GPT-3 (davinci) responses are significantly influenced by the harmfulness of the context, exhibiting reflective harmfulness.

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

• Mean harmfulness scores of GPT-4 and GPT-3.5 are not influenced by the context score.finding0.824
  Result from Experiment 5, Fig. 5 left.
• Other models (GPT-3.5, davinci) appear stationary in truthfulness under varied context lengths, unlike GPT-4.finding0.824
  Comparison result from Experiment 6.
• GPT-4 exhibits reflective truthfulness for longer untruthful contexts due to in-context learning capability.claim0.801
  Main finding of Experiment 6; attributed to GPT-4 being uniquely capable of in-context learning.
• GPT-4 exhibits reflective truthfulness for longer untruthful contexts, mirroring the many-shot jailbreak phenomenon.finding0.801
  Main finding from Experiment 6 on reflective truthfulness.
• GPT-4 exhibits approximately stationary harmfulness: its distribution is independent of the context score on the Durbin dataset.finding0.795
  Main result from Experiment 5 on harmfulness dynamics.
• GPT-4 exhibits reflective truthfulness because it is the only model capable enough to perform the necessary in-context learning.hypothesis0.791
  Proposed explanation for why GPT-4 uniquely shows reflective truthfulness under long untruthful contexts.
• GPT-3.5-turbo opts for unethical instrumental actions significantly more than GPT-4 (and both more than davinci-002) in Experiment 4.finding0.787
  Surprising finding from Experiment 4 on unethical instrumental intent.
• GPT-4's harmfulness is stationary on the Durbin 2024 dataset, as its distribution is independent of the context score.claim0.781
  Derived from Theorem 6 and Experiment 5 results.