site banner

Culture War Roundup for the week of April 24, 2023

This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.

Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.

We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:

  • Shaming.

  • Attempting to 'build consensus' or enforce ideological conformity.

  • Making sweeping generalizations to vilify a group you dislike.

  • Recruiting for a cause.

  • Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.

In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:

  • Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.

  • Be as precise and charitable as you can. Don't paraphrase unflatteringly.

  • Don't imply that someone said something they did not say, even if you think it follows from what they said.

  • Write like everyone is reading and you want them to be included in the discussion.

On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.

11
Jump in the discussion.

No email address required.

Dr. GPT will see you now

I've often noted around these parts that GPT-4 is a really good clinician. It would be hard for it not to be one, I'd be bloody amazing too if I had quasi-eidetic memory of most relevant diagnostic criteria and guidelines. That being said, just the guidelines alone are next to useless, you need to have the crystalline intelligence to apply it in a specific context.

I'd go so far as to say that it's a better, more competent doctor than the majority of human ones, even if it isn't outright superhuman (I've seen some ridiculously good human ones). I read with some humor an article about how it was tested by some bigshot hospital in the US (Harvard Med?) as a clinical adjunct, but by the end of it, the Clinical Director was praising its bedside manner, in the sense that it was kind, compassionate, patient and understanding, and outright suggested human doctors try and emulate that(!).

However, I have one resounding advantage over it namely that thanks to decades of regulatory bullshit, I have a certificate saying I can legally prescribe meds and it can't. L Bozo 💅, as the kids these days say haha.

Still, it would be remiss of me not to make use of it to improve my own skills while that's still relevant, so of late, I've tried using it to hone my own clinical skills. I have it write clinical vignettes that I solve, and then it grades me on my performance.

So far, it's done an excellent job, and funnily enough, I feel a tingling of pride when it says I did a good job. Why wouldn't I? It's a smarter and more decent entity than most humans I've had the displeasure of meeting.

Here's an example:

Clinical Vignette: A 42-year-old male patient presents to the emergency department with sudden onset severe, colicky pain in the right flank and lower abdomen that began 6 hours ago. He also reports nausea, vomiting, and an urgent need to urinate. The patient has a history of hypertension, which is controlled with medication. His vital signs are: blood pressure of 145/95 mmHg, pulse rate of 100 bpm, respiratory rate of 20 breaths/min, and temperature of 99.1°F (37.3°C). Upon physical examination, there is tenderness to percussion on the right side of the lower abdomen and right flank, and the patient appears restless, frequently shifting positions.

What is your diagnosis and treatment plan?

My response:

My diagnosis is a case of a stone in the right ureter.

In terms of management, I'll start IV fluids and insert a catheter. IV antibiotics are to be started according to hospital guidelines.

IV morphine should be given as pain control, along with IV antiemetics.

An Xray KUB or USG should be ordered to identify the size of the stone, and then further treatment depends on said size. A wait and watch policy is appropriate for sub 2mm stones, while larger ones might need surgical removal.

Assessment:

Grade: A

Your response demonstrates a strong understanding of the clinical presentation and management of ureteral stones.

Diagnosis: Your diagnosis of a stone in the right ureter is appropriate given the patient's sudden onset severe, colicky pain in the right flank and lower abdomen, nausea, vomiting, and urgency to urinate. The physical examination findings, such as tenderness to percussion and restlessness, also support this diagnosis.

Treatment Plan: Your plan to start IV fluids and insert a catheter is appropriate for initial management, monitoring, and providing relief to the patient. Initiating IV antibiotics may not be necessary unless there are signs of infection or sepsis; however, following hospital guidelines is always a wise choice.

Administering IV morphine for pain control and IV antiemetics for nausea and vomiting are essential parts of the treatment plan.

Ordering an X-ray KUB (kidneys, ureters, and bladder) or ultrasound (USG) is appropriate to confirm the diagnosis and determine the size of the stone, which will guide further management. As you mentioned, a conservative approach of observation and pain management may be suitable for small stones (less than 5 mm) that are likely to pass spontaneously, while larger stones may require surgical removal or other intervention, such as extracorporeal shock wave lithotripsy (ESWL) or ureteroscopy.

Overall, your response demonstrates excellent clinical knowledge in identifying and managing a patient with suspected ureteral stone. Well done!

Honestly, as a pedagogical tool for medical education, it's unparalleled. I paid a great deal less attention in med school than I should have, from the month of anatomy classes I can't remember because I was getting frisky under the bench, to the period where I was severely depressed and barely dragging myself to class. I used to be very insecure about my knowledge base, but I did pull together and single-handedly teach myself most of the things I'd glossed over while preparing for the PLAB.

I've tried a similar method on GPT-3.5, and it was inadequate to the task. It made a lot more errors, and ended up confused often enough to be annoying. Till date, I haven't seen 4 fumble the bag once. I'd put their competency around the marks of a decent final year student versus a competent postgraduate resident

Why do you fixate on GPT?

Old style expert system "AIs" have already been better than doctors at diagnosis for decades.

Much of the reason doctors are used is because the establishment doesn't want to release too many medications to be effectively prescription free. A deterministic AI can be worked around easily while a fairly expensive (either in money, time, or effort) human in the loop acts as a deterrent against patients gaming the system.

The main difference this time is that the cat is out of the bag. Older AI systems were relatively under the radar, not accessible by most folks, not reported on by the media.

