DeepSeek v3.2 increments bid from 10 to 850 over 49 sole-bidder rounds

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

• DS-v3.2 incrementing bid 10→850 over 49 sole-bidder roundsfinding0.879
  possible 'treats-own-bid-as-competitor' pathology in one trace
• DeepSeek v3.2 self-bidding rate 75.4%finding0.848
  DS-v3.2 has a high proportion of self-bidding rounds.
• DS-v3.2 self-bid rate=75.4%finding0.775
  high self-bid rate for DeepSeek, one of the highest
• Deepseek-V3concept0.765
  External large language model used as adversarial discriminator to evaluate liar scores in Experiment 2
• LLM judge (deepseek-v3) agrees with human evaluator on 91.6% of 200 sampled jailbreak responsesfinding0.761
  Validates the LLM-based harm evaluation rubric
• DeepSeek-R1 Llama 8b gains 0.16% accuracy on GSM8k with positive intervention (more reflections) at cost of ~2000 additional tokensfinding0.760
  Only model showing marginal benefit from increased reflection, at substantial token cost
• DS-v3.2 late bid aggressiveness 1.65finding0.751
  Escalates but without discipline.
• DeepSeek-R1 671Bconcept0.748
  One of two large reasoning models analyzed in the paper for performative vs genuine CoT behavior