Synthetic Life

AI answers depend on who's asking · a research prototype

MODE A — SYNTHETIC INPUTS, NEVER SYNTHETIC OUTPUTS · MODE B — SYNTHETIC RESPONDENTS, UNCALIBRATED

1 · Population

Population Purchase category Brand markers (comma separated, optional) Purchase query (Mode A — asked to the real assistant) Panel question (Mode B — asked to the synthetic respondents) Models (Mode A runs the query on each selected model · Mode B uses the first)

Prototype numbers are directional (1 repetition per run). API ≈ consumer app, not identical — directionally real. Mode B outputs are LLM-simulated opinions, uncalibrated in this prototype; the production rule is calibration against real human panels. Brand presence is a text-match heuristic, not the validated coder.

Synthetic Life

1 · Population

2 · Group profiles (researched with live web grounding)

3 · Results (cold baseline vs in-context, one row per model)

4 · Population report