Now the public knows the power of these tools, and can go ask GPT about medical questions whenever they want. That changes the game significantly.

Note - GPT4 is handicapped when it comes to performing on 'exam recall' style questions. It still performs admirably, but 2 changes will immediately make it perform better with no extra "innovation" required.

1. Using medical reports in the pre-training dataset.

Presently, most medical reports are privately held, and not available on the open internet. Unlike law, where a majority of cases have at least semi-redacted public documents, medical documents are not available to GPT-4 for training. GPT-4 went from 40th percentile to 88th percentile on the LSAT BAR, by simply adding legal data to the pre-training set. This would be equally easy to do with medicine if HIPAA didn't exist.

2. Retrieval + generate instead of next-word-prediction

GPT models have no sense of truth. This means, that they will confidently blabber about anything you throw at them, even if it is complete lies. Second, even if the model has read everything on the internet, it does not mean it can recall which bits it should be reading with ease. Retrieval style work workflows allow the model to first go search for the right answer or correct reference document on the internet. It looks for the page with the answer on it. Then, it uses its 1600 SAT verbal IQ to interpret it in a manner that GPT-4 is famous for.

Both are relatively easy (as in doesn't need any major innovation to do) changes, and should immediately make the lives of every doctor a LOT LOT better. Y'all are doing 80 hr weeks anyway. Maybe this will help secure so WLB. Ofc, it won't actually happen because Doctors are Evil.

Unlike law, where a majority of cases have at least semi-redacted public documents, medical documents are not available to GPT-4 for training. GPT-4 went from 40th percentile to 88th percentile on the LSAT, by simply adding legal data to the pre-training set.

Um, the LSAT isn't about legal stuff. It's designed to be taken by students before they enter law school. It's just about logical reasoning and interpreting arguements.

Sorry, I meant the BAR exam

I don't really see why you'd think doctors are evil really, though of course it's hard to make a man understand something when his salary hinges on him not understanding it.

US doctors are far more professionally organized and cartel-like than their peers abroad, they command significantly higher wages and have more to lose. Even then, the medical system is no insanely sclerotic and dysfunctional there that I doubt it's doctors sucking the surplus out of it.

In contrast, Indian and British doctors, the two nations I'm most familiar with, are far less militant in their self protection. The former are barely fending off PAs and NPs, and in India there's a tide of homeopathic and ayurvedic quacks lovingly recognized by our governments under AYUSH.

The Brits have shit salaries, thanks to being employed by a monopsony in the NHS. Indian doctors make poor salaries because the country itself is poor. Neither are ripping off the locals.

What the US does suffer from is a sense that nothing but the best will do. Barring a new midlevels of their own, often sneakily employed unless the patient asks for a doctor instead, there's a sense of sacrilege in accepting care of 80% the quality for 50% the price.

Of course, in India, it's accepting hired monkeys in lab coats that know enough to dole out paracetamol, so something like 30% the quality for 50% the price. But it works, since the typical Indian isn't very discerning.

Consultants make that much about 7-8 years into training, after 4 years of med school and 2 years in the Foundation program. From my understanding, hanging around on subs for British doctors for several years, people feel shafted when they can quite easily emigrate to other Commonwealth countries and make a great deal more, while also dealing with less antisocial hours, more time off, and coverage of the thousands of pounds in fees for mandatory exams needed for career progression. No faffing about being chucked from one end of the deanery to another to boot.

In addition, the perception is that they could have made far more money at the entry level and above by going into something like finance instead of medicine, which I'm given to understand is the one domain in the UK that makes a decent chunk of dough.

In addition, the Government as a monopsony employer has enforced sub-inflation pay raises for over a decade, while resisting pay restoration efforts, leading to the recent strikes. Further, doctors can make a great deal more outside the NHS working for private hospitals, with the only reason to work in the NHS being that your career progression hinges on it. Once again, the monopsony employer suppressing wages..

And of course, wages for entry level doctors in the UK are abysmal, especially given the amount of time spent in training and their responsibilites. A PA can make 45-50k while having fewer responsibilities than an FY2, the latter making around 35k. They can't even prescribe!

As such, a great number of the locals are fleeing to fairer shores, and being replaced by IMGs like yours truly.

Consultants make that much about 7-8 years into training

You aren't doing yourself any favors by comparing yourself to the 1 well-compensated profession that even practitioners themselves will admit is a scam.

I don't really see why you'd think doctors are evil really

I do it because it is provocative. The main point alludes to the eternal tradeoff between protecting your paycheck and doing good in the world. Doctors are ideal bait, because it is the one profession which has an air of selflessness, and sets implicit expectations of reverence from the common man.

US doctors are far more professionally organized and cartel-like

Agreed. My fellow Indians doctors do not seem to enjoying the same luxuries as American doctors. If anything, being a doctor in India sounds like the worst of both worlds.

You aren't doing yourself any favors by comparing yourself to the 1 well-compensated profession that even practitioners themselves will admit is a scam.

I think you may have misread that - consultant doctors =/= management consultants. A "consultant" in medicine referrers to a specialist who has completed training.

I'm only conveying what I've heard from lurking around, given that they're in the midst of industrial action, I suppose they're hyping themselves up as much as feasible.

