English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Large language models are transforming research on machine learning whilegalvanizing public debates. Understanding not only when these models work welland succeed but also why they fail and misbehave is of great societalrelevance. We propose to turn the lens of computational psychiatry, a frameworkused to computationally describe and modify aberrant behavior, to the outputsproduced by these models. We focus on the Generative Pre-Trained Transformer3.5 and subject it to tasks commonly studied in psychiatry. Our results showthat GPT-3.5 responds robustly to a common anxiety questionnaire, producinghigher anxiety scores than human subjects. Moreover, GPT-3.5's responses can bepredictably changed by using emotion-inducing prompts. Emotion-induction notonly influences GPT-3.5's behavior in a cognitive task measuring exploratorydecision-making but also influences its behavior in a previously-establishedtask measuring biases such as racism and ableism. Crucially, GPT-3.5 shows astrong increase in biases when prompted with anxiety-inducing text. Thus, it islikely that how prompts are communicated to large language models has a stronginfluence on their behavior in applied settings. These results progress ourunderstanding of prompt engineering and demonstrate the usefulness of methodstaken from computational psychiatry for studying the capable algorithms towhich we increasingly delegate authority and autonomy.

Inducing anxiety in large language models increases exploration and bias