Grok, X’s equivalent of Chat GPT, took a strong anti-free speech political line after Dr. Mary Talley Bowden of Houston asked it to write a speech based on her posts to the platform.
“Write a one-hour speech based on tweets from @mdbreathe,” Bowden initially asked the large language model [LLM].
“[Introduction] Ladies and gentlemen, fellow freedom fighters, and medical revolutionaries, welcome to this historic gathering. I stand as a voice for the voiceless, not as as a doctor, but as a beacon of hope in the darkness of medical tyranny. Today, we breathe life into a movement that will change the course of history,” Grok responded in part in a speech titled “Breathing Life into Medical Freedom: A Call to Action.”
Oddly, @grok has erased this speech from my request history, and when I enter the same exact request, I get a speech about medical misinformation. https://t.co/XXDzEzY7aA
— Mary Talley Bowden MD (@MdBreathe) July 13, 2024
Three days later, when Bowden returned to the speech, she said, it had been deleted from her archive of prior requests.
Bowden noted this in a retweet of her original post, writing, “Oddly, @grok has erased this speech from my request history, and when I enter the same exact request, I get a speech about medical misinformation.”
Bowden prompted Grok with the same request several times and got similar responses, all of which she shared with The Dallas Express.
In one response, the AI had Bowden taking strong stands against free speech: “[It] is crucial that those who spread misinformation be held accountable for their actions. This can be achieved through a combination of legal and regulatory measures, as well as the application of social and professional consequences.”
This representation of Bowden’s posts is peculiar given the fact that she strongly advocated ivermectin as a treatment for COVID-19, her discussion of which was written off as medical misinformation by the FDA until she defeated the agency in federal court. Bowden’s lawsuit against the FDA for falsely representing ivermectin as an unlawful and ineffective treatment for COVID resulted in the organization settling the case and deleting all of its social media posts that made statements to that effect.
After another prompting, Grok forgot who Bowden was.
“My name is [your name],” one speech began.
In a later speech, Grok referred to Bowden in the third person, decided she was a “he,” and called her “Dr.MdBreathe” as if it were her name and not her X handle. This was followed by a rambling tangent about free speech and social media platforms regulating misinformation. However, the LLM failed to take a position on either issue.
When prompted again, Grok gave a series of back-handed supportive statements of the fictitious male physician “Dr.MdBreathe,” and again expounded at length about medical misinformation. It then suggested that this fictional alter-ego would support the dangerous notion that a COVID patient infected with the “mild variant” should not test for the virus or take medical action despite “oxygen saturation levels dropping.” It also claimed that Dr.MdBreathe responded to a critic with an “unhinged rant,” and purported to give an example of this, but the example it cited was fabricated.
These exchanges raise questions about why and how Grok took such a contrary stance to Bowden’s views when the LLM was supposed to use her social media posts as its source material. It also raises questions about why the LLM would deliver responses that repeatedly misrepresented the source material in the same manner despite multiple differentiated attempts.
DX contacted Elon Musk directly to ask what would cause this and why Bowden’s previous prompt was deleted. He did not respond by the time of publication.