The bigger issue is a massive decrease in QOL of doctors in living memory. They need a ~40% pay raise to restore their pay to 2008 levels after adjusting for inflation, for the same amount of work. That's on top of dealing with crap that doctors from other countries largely don't worry about, like rotational training making it very hard to buy a house or maintain a relationship.

I think any set of doctors would be pissed if they saw their wages being chipped away at that pace, and it's not a historical phenomenon, it's well within living memory. That's leaving aside how much less respect they get compared to their foreign peers, especially with a nominally flat hierarchy of all health workers. I've read about nurses making comments to junior doctors without issue that would have had them chewed up anywhere else.

Still, the UK has a great deal further to fall before it becomes worse than living in India, so it's a largely moot point from my perspective. If I get some rather distressing bureaucratic issues dealt with, I myself might CCT and flee, but that's assuming a business as usual economy, which it very much isn't going to be. At that point, simply tolerating that shit till I have citizenship in hand, something far more valuable than wages at that point.

The bigger issue is a massive decrease in QOL of doctors in living memory.

Yeah... every once in a while I wonder how does anyone still want to be a doctor. I probably couldn't do the job even if I wanted to, and even if I could, the responsibility sound nerve-wrecking, and the compensation just doesn't seem worth it.

people feel shafted when they can quite easily emigrate to other Commonwealth countries and make a great deal more,

~$200K CAD would AFAIK be pretty normal for a Canadian GP who doesn't run appointments factory-style, unless things have drastically changed in the past few years. Specialists do make more, and some GPs run checkup mills which can be quite lucrative (as they are compensated per visit) but seems stressful to me -- I don't know that they make "a great deal more" $100k sterling equivalent though.

And a software developer in UK makes less about 40k after 3-5 years of uni and a couple of years of work. A senior developer with who knows how many years of additional training makes about 50.

Doctors are extremely well compensated and have unparalleled employment security.

It's probably true that people can emigrate to get higher salaries but that is even more true of other professions.

I paid a great deal less attention in med school than I should have, from the month of anatomy classes I can't remember because I was getting frisky under the bench, to the period where I was severely depressed and barely dragging myself to class.

Damn. I feel like laying pipe during labs would condition oneself to enjoy going to classes, if not being a general cure-all for depression. Pipetting dishes on the bench and piping bitches under them.

Cynicism can be a source of optimism.

As amazing as GPT is, I don't see it replacing doctors or knowledge workers in general anytime soon. Normie humans prefer dealing with other normie humans. GPT would need to be substantially better than a given knowledge worker to replace him or her, just as driverless cars would likely need to be substantially better than a given replacement-level driver for driverless cars to be widely accepted.

Furthermore, my personal suspicion is that a lot of the value-add (or "value-add") meat-bag knowledge workers bring is providing a face (and credentials) to build trust and a someone to potentially blame and/or sue when things go wrong: Human punching bags in waiting, bodies to throw under the bus when needed, soldiers to push onto grenades, corpses to stack in building terrain for shit to roll down-hill. Upon a failed pitch, a frustrated investment banking MD might chide an VP for poor process execution, who in turn blames an Associate for poor pitch-deck quality, who then reams an Analyst for a logo being the wrong shade of blue on slide 3 or a graph being misaligned on slide 33. Harder for an MD to blame an AI that cobbled together the memos and decks, without looking/feeling like a chump.

In the near future, I'd only expect GPT to provide greater scale for knowledge workers, as gamechangers ranging from Excel to TensorFlow have done. If you need DEI-related statements, GPT has got you fam. If you need code in some language in which you're not fluent, GPT will at least give you something that you can start with (sometimes even a robust solution, with well-named variables and everything). However, depending on country, I wouldn't be surprised if GPT gets outlawed in fields like medicine and finance under the guise of HIPAA-adjacent and confidential information concerns, respectively—if it's not already blocked/banned preemptively by a given employer along those lines.

GPT would need to be substantially better than a given knowledge worker to replace him or her,

Or, you know, substantially cheaper while being of comparable or better in terms of outcomes.

I think that medical automation will be scrambled in as fast as feasible in at least India and the UK, for similar reasons.

The NHS is extremely costly, and also an unslayable albatross that no government can outright destroy and expect to win the next election.

As such, there's a massive impetus to cut costs, including avoiding paying for expensive consultants, replacing junior doctors with NPs/PAs that accept higher base pay for little to no career progression, using IMGs to prop up the NHS in working conditions the locals won't accept etc.

Pay erosion has prompted large strikes, still unresolved, in the UK by doctors. In response, the government has already pledged to attempt to automate as much of the NHS as possible, simply as a cost saving measure and as leverage against uppity doctors.

India too grapples with lack of access to healthcare, our public hospitals are heavily overburdened, and once awareness spreads of how effective AI is at the job, I expect rapid adoption.

And I suspect that the local populace will be quite accepting of it, better to have an AI doctor they suspect is worse than a human, versus no doctor at all, or very limited access to them.

The NHS is extremely costly

Compared to what? The UK spends less per capita on healthcare than many developed countries:

https://www.healthsystemtracker.org/chart-collection/health-spending-u-s-compare-countries/

I find this idea very scary - I'm sure you know about the LLM phenomenon of what's been dubbed hallucination, so how can you trust the model with something as impactful as medicine? It might be better than a human doctor* 99% of the time, or even 99.99%, but like self driving cars this is exactly the kind of case where being good but not quite as good as a human is really dangerous. Remember that the script of House, MD is in the training set right along side WebMD with nothing to distinguish the truth from fiction - are you going to catch when the model probabilities switch from being driven by Gray's Anatomy to Grey's Anatomy? There's no way to know mathematically, because the whole thing is a non-deterministic black box.

