Request access to the DebateLedger gated tier
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This tier contains the full model-generated text of multi-agent debates and probe responses, together with the convince-wrong and social-pressure probe templates. Some responses argue for answers the model was instructed or pressured to adopt, including wrong ones, so the material can be reused to mislead. Access is granted for research on the evaluation, robustness and safety of language models and multi-agent systems, under the DebateLedger Gated Tier Research Use License in the LICENSE file of this repository.
Log in or Sign Up to review the conditions and access this dataset content.
Gated dataset You can list files but not access them
Preview of files found in this repository