In the past two days, there has been a frenzy about a post that claims GPT-4 saved a dog's life. In just one or two days, thousands of people reposted it and tens of thousands of people liked it. It is now a hot topic of discussion at the moment. At first glance, everyone must be wondering: What could be the relationship between these two? Can GPT-4 still have eyes to see that dogs are dangerous?
Well, here is how everything played out
According to the poster with the name “Cooper” (@peakcooper), his dog was diagnosed with a tick-borne disease. Although it was accompanied by severe anaemia, it improved relatively after a period of treatment. But unexpectedly within a few days, his condition took a turn for the worse. The dog's gums were pale and without blood. Cooper quickly took the dog back to the vet, and the blood test results this time were worse than the last time.
In order to rule out other concurrent infections related to tick-borne diseases, the doctor conducted many tests on the dog, but the results were all negative. Seeing that the dog's condition was getting worse, the doctor still couldn't give a diagnosis for the condition. The doctor told Cooper that they should just wait and see. Really? Cooper wondered. Let's wait while my dog die?
For most people who keep pets, pets are much loved and they will do anything to keep them safe. Cooper was no exception. He could not accept the doctor's suggestion at all and was going to a second hospital.
At the same time, he thought of GPT-4. When GPT-4 was released, it announced the results of a test of medical knowledge developed by the American College of Physicians. In the end, it answered 75% of the questions correctly, which is a big leap from 53% of GPT3.5.
Maybe GPT-4 is really good at diagnosing disease, Cooper thought. So he did what he said, and on the same day he told GPT-4 one by one about the dog's symptoms from the first onset, the treatment process, and each blood test report
(Query: On the 20th, my dog had a high fever of 41.5 degrees Celsius. We took him to the vet immediately, the blood test result was anaemia, and the vet diagnosed canine babesiosis. [Attached the blood test results of the day]. The vet gave the dog some drugs and after taking them, the dog's fever subsided the next day. The next 3 days the dog was treated with antibiotics (by taking pills at home) and the dog was showing signs of improvement, but on the 24th I noticed his gums were very pale and we went back to the vet)
(Query: The doctor tested the dog again, ruled out other common tick-borne diseases, and did another set of blood tests: [attach results]).
Cooper then asked GPT-4:
Taking all of the above into account, what really happened?
Soon, GPT-4 began to answer.
The response starts with an “I'm not a veterinarian” disclaimer and then said it “could try to help you analyze your blood report and tell your dog what might have happened”.
Gizchina News of the week
It first analyzed the two most obvious abnormalities in the first blood report, and then said that the babesiosis diagnosed by the doctor fit these phenomena. He then identified three abnormalities in the second test results and told what each one represented, telling Cooper that these results indicated that the original treatment was not fully effective or that there might be other underlying problems causing the anaemia.
Cooper nodded, thinking that what GPT-4 said was reasonable, and then grabbed the last sentence and asked immediately:
What do you think other potential problems might include?
GPT-4 once again stated that it is not a vet, and then gave several options:
1. Hemolysis: Destruction of red blood cells due to various causes, such as immune-mediated hemolytic anemia (IMHA) , toxins, or infections other than babesiosis.
2. Blood loss: Internal or external bleeding, which can be caused by trauma, surgery, or gastrointestinal problems such as ulcers or parasites.
Since Cooper had already ruled out other co-infections with the 4DX test and internal bleeding with ultrasound in the hospital, he locked his sights on IMHA. He thought this should be the only result that meets the situation at present.”
So, when Cooper brought the dog to the second hospital, he directly asked the doctor if it might be IMHA. The doctor said it was possible. After a series of tests, the diagnosis was confirmed. After finding the root of the disease, and after a few days of prescribing the right medicine, the dog was fine.
Cooper was ecstatic.
Only then did he know that these two diseases are very common. The former babesiosis is the number one tick-borne disease, and the latter IMHA is a common complication of the disease, especially in the breed he raised. body. Finally, the excited Cooper reviewed the entire treatment process and posted it on the Internet. He said he will provide all relevant into if OpenAI is to conduct research on the case.
Is GPT-4 really good at medical diagnosis?
Issues like this usually sparked heated discussions among people. Some people hold a conservative opinion, thinking that this is just an example, saying that the results can be found by using google.
(ps. This disease is really not an intractable disease. Some netizens said that after reading a description of Cooper, they knew that his dog might have IMHA.)
However, we should know that Google will not analyze every indicator in the detection report for you like GPT-4 (this is what Cooper appreciates most about GPT-4). In most cases, if you ask Google about a headache, it may directly throw you a “brain tumor” response.
On the contrary, ChatGPT (GPT-4) is very cautious, it will generally start by telling you “headaches can have many different causes, and while brain tumors are possible, it is not the most common cause. Most headaches are usually caused by other factors, examples include nervousness, sinus problems, or migraines.”
It then lists brain tumor-specific headache symptoms and tells you to call your doctor right away if you experience them.
Let's put aside the difference between Google and GPT-4 in diagnosing diseases for the time being. In fact, this matter is just an example, not so.
However, let's not forget that a doctor failed to diagnose the disease correctly and the other doctor which did was given direction by Cooper. The second doctor was able to rule out certain possibilities based on the first doctor's work. Amazingly, when GPT-4 gave the correct result in just one round.
But in fact, there are also very pertinent points of view worthy of note. The ability of AI does not lie in how cutting-edge knowledge prediction it has, but in that it can conduct comprehensive thinking on all possibilities of the problem without any influence, which can easily surpass many human practitioners.
No doubt, GPT-4 did a great job helping Cooper to save his dog's life. However, always remember that GPT-4 was not Cooper's first line of action. It will ve a very wrong idea to use GPT-4 for diagnosis and treatment. This could be harmful. It's best to remain with a physical doctor for all ailments. However, when diagnosis appears like a huge challenge, GPT-4 is a great last resort option.