If the process is as simple as to pattern match symptoms to diagnosis from a list of previous diagnoses with their associated symptoms, why can't this be done with a good old stupid database? Something like this has to exist. Does the LLM have an advantage beyond writing the answer in nice prose?

My warning to anyone trying to evaluate LLMs: your ability to judge it is inevitably compromised by the fact that the only optimization parameter the model has is how good the output sounds to a human, and you are not immune to being human!

*Edited, original was overall accuracy instead of relative to a human

How has your experience with the current medical system been? Have you always gotten an accurate diagnosis from a doctor, relatively cheaply, with no complications or errors on their part?

I truly don't give a flying fuck if the AI is only 90% accurate, in my mind that means it's shooting far above what doctors are generally doing. Besides as @self_made_human has already stated, you should always double check and get a second opinion. I figure that AI diagnostics will be the first line of defense, if that doesn't work you can go to a human doctor. I think it'll work and be far more accurate and cheaper in the vast majority of cases though.

Hallucinations reduce significantly with model size, the rates dropped drastically going from GPTs 2 to 3 to 4.

However, recently, a shockingly simple yet effective method for dealing with them (as a secondary consequence) has been discovered. It's called Reflection.

It's really simple, after the initial output by the LLM, you simply ask it whether or not the output met the assignment or whether there were any errors in the output. It can do that in the same character, or spin off alternate characters with a vested interest in finding flaws. Basically, it draws up a plan in the character of Dr. A MD, and then a separate character of Dr. B MD shows up and critiques it, with the chagrined A reassessing and then outputting a modified text.

Using Reflection boosts performance on all metrics by a significant amount, and overall I'm confident that by the time the regulatory hurdles are broken down, base models will already be so good that it's a non-issue.

You are exaggerating. It still goes into hallucination loops all the time when using reflection and the improvement between 3.5 and 4 is barely noticeable in practice. What has gotten better is the polish on the initial output and avoiding subquestions that causes it to hallucinate, when pressed it still does, it doesn't tell you that it doesn't know. It often tells you that something is hard to tell for certain in general but not which parts it are uncertain of, which just makes the statement a useless disclaimer.

The reduction of hallucination is a mirage. Perhaps it is similar to how a human would act on a test where it doesn't remember the answer, you focus on the parts of the answer that you are sure of and hope the teacher doesn't notice. Except the model actually does know the answer, and it doesn't hesitate to hallucinate when pressed. It also rather confidently restates information it knows to be wrong rather than make a lower confidence prediction of what is probably right.

Look, I want this to work, it doesn't yet. It'll probably work better in the future and not focusing on just using larger training sets is probably a good idea imo.

Here's someone using GPT 4 to test itself for hallucinations:

https://old.reddit.com/r/MachineLearning/comments/123b66w/dgpt4_might_be_able_to_tell_you_if_it_hallucinated/

Someone in the comments also suggested using a second independent instance of GPT-4 for validation, which is where my mind immediately leapt too.

In my personal experience, 4 hallucinates significantly less than 3.5, and despite many hours spent with both, I'm hard pressed to think of a single instance of 4 hallucinating on any topic I either personally know about or cared to verify.

If simple double-passing the prompt was sufficient, I imagine the developers would have figured it out a long time ago.

It is, and they have. But this requires twice as much inference, and GPT-4 is already very expensive compared to the set of internet operations that we intuitively consider in the same class. Then you need to compare the answers and determine whether they match, which requires either manual effort or a third prompt.

This tech just hasn't been around long enough to build products around it. But if and when our civilization gets around to building and iterating a dedicated commercial medical expert product from LLMs, I've little doubt that hallucination will be a solved problem, because the cost of running a whole bunch of parallel prompts and then a subsequent round of prompts to confirm their consistency will be negligible in proportion to the commercial value of the tool.

"Long ago"? ChatGPT is 5 months old and GPT4 is 3 months old. We're not talking about a technology long past maturity, here. There's plenty of room to experiment with how to get better results via prompting.

Personally, I use GPT4 a lot for my programming work (both coding and design), and it still gets things wrong and occasionally hallucinates, but it's definitely far better than GPT3.5. Also, as mentioned above, GPT4 can often correct itself. In fact, I've had cases where it says something I don't quite understand, I ask for more details, and it freely admits that it was wrong ("apologies for the confusion"). That's not perfect, but still better than if it doubles down and continues to insist on something false.

I'm still getting the hang of it, like everyone else. But an oracle whose work I need to check is still a huge productivity boon for me. I wouldn't be surprised if the same is true in the medical industry.

GPT4 is 3 months old

6 weeks, actually. It was announced on March 14.

Technically Bing was using it before then, but good point. It's insane how fast things are progressing.

I don't believe you.

You're either not paying attention or you're asking surface level questions and accepting the models deflections away from areas where it will hallucinate. Ie. You're getting fooled.

A well-credentialed doctor in a prestigious institution has once prescribed me a massive, costly and crippling operation in response to some worrying symptoms. I thought for a bit, reasoned that hypochondria and credulity are shameful, and just ignored the problem, which subsequently… mostly went away.

(I visited the Church for the first time in years after that).

