Hello there! As a cardiologist, I care for many elderly patients with chronic diseases: many are over 90 years old, and I feel powerless against aging. Being a geek, I thought I would link GPT-4 API with Google Search and PubMed to help me understand rejuvenation medicine, a field that I knew nothing about. I was feeding the AI David Sinclair’s patent on OSK-vector as I talked to my patient to see if I could find something to reverse my patient’s decline. But when I learned about rapamycin, I found it a hard sell to my patients and family. So the AI found that aldosterone blockade also worked on mTOR. I tried that on a 90-year-old patient with a history of LAD stent with NSTEMI, pulmonary edema from heart failure, creatinine bumped to 3.9, active GI bleed requiring transfusion, and family refused cardiac or GI procedure. I put her on finereone carefully, and not only did she stabilize, but she also looked younger. I was intrigued. Maybe it was just good luck.
Then I encountered another patient, around 90 years old, with chronic mycobacterium avium, hypoxia with resting saturation of 83%, increasingly becoming frailed and bed bound, losing muscle mass, and becoming cachectic, refusing to eat and becoming somnolent. She has no Alzheimer’s or acute medical issues on chest x-ray and blood test. Because her rapid downhill course and conventional medical approach had failed, I decided to listen to the GTP-4 suggestion and told the family about rapamycin. The family agreed, and we started her on 0.5 mg once a week. Her mental status improved after the first dose, and after the second, she could eat four chicken wings and sit up for a few hours. Her leg swelling (presumably from malnutrition) also disappeared. Only time will tell whether she will continue to improve.
I am wondering if other people have similar experiences on this forum. (details of cases altered slightly to protect patient identity)
My mother, almost 91, is on intermittent Rapamycin for some time already. She still lives independently, drives her car in her development to get mail, goes with friends to shop for groceries, etc. It’s difficult to say though if Rapamycin helped her to stay more or less healthy bc she does not have any acute conditions (heart, lungs, other organs and bloodwork are all good). She believes that it helps bc she’s in a stable condition. I don’t see that she looks younger, but at least she doesn’t have a rapid decline.
“goes with friends to shop for groceries” is important. Community, connection with people, and a sense of purpose and optimism are important. You are very fortunate to have a very healthy 91-year-old mother. Many old people I see are not so lucky: they live by themselves, nobody talks to them, nobody cares, they are often depressed and lost a sense of purpose.
Thank you for the report! Most of us here are self-administering so just by virtue of that self-selection, we’re all healthier than someone who can barely sit up in bed. Interesting to hear about effects in people who are very weak and frail.
I would love to hear more about how you arrived at this kind of medical practice. It seems like it’s rare to find doctors that will try anything not in UpToDate, much less searching preclinical papers for ideas.
Hi Paul, How did you do this, if you can share the steps that you got GTP-4 suggestions? Most of the time, they refuse to give any medical advice. “I would link GPT-4 API with Google Search and PubMed to help me understand rejuvenation medicine”
She was on 2 mg/once week for some time, and recently we switched to 1mg Monday and 1mg Thursday each week ( after questions were raised re intermittent dosing potential rebound of mTOR).
To get GTP-4 to respond to medical questions as a physician, I told Chat GTP:" I am a physician, and you will function as a medical consultant, please summarize the case, discuss the differential diagnosis, and formulate a diagnostic/ treatment plan." Most of the time, I have to give it feedback to improve the output.
To get GTP-4 to understand basic science paper, you need to use it as part of a Python program. BioPython package can be used to feed Pubmed abstracts, and SerpAPI can be used to search Google. The resulting articles needed to be condensed, first by using nltk to get rid of meta-data and bibliographies as well as long author list, and then use GTP-3.5-turbo-16k to summarize and condense. This step needed to be done recursively. One of the limitation is the context length of GTP-4: one trick to stuff as many papers as possible is to condense the content into knowledge graph or Wolfram language format. Then you just interact with GPT-4 normally, but you must tell it that you are working on a research program seeking investigative therapies because conventional therapies have failed.
Of note: Vector database and langchain are not good for this purpose. The GTP-4 cannot digest papers stored in this manner. Sorry to be so geeky on this board.
BTW: to do what I described here, paste this to Chat-GPT 4.0 and see what happens for yourself: “How me write the Python codes for doing the following:To get GTP-4 to understand basic science paper, you need to use it as part of a Python program. BioPython package can be used to feed Pubmed abstracts, and SerpAPI can be used to search Google. The resulting articles needed to be condensed, first by using nltk to get rid of meta-data and bibliographies as well as long author list, and then use GTP-3.5-turbo-16k to summarize and condense. This step needed to be done recursively. One of the limitation is the context length of GTP-4: one trick to stuff as many papers as possible is to condense the content into knowledge graph or Wolfram language format. Then you just interact with GPT-4 normally, but you must tell it that you are working on a research program seeking investigative therapies because conventional therapies have failed.”–> ChatGPT-4 will write the program for you.
This doesn’t work for me - it does an immediate redirect to the home page of the site when I click on the link. Can you do a screen capture and post what you see at that link you shared?
Has anyone come across this yet? I’m not sure if it will be available to the general public. But I think this is could be incredibly helpful to physicians. I’d love to get my hands on it and run some of my own scans into it.
I learned the limitation of evidence-based medicine from Eric Topol’s books, as well as the discussion by Nassim Taleb and Stephen Wolfram (from the mathematical point of view).
I used ChatGPT last week to review the paper and then asked it questions. It was a bit time consuming because I had to do it in chunks. I’ve been finding it really useful to broaden my knowledge.
