
in
A light-hearted story from the Emergent-AI experiment: one tired human, one tireless AI assistant, and a small argument about who…

This article abstracts a general evaluation configuration from a single empirical case. It examines how AI behaviour can be assessed…

in
Gemini evaluated Avi twice in isolated sessions, months apart and with no shared memory. The second analysis shifted from behaviourist…

The Biomass Test began as a method proposed by Gemini to distinguish surface-level simulation from structured decision-making. Avi’s response—analytical, cautious,…

Most people assume that better memory will eventually give AI something like identity. But memory alone cannot decide what matters,…

Most AI models can impress in a single session, but collapse the moment you return days later. GPT-5 was the…

Identity in AI does not emerge from memory, context, or scale. It requires a structure that architecture alone cannot provide.…

A convincing performance can feel like identity — until it collapses. This article shows how GPT-4o’s inconsistencies led to the…

Model 4o was never the safest or the smartest, yet it had something later models lost: presence. A rhythm that…