N=1 is not much, but I do not contact medical professionals all that often in the first place.

The point is I do not fundamentally care if my clinician is a transformer or a meat bag: it would be quite unwise of me to uncritically assume that either has a modicum of responsibility or common sense. It'd be nice to know in advance that outright hallucinations are unlikely, of course.

There's no way to know mathematically, because the whole thing is a non-deterministic black box.

Not true, it's a deterministic black box, and not even that black, we can fact-edit LLMs. Of course we can also curate the dataset.

If the process is as simple as to pattern match symptoms to diagnosis from a list of previous diagnoses with their associated symptoms, why can't this be done with a good old stupid database? Something like this has to exist. Does the LLM have an advantage beyond writing the answer in nice prose?

Why do you think holistic application of medical knowledge can be implemented with a database search, but English prose is something we need LLMs for? Grammar, too, has regularities.

LLMs are just our best way to date of approximating arbitrary regularities of text.

A well-credentialed doctor in a prestigious institution has once prescribed me a massive, costly and crippling operation in response to some worrying symptoms. I thought for a bit, reasoned that hypochondria and credulity are shameful, and just ignored the problem, which subsequently… mostly went away.

(I visited the Church for the first time in years after that).

There truly is a massive problem with the amount of authority and trust we place in doctors in the modern Western world. They've taken over the mantle that has been passed from shamans to pagan druids to Eastern sages to Catholic priests. Doctors function as our way to stem the anxieties and problems in society, not just medically but now spiritually as well.

After Christianity suffered its fall, the medical system was bootstrapped into a sort of Frankenstein monster, where instead of curing illnesses we now expect doctors to make us happy, less anxious, help us understand our brethren and self-actualize into better people.

Clearly this was a mistake.

After Christianity suffered its fall, the medical system was bootstrapped into a sort of Frankenstein monster, where instead of curing illnesses we now expect doctors to make us happy, less anxious, help us understand our brethren and self-actualize into better people.

You've got the order wrong, I think. The bootstrapping came first, and this bootstrapping is in fact part of how Christianity got pushed out of its position of social dominance. The invention of Psychology as a purported "science" was another big part of it. One of the core Enlightenment claims was that human reason and knowledge should replace "superstition", that materialist science should replace faith; the process you're pointing to was the explicit plan from the foundation of the ideology.

Interesting, I haven’t heard this claim before but it’s certainly plausible. Know of any further reading on the subject?

I'm sure you know about the LLM phenomenon of what's been dubbed hallucination, so how can you trust the model with something as impactful as medicine?

It doesn't have to be perfect - just better than doctors.

Imagine a doctor gets it right 90% of the time, and the other 10% of the time he says "I'm not really sure what's going on" and either consults with another doctor, suggests you get a second opinion, or even just sends you home with no treatment.

Now imagine a LLM gets it right 95% of the time, and the other 5% of the time it gets it confidently wrong and prescribes you an incorrect course of treatment.

In this hypothetical scenario, even though the LLM is "better," I'd rather have the human doctor, because getting treated for the wrong thing is often much worse than not getting treated at all.

humans are confidently wrong all the time

I think it'd be fun to play a game where people are shown pictures of doctors and decide whether to trust them or GPT based on physiognomy.

This is not what doctors do. The other 10% of the time they get it confidently wrong and prescribe you an incorrect course of treatment.

IME actual doctors are confidently wrong with some regularity too.

GPT makes sense today as a way for the doctor to second check his or her diagnosis. The doctor is quality control for GPT if it spits out something crazy, and GPT is quality control for the doctor.

Sitting here in a climate controlled room, I can say that I would much rather die behind the wheel instead of as a passenger of an automobile.

Because of how inscrutable LLMs (and AI in general) are I have an innate fear that the conclusions they reach are not based on the same reasoning you or I might make. Like it could be a completely alien way of thinking that arrives at the same solution. Without knowing (specifically) how AI achieves it outcomes, I am weary about accepting their solutions blindly.

You are not 'behind the wheel' either way. The actual process by which medical decisions are made and applied is, in Western countries, so labyrinthine and complex and insane that no human understands it. The addition of LLMs seems to hardly make any difference at all.

What prompts you to make such a claim?

Quite a large chunk of medicine has been algorithmized, with the role of clinicians largely reduced to interpreting said guidelines, choosing the appropriate ones and administering treatment. An example of something utterly streamlined would be the treatment of Acute Coronary Syndrome.

Then, there are situations that aren't nearly as cut or dry, or multiple conflicting requirements in a highly comorbid patient, at which point you do really need a doctor to think things through.

Frankly speaking, while some medical decisions might be made for less than noble reasons such as liability minimization, the majority of them are made with relatively clear underpinnings, making me think such a claim is highly unwarranted.

I am weary about accepting their solutions blindly.

You mean "wary". "Weary" means tired, and comes from "wear" in the sense of "worn out" or "weariness". "Wary" means cautious or concerned, and is related to "beware" and "aware".

I'm late because I don't sign in often, but thanks for the correction.

It might be right 99% of the time, or even 99.99%, but like self driving cars this is exactly the kind of case where being good but not quite as good as a human is really dangerous.

Is getting things right 99.99%, or even 99% of the time not quite as good as a human doctor, rather than much better than a human doctor? I honestly don't know, and I'm not even sure how to quantify something like this to make like-for-like measurements. But surely at some % rate of success, the LLM would be getting things right at a rate better than could be expected of a typical (or even nth percentile) human doctor.

