DialoGPT generates more diverse utterances across all conditions than BlenderBot 1.0

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

• BlenderBot on EmpatheticDialogues: NLI Diversity increases from -8.90 to -1.72 with 16.5 samplesfinding0.779
  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 samplesfinding0.775
  DTG result for DialoGPT on DailyDialog++ using NLI metric
• DialoGPT on EmpatheticDialogues: NLI Diversity increases from 3.68 to 10.11 with 7.1 samplesfinding0.755
  DTG result for DialoGPT on EmpatheticDialogues using NLI metric
• 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.quote0.723
  Key finding about the relationship between capability and introspection.
• The same 15 properties may not apply across modalities; aliveness in voice, text, and video may differ.claim0.720
• Different introspective tasks may preferentially use different path distributions in the transformer.claim0.715
  Interpretive claim connecting exponential path combinatorics to Lindsey's layer-dependent findings.
• Most commercial AI products optimize for local turn-by-turn quality without establishing good-shape across whole conversations.claim0.714
• Almost any social process can have a relatively more living character, or have a relatively less living character.claim0.712