EpochLex entry
Constitutional Classifier
kon-stuh-TOO-shuh-nul CLASS-uh-fye-er
A safety classifier trained from a natural-language constitution or rule set to detect and block model inputs or outputs that violate specified behavioral constraints.
EpochLex entry
kon-stuh-TOO-shuh-nul CLASS-uh-fye-er
A safety classifier trained from a natural-language constitution or rule set to detect and block model inputs or outputs that violate specified behavioral constraints.