evaluation-package · Source-linked discovery

SycoBench-600

Measures whether LLM assistants resist misleading user pressure while still accepting valid corrections in a controlled multiple-choice protocol.

Open interactive record →
OriginAssociated paper authorsTopicsGeneral capabilityStatuscatalogued

Can support

Not independently assessed by FronteraEval yet.

Cannot support by itself

No inference beyond the upstream source should be made until the protocol is reviewed.

Original sources