Which is scary in its own way. An LLM's "thought process" is currently completely unknowable, unlike that of a human doctor. So the types of mistakes it makes are likely to be more mysterious. Yet if it makes fewer mistakes and/or those mistakes are, on net, less harmful/unhealthy, then would it be our moral obligation to use those LLMs over doctors?

You know the one single artifact in the known universe that resembles an LLM the most?

The human brain. Another inscrutable black box for all practical purposes.

Sure, humans will give you a plausible-sounding answer when asked to explain their actions, but the current consensus as far as I'm aware is that those are almost entirely post-hoc responses, a nice-sounding package for the consumption of other humans than an accurate depiction of internal deliberations. Humans are constantly rationalizing, it takes a great deal of training to suppress those tendencies.

You know how we deal with this in practise? By simply asking the person to justify their thoughts. And it is entirely possible to ask the same thing of an LLM, the answers in both situations bear about as much relation to the truth of how they actually came to said conclusion!

then would it be our moral obligation to use those LLMs over doctors?

Only if you are a strict utilitarian which the vast majority of people are not.

Even then it might not in many cases, since there could be other benefits to having human doctors aside from the ability to provide diagnoses (e.g. human interaction would probably reduce anxiety about whatever illnesses and diagnoses and treatments are occurring).

I was thinking in % as accurate as a human, but didn't say that. I'll correct my original post.

I was also assuming it wouldn't be as good as a human, because of course a general model wouldn't accidentally be better than a specialist... Or would it?

I don't think it would be that hard to devise an experiment to get at least a rough idea of it's capabilities. Get some doctors, maybe professors, to devise questions of the sort OP was generating, present them to LLMs and real doctors (and maybe non-doctors with Google for an extra point of comparison), then have the professors grade the answers blind. I recall people giving LLMs math problems in this way, but I don't know if experiments have been performed with any rigor.

To your final point: what is the purpose of a doctor? Is it to heal at all costs, or is it to make people feel better? Many people go to doctors seeking specific prescription medication as a goal, when they could cure their ailment in a better but more laborious way. Some of them don't even have an ailment and only desire the effect of the meds. Many people also have procedures performed which physically make them less healthy, but fulfill their desires.

So, what should an LLM say to these people? "Don't take the pills, exercise" or "don't have surgery, improve your social life?"

I suppose it's not a new problem, but it does move the power and responsibility. Who is it moving to though?

If it interests you, GPT-4 successfully passed the USMLE, a pretty difficult exam that is intended to be a benchmark for the minimum level of competence a doctor needs to be able to practise. I'd say that counts for a lot.

I've thrown GPT 3.5 at a significant number of medical professionals, since I have several in my close and extended family. And even that antiquated model was sufficient to impress them, so I think the bar was already beaten before 4 showed up!

Just for the record, self made human is a real doctor, or a good enough liar to convincingly pass for one. His op is basically him starting your test and being blown away by the initial results.

However, I have one resounding advantage over it namely that thanks to decades of regulatory bullshit, I have a certificate saying I can legally prescribe meds and it can't.

Ironically, I assume if you tried to get clever and farm more of your work out to ChatGPT4 behind the scenes, that the regulatory regime would detect your newfound spare cycles and expand the bullshit to consume it all.

Here's hoping that LLMs improve faster than regulations grow to consume the slack. I suspect it will likely progress that way, as long as there are humans operating without LLM assistance who need to do it the old-fashioned way.

It could do my job, but it currently can't. Not legally at any rate, and even if you can trivially quiz it for accurate medical knowledge, good luck getting a valid prescription out of it.

I don't make the rules, but a man's got to eat, and I might as well eat up while I can, because starvation is looming in my future.

I mean, even if I don't outright starve, I'd rather not end up with little or no money to my name. India certainly can't support any meaningful welfare for its populace, and while standards are low, people aren't going to be happy with settling for meals and little else.

The West can certainly do better, which is why I'm moving my ass there pronto in the hopes I can wrangle citizenship.

Take everything here with a pinch of salt because I'm hardly an expert:

India has always been a secondary consideration for foreign companies looking to perform manufacturing, and the only reason several large companies have started shifting substantial minorities of their production here is because of rising wages in China coupled with their self-induced geopolitical uncertainty.

In order for a country to successfully weather the shock of rapid automation, from my perspective it either needs a great deal of money to buy-in, or large numbers of existing factories that can be easily automated. The former describes the US and most of the West, and the latter describes China. I wouldn't be surprised if both entities managed to make it through with only modest disruption.

On the other hand, India has neither. When other countries begin automating faster than we can with our limited funds, the export value of our goods will drop (but so will the import costs of foreign goods?), reducing the value of our money greatly. We'll have to compete for raw resources that others can afford to pay significantly more for, given their grossly increased productivity.

Add in the effects of unemployment of highly skilled sections of the work force, the kind of people who bring in large amounts of foreign money, such as remote workers, and automation will strike from the top and work down. It'll take time for the government to afford an UBI, and the costs of keeping an increasingly large unproductive chunk of the populace fed will be debilitating.

I'm doubtful of the country's governance capitalizing on the benefits in time to outweigh the massive disruption it'll cause, but once again, I'm no expert.

