Aleksandr Altergot
Researcher, LabAI — an independent AI lab (2025-present)
Working on whether a population of LLM agents keeps a stable identity over long horizons, and on how much of a measured change comes from the agents rather than from the evaluator that scores them.
Research
We run a population of persistent LLM agents with episodic memory and a deterministic psychometric layer, and study whether their identity holds over months — under preregistration, with randomized arms and a public deviation log. The open question: what actually keeps an agent the same character over time, and whether that can be measured reliably at all.
Work
- Longitudinal identity of LLM agent populations — preprint in preparation, 2026.
- Instrument diagnostics for longitudinal LLM evaluation — code and dataset to be released.
Background
Omsk State University, Computer Science Faculty — PhD Student, 2012–2015.