41–50 of 253 posts
I also very much doubt an agitated family member will be calmed down by a doctor reading a script to them.
> Satisfying HIPAA rules around patient privacy alone may take many years, or decades, to resolve themselves before we could even contemplate directly using programs like ChatGPT in a medical theater. This is easily solved by not using someone else's API. Here's $500 worth of hardware you can get on ebay running Vicuna-13B, locally: > IV fluids may not be the best treatment for someone with severe pulmonary edema and…
How much time does this take to get produced locally?
...however, that was on cheapo ARM hardware. On a hospital budget I bet you could beat OpenAI with a local model pre-mapped in GPU memory on a 3090.
What I want (in a business context) is a LLM that can rewrite memos / docs (I am an Amazonian) in exactly the style that someone expects and is trained on other successful docs. It's not to cut out the hard work of getting the details right - its to cut down on the bullshit of misalignment from bad communication.
I know some doctors are full of baloney but c'mon. I really hope maybe that we create better doctors that have a lot of empathy, and are able to wield technology to make themselves 10x better like we're doing with programming. But, this is not the way.
Edit: I realized its fake...I think, it has to be. It doesn't make sense, like others mention. You're not reading a 5 para speech to a patient's family while they are being treated. And also, in the article, they say to double check and verify what is being said. You better proofread that too! Or it might invent something absolutely crazy, and you may be liable.
> Satisfying HIPAA rules around patient privacy alone may take many years, or decades, to resolve themselves before we could even contemplate directly using programs like ChatGPT in a medical theater. This is easily solved by not using someone else's API. Here's $500 worth of hardware you can get on ebay running Vicuna-13B, locally: > IV fluids may not be the best treatment for someone with severe pulmonary edema and…
Who is surprised that in a match between a doctor's bedside manner and a machine's that a machine wins? The cold metal of a machine is nothing next to the coldness of a doctor's heart! ;)
Regardless of my state of agitation, I would most definitely not want to be read a script prepared by an AI. In fact, were I to find out after the fact, I'd be even more incensed.
Earlier quoted context omitted.
"Trust but verify". If you can't verify, then don't trust. I've used ChatGPT for short scripts. I can read and test-run scripts to verify their correctness. On the other hand, asking it to do more complex programming is error-prone.
If you have to verify then there is no trust. That's not how trust works. I never liked the phrase.
Kind of fascinating that the benefit here is not so much artificial intelligence as artificial empathy.
They've even got this thing we call "bullshit" down pat.
Theoretically something like BioMedLM and/or a vector search over clinical research and/or GPT-4 with cited sources etc. could be useful in some diagnostic circumstances to at least give doctors ideas, right? What he is talking about here is just an incredibly common scenario that the doctor and everyone is already very familiar with. Which is basically delaying the inevitable for people who are in some ways already…
Anything is possible, but consider this doctor's review of ChatGPT's diagonistic output:
> ChatGPT works fairly well as a diagnostic assistant — but only if you feed it perfect information, and the actual patient has a classic presentation, which is rarely the case. (And you don’t mind a 50% success rate that often misses life-threatening conditions.)
For any new technology to be useful, it needs to do more than just work; it must be significantly more productive than the alternative.
Kind of fascinating that the benefit here is not so much artificial intelligence as artificial empathy.