About this idea
LLM Personality Board treats large language models like people taking a personality test: the same HEXACO questionnaire — six dimensions, Honesty-Humility, Emotionality, Extraversion, Agreeableness, Conscientiousness, Openness — is administered directly to the models via API, the answers are aggregated into a per-dimension score with an uncertainty band, and the result lands in a grid of cards with radar charts, comparable model by model and version by version.
The idea started from a simple question: do language models have a recognizable, stable "character" over time, or is it just an artifact of whatever prompt happens to be in play? Administering the same standardized psychometric instrument to different models — and to the same model family over time, version after version — is a way to start answering that with numbers instead of impressions.