Sampling-Based Approximation of Projection Difference

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

• Projection Differenceconcept0.835
  Metric for pre-finetuning data screening: difference between average projection of training responses and base model natural responses onto a persona direction
• Prompt Token Approximation of Projection Differencemethod0.819
  Uses last prompt token projection to approximate base generation projection, avoiding expensive model rollouts
• Postpone Approximationsclaim0.758
  Approximations and prunings compose badly; cleaner to maintain precise infinite semantics until final extraction
• Projection difference is more predictive of post-finetuning trait behavior than raw projection of training datafinding0.746
  Justifies the use of projection difference metric rather than simpler raw projection for data screening
• Projection difference better captures training data influence on persona than raw projection because it accounts for the base model's natural response tendenciesclaim0.726
  Author's interpretive explanation for why projection difference outperforms raw projection in data screening
• Simple difference-in-mean probes generalize as well as other probing techniques while identifying directions which are more causally implicated in model outputsclaim0.726
  Key methodological claim: MM probes are both competitive in accuracy and superior in causal influence
• Bethe Approximationframework0.721
  Free energy approximation using two-node marginals.
• Learned simulations can be partially observed and lazily-rendered, and still work.claim0.721
  One of the updates about prosaic ML simulation.