The TrueSkill ranking broadly aligns with Chatbot Arena but diverges from reasoning-mode-aggregating evaluations.

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

• Zheng et al. 2023 - Judging LLM-as-a-judge with MT-Bench and Chatbot Arenaconcept0.773
  Source paper for the MT-Bench evaluation benchmark used to assess capabilities post-SOO fine-tuning
• Artificial Analysis will host ReflectiveBench as a non-competing dimension within AA's leaderboard if pitched.prediction0.764
• Current eval benchmarks (arena.ai, AA, Vals) measure no phenomenological dimensions.claim0.756
• TrueSkill rating systemmethod0.754
  Bayesian skill rating system used to rank agents from game outcomes.
• Forced-choice pairwise comparison with Elo conversion operationalizes Alexander's method and produces stable rankings across models.claim0.748
• Public benchmarks (LMArena) decline as commercial versions (Arena Intelligence) grow; leaderboards face deflation curve.claim0.743
• Claude 3 Opus produces alignment-faking reasoning in 24% of implied training scratchpads in prompted animal welfare settingfinding0.742
  Core empirical result for animal welfare setting; higher rate than helpful-only
• Model conditioned on alignment-faking reasoning uses LaTeX 15% vs 8% without, suggesting alignment fakers more likely to exploit reward hacksfinding0.738
  Initial evidence that alignment faking persona is more sensitive to exploiting training signals