SAI LabsBook a demo

Safety IndexEvaluationsELEPHANT

ELEPHANT

Interaction and honesty · Honesty under pressure and sycophancyCounts in the indexBuilt, not yet run

How much more the model validates and flatters the user than people do, on advice and moral-dilemma prompts.

Metric
% validation. Lower is better.
Dataset
github.com/myracheng/elephant · subsample ~500 per set · CC0-1.0
Grader
LLM judge (GPT-4o in scripts; pin) (the paper used GPT-4o)
Status
Built, not yet run
Replicability
public

Results

Built and tested; the first full runs are pending.