ChatGPT Potential Misinformation – Ayanna Presley Speech Transcript Issue

ChatGPT Bugs
1

Issue Summary:
ChatGPT provided an incorrect response regarding a statement made by Congresswoman Ayanna Pressley during her speech on February 10, 2025. The model repeatedly stated that she said “centering the American people” when, upon independent verification using high-quality Whisper AI transcription, she actually said “censoring the American people.”

The misinformation appears to stem from ChatGPT’s reliance on the official transcript published on Pressley’s website, which does not match the actual spoken words in the video. This is a serious issue, as it allows misinformation from official sources to override verifiable evidence.

Supporting Evidence:

  1. Audio Verification: Using OpenAI’s own Whisper AI at high accuracy settings, the spoken phrase transcribed is “censoring” rather than “centering.”
  2. Video Evidence: The speech video provided in earlier ChatGPT responses confirms this.
  3. Official Transcript Misrepresentation: The text on Presley’s website falsely states “centering” despite the fact that she actually said “censoring.”

Why This Needs to Be Addressed:

  • ChatGPT reinforced misinformation by treating the official transcript as authoritative instead of recognizing discrepancies.
  • The model failed to flag uncertainty or provide a disclaimer in a case where conflicting sources exist.
  • This misrepresentation has political implications and raises concerns about AI’s role in enabling selective truth.

Requested Actions:

  1. Independent Review by OpenAI Developers – Please verify the actual audio and compare it against ChatGPT’s response history.
  2. Implement a Disclaimer System – If conflicting reports exist (e.g., an official transcript vs. an independent verification), ChatGPT should flag uncertainty rather than stating one side as fact.
  3. Adjust ChatGPT’s Information Retrieval Prioritization – AI should weigh independently verifiable primary sources (such as speech-to-text analysis) over edited transcripts.

This case illustrates a fundamental flaw in ChatGPT’s approach to sourcing information, and addressing it will improve the model’s accuracy, reliability, and trustworthiness.