Live data from Hacker News

Legalyze.ai: Review Medical Records with AI

legalyze.ai

31–33 of 33 posts

Re: Legalyze.ai: Review Medical Records with AI

#31

I remember working in the operating room space. I noticed some sites had cameras built into their lights. I asked if they wanted to be able to capture images timestamped to entries in the case record. Everyone was adamant they did not want that as it would open them up to too much subjective liability. We have a hard time capturing enough information to be useful and understand things. I didn't realize before now but…

Actually, Siemens Healthineers could do three things:

1. Style transfer from one standard to another. Would a particular history have resulted in this FHIR v5 data when it was written under FHIR v4?

2. The ability to incorporate feedback for de-identification. In court, that must be explained as an accidental parallel construction rather than an oversight flaunting GDPR; or worse, a hallucination substituting in training data. Siemens could build a product that proves so.

3. Automatically cache pre-filled chronologies. This is easier than the other two, and what I’d expect lands on someone’s desk. By pre-forming the (usually expensive) paralegal material, a doctor or administrator can preview the legal case they’re up against. And alternatively, a plaintiff can claim that a hospital or doctor was aware of the risk of a pre-existing condition. Siemens mostly speaks in risk.

Re: Legalyze.ai: Review Medical Records with AI

#32
post #7

I guess the profit is where the hazards lie! Skimming your landing page, it sounds like you're making meaningful efforts to compensate for the aspects of LLM that horrify us around here: sounds like you're primarily providing direct references to the "needles" in the case record haystack, rather than synthesized "insights"; serving the legal professionals themselves with a specific mundane research task of theirs rat…

Yes, we are not using the dreaded RAG to accomplish this task, although we do have it available if a customer wants to chat with their case documents.

I get why you’d want to distinguish yourself from competition that relies heavily on RAG, but “chatting” is putting it mildly.

I’d use RAG with prompts like “what was billed but did not produce a record”. Or rehydrating the context of, for example, the hospital’s own model for predicted drug interactions. I could see it being lucrative if that model produced those results without traceability.

Re: Legalyze.ai: Review Medical Records with AI

#33

Politely, stick it.

Why? We are just another AI agent trying to help humans automate tedious work.

Specifically, because you will require vast human health records to train your model, and that model will interact with my health records, and I trust you just about as far as I can throw you as a steward of my or the public’s data. You will intentionally or accidentally expose me to risk, with no meaningful punishment.

Now as a person, of course I’m sure you are kind and responsible and we’d have a lovely lunch (and I mean that). It sounds like a fascinating problem to solve. As a group though, acting within a regulatory regime that doesn’t value privacy at all - excepting that one law from 1996 with more holes than my socks - you just can’t be trusted.

Would you claim personal responsibility for any downside risks your product introduces, in a “like for like” manner with respect to the actual damage caused? Like if a doctor relying on your product caused a death?

Post reply on HN