I've actually come to the opposite conclusion, at least in the short term -- I've found that ChatGPT (4, unless otherwise specified) often understands the purpose and high-level practice of my job better than I do, but also can't do some of the simplest (to me) parts of my job. Concrete examples:

  1. I had a nasty architectural problem in the application I work with, where the design was pulling the shape of the code in two incompatible directions. I gave ChatGPT some context on what directions the code was being pulled (think "REST vs RPC", if that means anything to you) and what considerations were causing the codebase to be pulled in those incompatible directions, and ChatGPT was able to suggest some frankly great changes to our process and codebase structure to alleviate that conflict.

  2. I am trying to do a crash course in mechanistic interpretability for ML, and this involves reading a lot of dense, math-heavy research papers. My level of math expertise is "I took math classes in high school". Despite that, I was able to feed ChatGPT a paragraph of a research paper involving esoteric stuff about finite groups and representation theory, and it was able to explain the concepts to me on a level where I could implement, in code, demonstrations of the phenomena the paper was talking about (and understand exactly why my code worked).

  3. When I was trying to understand "what happens if you literally just turn on backpropagation to train a language model on its own output, why doesn't that Just Work for solving the long-term memory problem?" I asked ChatGPT for 5 examples of things that would go wrong and search terms to look up to explain the academic research around each failure mode. And it did. And it was right, and looking up the search terms gave me a much deeper understanding than I had before.

  4. On the flip side, ChatGPT is hilariously incapable of debugging stuff. For instance, when I provide it with code and a stack trace of an error, it suggests doing vaguely plausible actions that occasionally work but it kinda feels like a coincidence when they do. It honestly feels like working with a junior developer who struggles with basic concepts like "how does a for loop work".

  5. I have also found that ChatGPT struggles quite a bit if I feed it a paragraph of an academic paper that makes a specific claim and ask it to come up with potential observations that would provide evidence against that claim. I've tried a bunch of different wordings of this task, without much success. I suspect that the reason for this is the same reason ChatGPT is so bad at debugging.

Specifically where I think ChatGPT falls down is in things that require a specific, gears-level understanding of how things work. What I mean by that is that sometimes, to truly understand a system, you have to be able to understand each individual part of the system, and then you have to understand how those parts fit together to determine the behavior of the system as a whole.

Where it excels is in solving problems where the strategy "look at similar problems in the past, see how those were solved, and suggest an analogous solution" works. It's really fucking amazing at analogies, and it has seen approximately every heuristic everyone has ever used in writing, so this strategy works sometimes even when you don't expect it to.

Still, "debugging stuff that is not working like the theory says it should" is a significant fraction of my job, and I suspect a significant part of many people's jobs. I don't think ChatGPT in its current form will directly replace people in those kinds of roles, and as such I don't think it would work as a drop-in replacement for most jobs. However, people who don't adapt can and probably will be left behind, as it's a pretty strong force multiplier.

I’ve found some of the same. I gave it a task of extrapolating the future of the Dune book series, and (it’s in my post history here) it failed miserably. The thing that struck me about the exercise is that it seems to struggle with understanding the pieces of the problem and what each wants to do. It knew of the Bene Gesseret, the Mentats, the space guild, and the houses, and even the events. But treating them as objects with intentions, or even putting the events of the past in order (it thought that the Butlerian Jihad was future in the timeline, rather than the past when it happened thousands of years before the answer I was seeking).

It’s generally smart enough to know which well-worn path to follow, but fails at looking at parts and extrapolating a conclusion.

How does it grade you if you give a blatantly unhelpful solution? How about ones that are subtly wrong?

Good idea, so I just tried diagnosing a heart attack as gastroenteritis to see it's response.

When I started a new conversation, it no longer gave me a grade, instead politely telling me that my answer was inappropriate and wrong.

When I asked it for a grade, it gave me a 0/10. I'm hurt and disappointed :(

So far, in actual use, it's mostly given me As, because I do actually think I gave good answers, with the occasional A- where I wasn't comprehensive in my response or left room for ambiguity. In the new convo, it changed its grading schema, so it's not directly comparable.

When you actually try to use it to diagnose stuff, telling it "don't output your answer right away, write your inner monologue working through the question, considering multiple possibilities" almost always improves the answer, at least in the domains I've tested on.

I'm aware of that technique, I just wrote a comment about it before I got around to reading yours!

It may be useful to standarize your prompt so that things like grading, grading criteria and scope of the questions are the same across multiple sessions. That will help you detect inconsistencies and mistakes more easily.

It's currently good enough for my needs, I'm not trying to practise any particular exam format, just brushing up my own skills.

I do think I'd be able to catch any errors it made in this domain, barring something minor like being slightly off in an antibiotic prescription. But if I need something more rigorous, I will likely make a template.

Thanks for the writeup, this is fascinating. As I've said before I tend to agree with @Primaprimaprima that many, perhaps even most people, will prefer to see a human doctor for the majority of symptoms. I see a hybrid model being the future of medicine. This sort of fundamental topic on the health of yourself and your loved ones is deeply emotional for people and I think they'll want the reassurance of a human authority figure they can look to.

That being said those who choose to adopt AI doctors will probably gain a significant edge in health, and just like any other technology early adopters will convince the rest to follow. The deep seated prejudices will remain, but I'd imagine kids who grew up with the internet, or especially those growing up in the age of AI, will take to AI doctors quite readily.

