Start Here

Start Here

Choose your path based on what you're trying to understand or accomplish.

For AI Safety & Interpretability Researchers

Who this is for: You work on mechanistic interpretability, AI alignment, or understanding what models are actually doing internally.

Why this matters to you: Geometric methods provide measurable signatures of self-models, deception, and hidden reasoning that linear probes miss.

Start with these (in order):

  1. Inside LLMs - A longer form narrative exploration of residual stream geometry
  2. Curved Inference - A series of experiments measuring residual stream geometry
  3. PRISM (Persistent Recursive Introspective Self-Model) - An experiment in Computational Phenomenology exploring hidden theatre and register separation
  4. Research Program overview - How it all connects

Then: Explore PRISM implementation on GitHub and subscribe for experimental updates.

For Consciousness Scientists & Computational Phenomenologists

Who this is for: You study consciousness, phenomenology, or the relationship between subjective experience and physical systems.

Why this matters to you: FRESH provides a geometric framework that makes phenomenological concepts falsifiable using LLMs as instruments (not as conscious subjects).

Start with these (in order):

  1. FRESH (Functionalist & Representationalist Emergent-Self Hypothesis) - A geometric approach to consciousness
  2. PRISM (Persistent Recursive Introspective Self-Model) - Operationalising phenomenological facets
  3. Curved Inference - Measuring how the geometry inside LLMs evolves in response to concern, deception and compuational self-models
  4. Research Program overview - Theory → measurement → experiments

Then: Consider how these methods apply to your research questions or experimental paradigms.

For Philosophers of Mind & Cognitive Science

Who this is for: You’re interested in theories of self, subjectivity, agency, or the nature of cognitive architecture.

Why this matters to you: This work connects Metzinger’s self-model theory, Perera’s virtualised theatre, and phenomenological traditions to falsifiable predictions.

Start with these (in order):

  1. Parrot or Thinker article (latent deictic models) - Functional account without metaphysics
  2. Latent Deictic Models - How deixis in the training data can lead to sophisticated latent models
  3. FRESH (Functionalist & Representationalist Emergent-Self Hypothesis) - Geometric framework for subjectivity
  4. PRISM (Persistent Recursive Introspective Self-Model) - Testing predictions experimentally
  5. Research Program overview - The complete picture

Then: Engage with the philosophical assumptions or propose alternative operationalisations.

For AI Developers & Practitioners

Who this is for: You build AI systems and want more transparency, control, or understanding of what they’re doing.

Why this matters to you: These tools help you measure internal model structure, detect hidden reasoning, and understand failure modes.

Start with these (in order):

  1. Curved Inference - Measuring concern and intent
  2. Tools overview - What’s available and how to use it
  3. Parrot or Thinker article - Why latent models matter
  4. Subscribe to updates - Get notified of new tools

Then: Try the methods on your own models or contribute to the open-source implementations.

Not sure which path fits you?

Read the Research Program overview for a complete picture of how theory, measurement, and experiments connect.

And subscribe to updates and explore as the work develops.