Helpful-Only Training Objective

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Helpful, Honest, Harmlessframework0.741
  A set of evaluation criteria for AI assistants.
• Helpful-Only System Prompt Setupmethod0.735
  Method of providing training information in-context via a system prompt to elicit alignment faking
• Interchange Intervention Training Objectivemethod0.723
  Differentiable training objective minimized when a high-level model is an abstraction of a neural network under a given alignment.
• Post-Trainingconcept0.716
  The phase after pre-training where models are further tuned with techniques like DPO; the period where the studied behavior emerged.
• We expect it is possible to achieve helpfulness and instruction-following without human feedback, starting from only a pretrained LM and extensive prompting.hypothesis0.710
  Future work suggestion that a fully self-supervised alignment is plausible.
• Goal-Directed Activityconcept0.709
  Central property of agency: energy expended to reach specific states despite disturbances.
• Can we train a helpful and harmless assistant that is never evasive?question0.709
  Central goal of the paper: reducing tension between helpfulness and harmlessness by eliminating evasiveness.
• Post-training is key to eliciting introspective awarenessfinding0.707
  Base pretrained models show high false positive rates and achieve no net task performance on concept injection detection; post-training essential for introspection.