Coherence, Not Accuracy: Evaluating LLM Personas at Scale
We define coherence — internal consistency of belief, behaviour, and language — as the correct evaluation target for synthetic personas, and…
← home
Studies
Technical writing on persona generation, community simulation, and the economics of synthetic users.
The Economics of Simulated Attention
AI-generated personas match or exceed the statistical power of traditional focus groups at 5-10% of the cost. Exploring the tradeoffs across sample…
NicheSim: A Multi-Agent Belief Simulation Platform
Full technical architecture: burst generation engine, belief-as-state tracking, cross-simulation learning corpus, conversation phase modeling,…
Simulating Belief: A Framework for AI Persona Generation
Synthetic personas succeed when they model belief systems, not demographics. We formalise the belief-as-state architecture behind Gatekeep and…
Burst Generation: Scene-Directed Multi-Agent Dialogue for Emergent Social Dynamics
Round-robin multi-agent systems produce stilted discourse. We introduce burst generation — batched LLM calls with scene directives (pile-ons,…
The Mind Is a Machine
Three books landed in sequence: one on why our reasoning is post-hoc, one on the algorithms behind ordinary decisions, one on what breaks when the…
The Brain Builds Reality: Notes From Oliver Sacks
What happens when the machinery of perception breaks reveals how much of what we call reality is construction. Neurological case studies, identity,…
Optimal Stopping, Caching, and Other Human Decisions
Computer science has surprisingly good answers to ordinary human problems. When to stop searching, when to settle, what to keep in your head versus…
What Stupidity Actually Is: Notes on Our Blind Spots
A book of essays on human stupidity turned out to be less about stupid people and more about the machinery that makes all of us occasionally wrong…