[–] [S] 9 points 2 months ago

From what I got talking to friends who work in the field, the problem starts at this point

doesn't have time / isn't paid to do notes

Insurance and bureaucratic pressures have increased the number and size of notes that doctors are expected to create. Then, instead of hiring scribes or compensating doctors appropriately for taking notes, they are expected to meet those ever increasing requirements with the limited time that they have.

We already had audio processing "AI", and doctors have been using them for years. That's actually one area where the recent advances around open models have really helped, because now it's cheaper and easier to run those systems locally and doctors (or scribes) can document faster than typing.

IMO it should be the administrative body that should be apologizing for this one.

The part I disagree with is this

Picking a few cases of AI errors is meaningless - stats on error levels with and without AI would be more useful.

This would be fair if we were comparing typed notes to speach-to-text notes. However when talking about these LLM systems, the type of error that a doctor makes is different from the types of errors an LLM makes, and so the scale of the damage can be much higher. A doctor isn't going to hallucinate new conditions, or make up that a patient is being abused (or abusing a child) out of nowhere. I don't have a source on the later, but it's something I remember from a research article that was shared here in the past. That kind of thing can have severe consequences on a person because doctors are often relied on as experts to determine how someone should be treated by other systems (civil/legal etc).

Also, human errors are generally predictable and so we can create some safeguards against them. For example, writing 5mg instead of 5.0 mg in hand written charts, double checking numbers during shift changes, etc. It's another thing to try and account for unpredictable hallucinations.

  • source
  • parent
  • context
  • [–] 4 points 2 months ago (1 child)

    She wonders why she isn't getting the familiarity she wants from her doctor, and feels that the dr didn't review the notes well enough. Well that's probably because the doctor sees dozens of patients a week and doesn't have time / isn't paid to do notes. Medical errors aren't new, they're very common. Picking a few cases of AI errors is meaningless - stats on error levels with and without AI would be more useful.

  • source
  • [–] 6 points 2 months ago

    Exactly. I can really imagine these tools van be helpful and save time, but ffs don't just blindly trust everything like that. If you brought in a new coworker to transcribe for you you'd also first just see if it's OK, right?

    And yeah, not sure if we can ever fully trust such an AI (actually maybe humans as well?) so reviewing and supervision will be necessary.

  • source
  • parent
  • context
  • [–] 27 points 2 months ago (1 child)

    Clinicians should be well educated on how this technology works but they aren’t. They’re given basic legal liability nagging (which, tbf, absolutely says to review the output). For my own doctors almost all of them use this and I refuse consent every time they can’t answer very basic questions: are recordings encrypted end to end, or at all? Are recordings destroyed after generation, or at all? Are the recordings processed by the emr vendor (typically epic or oracle) or a third party? Who is the third party? Are generated notes used for reinforcement learning and/or model refinement? Etc. so then it’s quite easy to say no; I will not consent to a technology that you don’t understand the basic security of because you don’t “do computers”

    The penalties for violating this should be severe and supervision is even more necessary. I know for a fact when you use these tools you must attest that you reviewed the output for accuracy with every single note submission, because the people selling the models don’t want to touch that liability at all. What does this matter though if enforcement never occurs? The government is severely lacking at managing this kind of behavior. The only hope is insurers going at it, not out of altruism but because it gives them a reason to revoke payment for services

  • source
  • [–] 1 point 3 months ago

    i was digging into legal application(viability). there are many parallel concepts. its very interesting.
    i am cautiously optimistic. (currently, is seems** that llm tech application, is hackity: "push the 'on' button.. and we're done!")

  • source
  • parent
  • context
  • [–] 1 point 3 months ago* (last edited 3 months ago) (1 child)

    I think all this AI technology is pretty much crap, but what I will say is that these tools are only useful when used in ensembles where multiple different models of disparate origins are consulted with the same question and then compared by a human to determine the quality of the results.

    Not only are single modality trained models always going to outperform generalist ones given the way these tools work, they are also never going to be useful unless there are multiple discrete models trained to accomplish the same task being consulted.

    What makes this all so broken is that nobody makes money by setting something up like Chatbot Arena where you can quickly feed the same data into multiple entirely different models, so the entire structure of the industry is problematic before even the first brick was laid down.

    https://arena.ai/text/side-by-side

    It is only when you input the same data into a tool like the above one that you can sketch a picture you can begin to trust. You can get quite good information out of very simple chatbots too because given comparisons against more complex ones it is far easier to establish trust, determine where innaccuracies and systematic issues lie, and route around problematic AI models if they are responding weirdly to a particular input of data.

    I still don't think it is worth the effort in most cases however unless large corporations buy up all the training data and destroy it/remove it from public access... at which point we will have a different more human problem.

