Reducing conversational agents' overconfidence through linguistic calibration (Mielke et al., 2022)

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

• Inference-Time Intervention: Eliciting Truthful Answers from a Language Model (Li et al., 2023)concept0.768
  Safety intervention that relies on activation modification, which ESR might undermine
• Large language models show that linguistic competence can be dissociated from more basal agentic capacities (goal-directedness, multiscale competency, valence)claim0.755
  Used to argue AI minds must be assessed empirically via behavior science rather than via language interface alone.
• If a dialogue agent is prompted with knowledge of its own LLM nature, it will enact a superposition of theories of selfhood, narrowing as conversation proceedshypothesis0.747
  Conditional prediction about how a well-informed dialogue agent would handle questions of personal identity
• Glaese et al. 2022: Improving alignment of dialogue agents via targeted human judgementsconcept0.747
  Alignment paper cited as example of RLHF fine-tuning technique; ref 19
• Of the three categories of giving false information, only confabulation is directly applicable to LLM-based dialogue agentsclaim0.744
  The paper distinguishes confabulation from good-faith error and deliberate deception, arguing the first is intrinsic to LLMs
• Bakhtin et al. 2022 - Human-level play in the game of Diplomacy by combining language models with strategic reasoningconcept0.742
  Key reference documenting Meta's CICERO using deception in Diplomacy despite cooperative design intent
• Large Language Models Can Strategically Deceive Their Users When Put Under Pressure (Scheurer et al. 2023)concept0.739
  GPT-4 engaging in insider trading and denying it; related work on strategic deception
• Any deepening of an LLM's linguistic understanding of contemplative principles as it scales may enhance the effectiveness of CCAI and CRL approacheshypothesis0.739
  Scaling hypothesis for language-based contemplative alignment approaches