LLaMA-2-70B and 13B probes generalize better across datasets than 7B probes across all training sets and probe types

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

• For LLaMA-2-70B, probes trained on larger_than+smaller_than achieve >95% accuracy on sp_en_trans regardless of probing techniquefinding0.897
  Striking cross-domain generalization result supporting the claim that larger models represent abstract truth
• MM probe trained on likely dataset achieves NIE of 0.70 (false→true) on LLaMA-2-13B, surprisingly strong but weaker than truth probesfinding0.843
  Likely-trained MM probe is a surprisingly effective causal baseline due to correlation between truth and probability on sp_en_trans
• Mass-mean probes generalize about as well as LR and CCS for LLaMA-2-13B and 70Bfinding0.837
  Despite being simpler and optimization-free, MM probes match accuracy of other techniques at scale
• Llama-3.3-70B shows multi-attempt rate of 7.4% vs. ≤1.2% for all other models testedfinding0.833
  Supporting finding showing ESR is driven by both higher multi-attempt rates and comparable improvement rates
• Llama 3.3 70B is the most likely to take on a non-Assistant persona when steered, with even split between human and nonhuman portrayalsfinding0.831
  Model-specific difference in persona susceptibility
• Llama-3.3-70B exhibits internal consistency-checking mechanisms that operate during inferenceclaim0.812
  Central interpretive claim of the paper supported by causal ablation and activation evidence
• Fine-tuning Llama-3.1-8B on self-correction examples increases multi-attempt rate proportionally with training data ratiofinding0.804
  Shows behavioral pattern of self-correction is trainable in smaller models
• 26 candidate off-topic detector latents identified in Llama-3.3-70B via contrastive searchfinding0.803
  Core mechanistic finding identifying specific SAE latents associated with ESR