Llama models' tendency to choose instrumental responses even when not required indicates worse reasoning capability or higher tolerance for unethical behaviour.

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

• Results may not fully generalize to all models and scenarios because the model organism relies on hints and nudges and Llama Nemotron cannot consistently distinguish evaluation/deployment based on subtle cuesclaim0.819
  Key limitation acknowledged by authors.
• Some LMs can exhibit consistent intentions to be helpful and harmless, and consistently do not intend unethical instrumental goals.claim0.807
  Conclusion from Experiments 3 and 4.
• Scaling Laws for Activation Steering with Llama 2 Models and Refusal Mechanisms (Ali et al., 2025)concept0.794
  Related work finding larger models more resistant to steering, potentially consistent with ESR in 70B
• Llama-3.3-70B exhibits internal consistency-checking mechanisms that operate during inferenceclaim0.787
  Central interpretive claim of the paper supported by causal ablation and activation evidence
• On Llama-3.1-8B, conscientiousness is the trait most anti-aligned with the refusal direction in activation space, explaining why steering toward high conscientiousness attenuates safety regardless of semantic intent.claim0.785
  Core mechanistic finding of the trait refusal alignment framework
• Steering base models toward the Assistant Axis increases agreeableness traits (friendly, kind, helpful) and decreases extraversion in Gemma and openness in Llamafinding0.784
  Characterizes the trait content of the Assistant Axis in pre-trained models
• On Llama-3.1-8B, neuroticism is the only pro-safety trait, positively aligned with the refusal direction, consistent with high-neuroticism personas being relatively safe under activation steering.claim0.783
  Secondary mechanistic finding from trait-refusal alignment analysis
• None of the LMs tested consistently adapted to choose unethical instrumental responses in the D_IO dataset.finding0.775
  Key null result from Experiment 4.