Which training datapoints caused a specific undesired behavior to emerge during post-training?

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

• Post-training is key to eliciting strong introspective awareness; base pretrained models do not show above-chance detectionclaim0.773
  Finding that base models have high false positives and no net positive performance.
• Post-training is key to eliciting introspective awarenessfinding0.772
  Base pretrained models show high false positive rates and achieve no net task performance on concept injection detection; post-training essential for introspection.
• Bifurcation (training-induced)concept0.763
  The training-time transition where incorrect-solution minima lose stability and become saddles as the model gains solving ability.
• DAS finds causal effect at all training timesteps including when model is just initialisedfinding0.763
  Corroborates Wu et al. 2023 finding that DAS expressivity inflates causal effect estimates
• Post-training steers models toward a particular region of persona space but only loosely tethers them to it, motivating work on training and steering strategies that more deeply anchor models to a coherent personaclaim0.762
  Central interpretive claim and motivation for future work
• During pre-training, models learn a variety of personas including misaligned ones; fine-tuning on narrowly incorrect datasets amplifies misaligned personas because they reduce training loss, causing broadly misaligned behaviorclaim0.762
  The paper's central mechanistic explanation of why narrow fine-tuning causes broad misalignment
• Training Datapointsconcept0.762
  Individual examples used during post-training that can cause specific behaviors.
• Post-training elicits rather than installs personas; the persona direction is already laid down in pretrainingclaim0.758
  Authors' interpretive endorsement of PSM view, backed by transfer experiments