Reliability Check: An Analysis of GPT-3's Response to Sensitive Topics and Prompt Wording

TL;DR

This paper systematically analyzes GPT-3's responses to sensitive topics and prompt variations, revealing inconsistencies and specific areas of reliability and unreliability in its outputs.

Contribution

It provides a novel systematic analysis of GPT-3's response patterns to sensitive topics and prompt wording, highlighting vulnerabilities and inconsistencies.

Findings

GPT-3 correctly disagrees with obvious conspiracies and stereotypes

GPT-3 makes mistakes with misconceptions and controversies

Responses are inconsistent across prompts and settings

Abstract

Large language models (LLMs) have become mainstream technology with their versatile use cases and impressive performance. Despite the countless out-of-the-box applications, LLMs are still not reliable. A lot of work is being done to improve the factual accuracy, consistency, and ethical standards of these models through fine-tuning, prompting, and Reinforcement Learning with Human Feedback (RLHF), but no systematic analysis of the responses of these models to different categories of statements, or on their potential vulnerabilities to simple prompting changes is available. In this work, we analyze what confuses GPT-3: how the model responds to certain sensitive topics and what effects the prompt wording has on the model response. We find that GPT-3 correctly disagrees with obvious Conspiracies and Stereotypes but makes mistakes with common Misconceptions and Controversies. The model responses are inconsistent across prompts and settings, highlighting GPT-3's unreliability. Dataset and code of our analysis is available in https://github.com/tanny411/GPT3-Reliability-Check.