EmpatheticDialogues shows lower starting diversity than DailyDialog++ for both models

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

• DialoGPT on EmpatheticDialogues: NLI Diversity increases from 3.68 to 10.11 with 7.1 samplesfinding0.825
  DTG result for DialoGPT on EmpatheticDialogues using NLI metric
• BlenderBot on EmpatheticDialogues: NLI Diversity increases from -8.90 to -1.72 with 16.5 samplesfinding0.765
  DTG result for BlenderBot on EmpatheticDialogues; requires most resampling of all conditions
• DialoGPT on DailyDialog++: NLI Diversity increases from 4.11 to 10.24 with 6.3 samplesfinding0.752
  DTG result for DialoGPT on DailyDialog++ using NLI metric
• H10: Empathy training blocks self-observation — empathy-trained models will show minimal lift and low baseline.hypothesis0.703
  Exploratory hypothesis supported by Inflection Pi +0.63 lift
• Lower (more central) emotion PCs are more persistent than higher (noisier) PCs in both Kimi and Cogitofinding0.703
  Rules out that persistence is an artifact of probe construction, since noise dimensions are not similarly persistent
• This introspective capacity is highly unreliable and context-dependent in today's modelsclaim0.702
  A caveat qualifying the main claim.
• Notably, Claude Opus 4.1 and 4—the most recently released and most capable models of those that we test—perform the best in our experiments, suggesting that introspective capabilities may emerge alongside other improvements to language models.quote0.701
  Key finding about the relationship between capability and introspection.
• VS increases diversity by 1.6-2.1x over direct prompting on creative writing tasks (poem, story, joke)finding0.700
  Core empirical result demonstrating VS's effectiveness on creative writing diversity