On The Challenges Of Using Black Box Apis For Toxicity Evaluation In Research 2023 04 24
Source
cohere.com
Author
Cohere
Date
Context
A model's position in a toxicity ranking can change when the scoring service changes. A research abstract published by Cohere reports that rescoring models in the HELM benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition → with an updated Perspective API produced a different ranking. Because that service is retrained over time, older stored scores and newly obtained scores may reflect different evaluators. The finding makes the evaluator's version part of interpreting a comparison: a changed ranking alone does not show that the underlying models became more or less toxic.
Terms in this piece · Glossary
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
determinism — Whether the same input reliably produces the same output — something LLM systems mostly lack, which changes how you test and debug them.
postCohere releases open-weights translation model North Small Translate
articleThe Data Provenance Initiative A Large Scale Audit Of Dataset Licensing And Attribution In Ai 2023 10 25