EigenBench
A Comparative Behavioral Measure of Value Alignment
Models answer the same scenarios and judge each other’s answers against a written set of values.
Research
Papers from LAISR Lab on measuring the values of language models and on character training.
Side Effects of Character Training
Quantifying Cross-Constitution Drift in LLMs
Training a model on one trait shifts others too. We measure those side effects and what it costs to reduce them.