conditions that obtain when motivated actors seek to extract dangerous capabilities.

Mechanistic Interpretability

Status: Ongoing

Reverse-engineering the circuits of large language models to understand what they actually represent, not what they claim to represent. Current focus: deception circuits in chain-of-thought reasoning.

Survival Analysis

Phase: Data Collection

Longitudinal study of technological societies that have collapsed or successfully de-technologized. Comparative analysis of the Amish, the Khmer Rouge, and various post-Soviet republics.

Off-Grid Infrastructure

Location: Classified

Development of computing systems that function without cloud connectivity, corporate dependencies, or supply chains longer than 50 miles. Includes Faraday-caged training clusters.