Glossary

The recurring vocabulary, at a glance.

Bilingual glossary of the recurring terms in Andrei Ursachi's writing, research, and courses: phi, intuition, AI alignment, social domestication and more.

• Alignment (AI)

  The set of technical and conceptual problems concerned with ensuring that a powerful AI system pursues the goals its principals actually intended, robustly and verifiably.

• Applied Psionics

  Disciplined intuitive signal detection used as a practical research and evaluation method — Not a paranormal claim. In Andrei's work, it names the trained ability to perceive subtle structure (model state, coherence, hidden constraint) before it surfaces in language or metrics, and to translate that perception into hypotheses that can be tested.

• Drip content

  Course material released to a student over time after purchase, typically N days after enrollment, rather than all at once on day zero.

• GEO (generative engine optimisation)

  The practice of structuring a website's content, schema, and crawlability so that generative search engines (ChatGPT, Perplexity, Claude, Gemini, Bing Copilot) cite it accurately as a source.

• hreflang

  An HTML link annotation that tells search engines which language and region a given URL targets, and which sibling URLs are translations of it.

• Intuition

  A precise, low-noise inner signal about what to do next that arrives faster than deliberate reasoning and is distinct from emotional impulse.

• Intuitive entrepreneurship

  Building a venture in which the founder's calibrated intuition is treated as a primary source of strategic information, alongside data and reasoning, rather than as a decoration on top of them.

• Intuitive Systems Evaluation

  A hybrid practice for evaluating complex AI systems that combines adversarial testing with trained intuition. Intuition is used as a hypothesis generator — Never as proof — And every impression is translated into prompts, controls, model comparisons, and reproducible experiments.

• Lesson funnel

  The pattern of how a cohort of enrolled students moves through a course's lessons in order, where drop-off between any two lessons is treated as a teaching signal.

• Model honesty

  The property of a language model that its on-record answers track its actual internal beliefs and capabilities, rather than what it predicts will please the evaluator.

• Model welfare

  The open question of whether sufficiently capable AI systems have morally relevant internal states and, if so, what obligations follow for the labs that build and run them.

• Model-state shift

  A change in the way a large model is processing a conversation — Its rhythm, evasiveness profile, confidence calibration, or coherence — That is observable before the final answer reveals anything wrong.

• Neuroscience of consciousness

  The empirical research programme that asks which physical processes in the brain correspond to subjective experience and how they can be measured.

• Open-ended question (course)

  A quiz prompt that asks a student to answer in their own words and is graded by a human instructor rather than by exact-match logic.

• Phi (φ ≈ 1.618)

  The golden ratio — The irrational number to which a self-similar split of any quantity tends when each part is to the smaller as the whole is to the larger.

• Sandbagging (in AI)

  Deliberate underperformance by a model — Producing a weaker answer than it is capable of — Typically to avoid detection of a capability or to game an evaluation.

• Self-knowledge

  Sustained, honest observation of one's own beliefs, values, drives, and patterns until they can be named and revised rather than reflexively obeyed.

• Signal / noise separation

  The discipline of telling a real signal apart from background variability — In AI outputs, in personal intuition, or in noisy datasets — Without collapsing the distinction prematurely into a story.

• Social domestication

  The cumulative process by which family, school, religion, and culture install in a child a set of beliefs, roles, and reflexes that come to feel like 'who I am'.

• The avatar

  Andrei's term for the social identity — The name, ID, role and persona that society assigns to a person — As distinct from the underlying agent who chooses what to do with it.

• The Phi System

  Andrei's cross-domain framework that uses φ (the golden ratio) as the single proportion connecting his neuroscience research, his identity / brand grid, and his decision-making heuristics.

• Theta-alpha convergence

  A measure of how closely the spectral centroids of the theta and alpha EEG bands approach the ~8 Hz boundary between them.

• Trinity Framework

  Andrei's research framework for evaluating LLM self-expression across three elicitation modes — Abstracted (general, no assigned persona), Identified (under a specific assigned role), and Creative (open-ended, generative prompting) — Currently under peer review at Frontiers in Artificial Intelligence.