A Twitter user claims GPT-4 saved their dog's life after a vet couldn't correctly diagnose her symptoms

• A Twitter user claimed GPT-4 saved their dog's life after they asked it to diagnose her symptoms.
• After treatment for a tick-borne disease, her condition worsened but the vet couldn't diagnose it.
• GPT-4 made a diagnosis from blood work and test results, and another vet said it was right.

A Twitter user claimed the latest version of ChatGPT Plus saved their dog's life after a veterinarian couldn't correctly diagnose her worsening condition.

The user, who goes by Cooper on Twitter, said in a thread that their border collie named Sassy was diagnosed with a tick-borne disease and was improving after being treated by a vet. But after a few days, Sassy became worse, so Cooper took her back to the vet.

—Cooper ☕ (@peakcooper) March 25, 2023

"The blood test revealed an even more severe anemia, even worse than the first day we came in," Cooper wrote.

After running more tests to find co-infections related to tick-borne diseases, Cooper said the tests were negative and that "the vet had no clue what it could be." The vet suggested waiting to see what happened, but Cooper said that "wasn't an acceptable answer," and sought a second opinion from another clinic.

"In the meantime, it occurred to me that medical diagnostics seemed like the sort of thing GPT4 could potentially be really good at, so I described the situation in great detail," Cooper wrote. 

GPT-4 is the latest improved AI model from OpenAI, the company behind the viral ChatGPT chatbot that launched in November. Sam Altman, the CEO of OpenAI, said the new model is "more creative" and "less biased" than previous versions, and that it could pass the bar exam and score a 5 on certain AP exams.

Per his tweets, Cooper told GPT-4, which is currently available through ChatGPT Plus, that his dog had a high fever and that blood work from the vet indicated anemia. They told the AI that the vet diagnosed the dog with babesiosis and inputted two sets of blood work results from that day.

Cooper continued describing the situation, telling ChatGPT Plus that the dog's fever went away after a vet treated her for babesiosis, and the dog improved after receiving antibiotics, but after a few days, the dog's gums were pale, so they went back to the vet.

ChatGPT Plus responded that while it is "not a veterinarian," it would try to help Cooper "understand the blood work results and what might be happening with your dog," per Cooper's tweets.

The first blood test showed anemia and low platelet count, the AI said, while the second blood test showed "an increase in WBC," "a further decrease in RBC, HGB, and HCT," and "a significant increase in platelet count," per screenshots that Cooper posted Twitter.

According to the screenshots, ChatGPT Plus pointed out that the dog's pale gums were a result of worsening anemia, which "could indicate that the initial treatment was not entirely effective, or that there might be another underlying issue contributing to the anemia."

Cooper asked the AI what other underlying issues could possibly be, and it gave them a list, including IMHA, or immune-mediated hemolytic anemia. ChatGPT Plus noted again that it isn't a vet, but that it could "provide some general information about possible underlying issues that might contribute to anemia in dogs." It also told Cooper to consult a vet for "a proper diagnosis and treatment recommendations" for the dog's situation, according to Cooper's tweets.

After asking the second vet if IMHA could be a possible diagnosis, the vet agreed, Cooper wrote. After more blood tests, "the diagnosis was confirmed. GPT4 was right."

Cooper did not immediately respond to Insider's request for comment ahead of publication.

"I can't imagine what medical diagnostics will look like 20 years from now," Cooper wrote on Twitter, adding that GPT-3.5, which powers ChatGPT, couldn't properly diagnose Sassy. In a response to another user on the thread, Cooper clarified that GPT-4 "should be a tool to be used by professionals to help them save more lives."

While ChatGPT is not as powerful as GPT-4, it was found to have "moderate accuracy" and was "comfortably within the passing range" in all three parts of the US medical licensing exam. Some researchers excluded "indeterminate" answers because ChatGPT seemed to be programmed not to give medical advice, Axios reported.