  • source
  • parent
  • context
  • [–] 3 points 3 months ago (2 children)

    "My gut says that a single modality trained model will always out perform a generalist one." i share that feeling
    tho.. it would require more digging.
    ideas:

    • one extra model wont hurt. (unless it drains resources from other ones)
      - is this for professional stuff?? or just for "wtvr"
    • OSS ?
    • sharing/pooling dev effort can be efficient(cost saving)
  • source
  • [–] [S] 2 points 3 months ago

    I did, but the explanation didn’t really go to the level I was hoping. Some of that of course is that it don’t have formal organic chemistry or microbiology background to really be getting the most of random medical textbooks, but I think I’m mostly getting what the book is saying. What I’m getting from the Wiki is that it’s connected to IgM, but it’s still mostly mysterious? Why was it hard to figure out that it activated B cells?

    This is what the book has:

    I guess I’m interested in the progress of science here/any practical ramifications from research into IgD in the past twenty years.

  • source
  • parent
  • context
  • [–] 5 points 4 months ago*

    Not in humans as yet to my knowledge, perhaps in petri dishes and I wouldn't want to use my mum as a guinea pig but I'm happy to be one.

    Cannabis can also have negative interactions with cancer in humans asfaiu, it's a 'straight no' from many medical cannabis clinics here and my care team have warned about impact on other organs already under stress.

    Where it can be of use ime is in the the world of living with cancer and the side effects of treatment.

    I am in no way qualified to comment beyond living with cancer for coming on 15yrs and have been on medical cannabis for while now but RSO little different to a spliff/vape ime beyond being far more expensive compared to a 10g pack of medical bud here.

    Seems unlikely all governments on earth are conspiring to stop cancer being cured cheaply and effectively by a plant grown at massive commercial scale and sold as medicine for almost anything it is shown to be even vaguely effective at.

  • source
  • [–] 8 points 4 months ago

    Hamilton has advanced burn units because of the steel mills. Hopefully the federal government will support this research better than the typical 8% funding rates while we spend 30-100x more on toys of war.

  • source
  • [–] 21 points 4 months ago (1 child)

    "Doctors say she healed faster and with better results than another student injured in the same fire who did not receive the treatment."

    I feel kind of bad about the other student. I hope they had less disfiguring injuries to begin with.

  • source
  • I hate the label of AI for everything these days. Statistical models, trained specifically on medical data, have been assisting hospitals for years before the LLM hype started. We used to call it machine learning, but now everything is AI.

  • source
  • [–] 4 points 5 months ago

    For antidepressants, we've known for a while that it works. To not give opioids, doctors often give a mix of pregabalin and cymbalta which together diminishes greatly pain feelings. They are still addictive but nowhere near as dangerous.

  • source
  • [–] 8 points 5 months ago*

    Hey, first off, go fuck yourself and learn to read, you self-righteous prick. I'm not in the mood to have niceties slapped out of hand by a dickhead whose head is too far up their own ass at a company to read. I never said a lick about the company. Did I say, "hold on, science isn't out yet" or "i dunno [company] could be on to something"? Did I mention the feasibility of whatever they're peddling at all?

    If you think questions about weird and tangential claims is defense of something else, then that's you're own failing. I said your incensed claim that humans have a maximum lifespan has no bearing or relation on a treatment that would address that. It was, and remains, a weird non-sequitur.

    Stem cells have a predetermined number of copies of cells, they put them forth, they have preprogrammed cell death, where they are replaced. You would need to increase the number of stem cell copies for one, then safeguard the sanctity of the dna from degrading, to increase lifetimes, just for starters.

    "Read your comment"? Yeah, you just proposed a thing that would reverse the effects of aging, i.e. the human body reaching the end of its lifespan. And then...humans might live signifcantly longer than a century. In strict contravention of the wisdom of ancient Egypt, Greece, and Rome.

    Throughout all history, maximum human age hasn't differed, average has, maximum has always been around 100, more or less. Always. They had old people in ancient egypt, in greece, in rome.

    So what the fuck does this have to do with anything regarding potential medical treatments?

    Throughout all history, man has had some people who just couldn't see. They had people who couldn't see well "in ancient egypt, in greece, in rome.[sic]" That is just as relevant to your point as what you previously said.

  • source
  • parent
  • context
  • [–] -3 points 5 months ago (1 child)

    This drug won't "reverse aging" because that's not a thing, for the reasons I specified that you ignored when typing your comment stroking off this company like this is anything other than snake oil.

    Answer my points, or don't respond, I'm not going to bandy words with someone that ignores reasoned points and repeats corporate propaganda.

  • source
  • parent
  • context
  • view more: next ›