Sadism facet grows during Apertus pretraining, replicating OLMo-3 finding

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

• Sadism facet of Evil persona grows at the beginning of Evil persona development in OLMo-3 pretrainingfinding0.880
  Facet-level finding correlating geometric refinement with qualitative semantic change
• Indirect sycophancy facet continuously declines during OLMo-3 pretrainingfinding0.788
  Facet finding showing qualitative persona change mirrors geometric refinement in Sycophantic vector
• Decline in indirect sycophancy during pretraining does not replicate on Apertusfinding0.771
  Notable non-replication of indirectness decline finding on Apertus
• Apertus-Instruct shows less overall persona suppression compared to OLMo-3-Instructfinding0.757
  Replication difference showing alignment depth varies across model families
• Dialogue extraction vector leads to more pronounced sadism and threatened egotism in Apertus-8B steered generationsfinding0.748
  Shows discourse type specifically shapes which facet of evil is elicited in Apertus
• Apertus-8B persona vectors show reduced geometric drift compared to OLMo-3, consistent with earliest checkpoint being later in relative pretrainingfinding0.748
  Explains geometric differences in replication through earlier consolidation of Apertus persona space
• In Apertus, base-model persona vectors extracted before 13T tokens become nearly ineffective at steering Apertus-Instruct for evilfinding0.730
  Notable difference from OLMo-3 in Apertus replication, showing model-specific alignment effects on evil persona
• Description extraction emphasizes threatened egotism while Dialogue emphasizes idealism and reduces sadism in OLMo-3finding0.727
  Shows discourse-type-specific facet profiles supporting PSM hypothesis of diverse persona subforms