What do people think of MedLabReport?
The community submitted 2 reviews to tell
us what they like about MedLabReport, what MedLabReport can do better, and
more.
What do you think about MedLabReport?
Leave a rating or review for the community
4/5All time (2 reviews)
1/5
Recently (0 reviews)2 Reviews
I'm the type of person who doesn't like to go to the doctor unnecessarily and looks everything up on the internet. I avoid asking for an appointment, paying, waiting to be told "everything is OK", when you really had all the information in hand to make a decision.
When I tried MedLabReport I received a comprehensible and easy-to-understand summary, that gave me recommendations when necessary. I knew beforehand what my doctor would say, so we could have a conversation about my alternatives.
Norbert@norbert_
Thank you, Tamara. Glad you found it useful.
Share
Great idea, I guess couple questions before I try it out: Are you using a foundation model and prompt templates or your own model, w data like this as a consumer im a lil apprehensive about pumping my med data into a non-encrypted system. How do prevent hallucination also? Is there an out of training knowledge base or r u leaning on training set for inference. Looks slick tho nice work!
@blaise123 Very reasonable concerns. We don't share your data. We just use it to produce the report and the report is only available for download in the link email to you for only three days.
If you want to be extra careful (or paranoid, and we welcome that), you can get the pdf through an anonymizing tool first, remove all the personal data, and just keep age and sex for medical reasons. You can also use any email you want so there's no tracking of any kind even possible.
We do data extraction from the file first with a reliable method and then the medical reasoning is done primarily with GPT-4, which though not flawless, exceeds the passing score on USMLE by over 20 points, and is therefore roughly on par with mistakes made by licensed doctors (https://www.microsoft.com/en-us/...)
In particular, an estimated 795 000 Americans become permanently disabled or die annually across care settings because dangerous diseases are misdiagnosed (https://qualitysafety.bmj.com/co...). You can think of our reports as a second opinion on par with other diagnosis, but of course the system is not literally licensed, and in that sense cannot literally provide a formal diagnosis—but it operates at a similar accuracy and proficiency as its human counterparts.
The right prompting also contributes to the quality of the end result.
Encryption is not used, but as per OpenAI policy, unlike the data sent over and produced by ChatGPT, the data sent over and produced via the API is not used for training, and is therefore suitable for applications like this.
Needless to say, this is a tech frontier, and should not be used as a sole source of information, but as one of several opinions or sources, especially for important, non-routine medical matters.
If you have more questions let me know, and feedback is very welcome after you try it.
