Long-form interaction
What becomes visible only at scale?
Accumulation changes what can be measured. Repeated motifs, shifting coordination, and long-range structural variation only emerge when exchanges persist long enough to develop them.
About Axiom Drift AI.
At the intersection of human-computer interaction, evaluation research, and the empirical study of long-context language model behavior.
Axiom Drift AI studies the structural dynamics of long-form human–LLM interaction
The question is narrow: what measurable structure emerges when humans and language models interact across hundreds of turns, and what can that structure tell us about the interaction?
Measurements come from temporal organization and coupling behavior — not content, sentiment, or meaning. The approach is non-semantic by design. Findings are published openly. Instruments are publicly available.
Four recurring lines of inquiry.
Coupling regimes and temporal structure
How does the interaction between user and model evolve over time?
Interactional tempo, shifts in responsiveness, and local stabilization are not fixed properties — they can vary across the length of the interaction. We examine that variation and its broader temporal architecture.
Synthetic nulls
What remains after baseline noise is accounted for?
Synthetic baselines help distinguish structured interactional signatures from patterns expected by chance, format, or model habit.
Cross-human comparison
How do different humans shape different interaction trajectories?
Comparative analysis helps identify stable differences in pacing, elicitation, structure, and model response patterns.