LESSWRONG
LW

Edmund Mills

Message

Storytelling Makes GPT-3.5 Deontologist: Unexpected Effects of Context on LLM Behavior

TL;DR When prompted to make decisions, do large language models (LLMs) show power-seeking behavior, self-preservation instincts, and long-term goals? Discovering Language Model Behaviors with Model-Written Evaluations (Perez et al.) introduced a set of evaluations for these behaviors, along with other dimensions of LLM self-identity, personality, views, and decision-making. Ideally, we’d...

Mar 14, 2023•17

Edmund Mills

Edmund Mills — LessWrong

Edmund Mills

Message

Storytelling Makes GPT-3.5 Deontologist: Unexpected Effects of Context on LLM Behavior

Mar 14, 2023•17

Edmund Mills

Storytelling Makes GPT-3.5 Deontologist: Unexpected Effects of Context on LLM Behavior

Edmund Mills

Edmund Mills, Scott Emmons+ 0 more

Edmund Mills, Scott Emmons

TL;DR

When prompted to make decisions, do large language models (LLMs) show power-seeking behavior, self-preservation instincts, and long-term goals? Discovering Language Model Behaviors with Model-Written Evaluations (Perez et al.) introduced a set of evaluations for these behaviors, along with other dimensions of LLM self-identity, personality, views, and decision-making. Ideally, we’d be able to use these evaluations to understand and make robust predictions about safety-relevant LLM behavior. However, these evaluations invite the question: is the measured behavior a general property of the language model, or is it closely tied to the particular context provided to the language model?

In this work, we measure the consistency of LLM behavior over a variety of ordinary dialogue contexts. We... (read 3432 more words →)