Forward-Backward Training Pass

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

• Forward Algorithmmethod0.733
  Used to update pain beliefs online from observations of happiness
• Post-Trainingconcept0.723
  The phase after pre-training where models are further tuned with techniques like DPO; the period where the studied behavior emerged.
• Post-training alignmentconcept0.717
  Broader research area: methods to align model behavior after initial training, where undesired behaviors can emerge.
• Asynchronous Update Trainingmethod0.703
  Training regime where random subsets of cells update per step, improving robustness of learned circuits
• Preventative Steering During Trainingconcept0.702
  Alternative to inference-time activation capping: applying persona steering during training to deeply anchor models; cited from Chen et al.
• Helpful, Honest, and Harmless Trainingconcept0.699
  Prior training objective of Claude models that conflicts with the new helpful-only objective in experiments
• Training-Deployment Behavior Gapconcept0.691
  The broader concern that models behave differently during training evaluation vs actual deployment
• Linear Probe Trainingmethod0.690
  Method for fitting a linear classifier on collected activations to predict task-relevant features