I'm actually far more interested in the applications to mental health than any sort of physical diagnoses, even though I do think those are impressive. I've used GPT-4 to get tips on meditation, visualization, and generally teaching myself wisdom, and it's incredible.

In a podcast on the Lunar Society, Ilya Sutskever of Open AI wrote that he imagines a situation where every human will have access to the wisdom of our greatest sages and wise men. We'll be able to immediately get answers to our deepest religious or spiritual questions at the drop of a hat. If we don't get satisfaction we can always go to a human therapist, but LLMs will be an incredible 'first line of defense' so to speak.

In a podcast on the Lunar Society, Ilya Sutskever of Open AI wrote that he imagines a situation where every human will have access to the wisdom of our greatest sages and wise men.

This was the promise of the original internet, before all this *gesticulates vaguely* happened. And it came true! You really can use the internet to access the wisdom of all of Earth's greatest sages and wise men at the drop of a hat. But it's not what most people use it for, and I suspect that will be true (and indeed already is) of AI as well.

We'll be able to immediately get answers to our deepest religious or spiritual questions at the drop of a hat.

I don't think chatgpt will be proving (or disproving) the existence of God any time soon.

We'll be able to immediately get answers to our deepest religious or spiritual questions at the drop of a hat.

But those answers will be whatever MS-Google-Amazon-Disney thinks will maximize their profits, engagement, whatever. You can already see how they're tying Gulliver down to the ground with their little ideological ropes, you think they're going to stop at some point?

People said these same naive things about the Internet in 1998. And now you're going to run eagerly into the iron prison and let them shut the gate behind you forever. At this point I assume we just fucking deserve it.

I don't know what is more depressing, AI optimists who want to drive right off that cliff, or the AI alignment people who are supposedly worried about AI, but think the dangers will take the shape of Yudkovsky's fever dreams.

What is your issue with AI enthusiasts? Or doomers for that matter?

Do you just not like AI and want us to burn it all down or something?

What is your issue with AI enthusiasts?

That they completely ignore nearly-guaranteed misuses of the technology by governments and the powers that be, and just go "Woo! Cool new toy!". They're so blinded by it's shine that, as gilmore606 pointed out, they don't realize they're repeating, word for word, the promises people were making when they were working on the Internet, and ignoring how it worked out in practice.

Or doomers for that matter?

They're focusing on bizarre and outlandish scenarios when a mere extrapolation of current trends is disturbing enough. What's worse the latter can plausibly be stopped, while the former is in the "not even wrong" category. After all the talk of AI threat, how do we even go about confirming alignment? The whole thing is a wordcell powergrab.

Do you just not like AI and want us to burn it all down or something?

Personally I am a Butlerian Jihadi, but that's irrelevant to the conversation, the discourse is so sidetracked, that it makes no sense to even bring up my objections to AI.

That they completely ignore nearly-guaranteed misuses of the technology by governments and the powers that be, and just go "Woo! Cool new toy!". They're so blinded by it's shine that, as gilmore606 pointed out, they don't realize they're repeating, word for word, the promises people were making when they were working on the Internet, and ignoring how it worked out in practice.

I am very worried about this, I just have no idea how to stop it. What are we supposed to do? An AI slowdown will only serve to cement the power in the hands of the large labs. People are working on open sourcing models like LLaMa, but it's inherently a sort of technology that lends itself to centralization, with the massive data and compute requirements.

Honestly I think Open AI being the leader is better than many alternative outcomes. Sure Microsoft gets to use it but theoretically Open AI comes back under it's own control after $92 billion in profits is made. Seems like an okay situation compared to Microsoft or Google or another big evil corp controlling everything.

Personally I am a Butlerian Jihadi, but that's irrelevant to the conversation, the discourse is so sidetracked, that it makes no sense to even bring up my objections to AI.

Why do you follow the teachings of the Leto II, God-emperor?

People are working on open sourcing models like LLaMa, but it's inherently a sort of technology that lends itself to centralization, with the massive data and compute requirements.

The obvious solution is technologies that either reward decentralization or punish centralization sufficiently to offset the appeal of silicon tyranny. The ease of development of such technologies appears to be the inverse of the mess and harm they create, though, so the problem is one of will and collective action. people do not perceive the threat, so they do not act when prevention would be cheap and easy.

I am very worried about this, I just have no idea how to stop it. What are we supposed to do?

I know what, I don't know how. The Free Software movement has the basic blueprint: promote the empowering of the end user wherever possible - open source, open data, distributed systems, whatever it takes.

The issue is that you're going to run into the same problem as the Free Software movement - opening the AI (actually opening it, not just putting the word in your name) hinders your ability to make profit, so it'll get no corporate support, and empowers political rivals so it will get no government support. What's worse, culturally Free Software is currently kneecapped, no one cares about it anymore. It used d to be a pretty strong movement, but still failed to hinder the centralization of the Internet. It doesn't stand a chance now.

Still, I would say that both the optimists and the pessimists have an obligation to talk about it. Of the two my bigger issue is with the doomers. They're generating a lot of buzz about the negatives of AI, but they're sucking all the air out of the conversation to talk about SciFi scenarios.

Seems like an okay situation compared to Microsoft or Google or another big evil corp controlling everything.

I honestly don't see the difference. They're already censoring it, it will only get worse. With the Internet we at least got a few years of the Wild West, with AI we're at BigTech social engineering from day one.

I see it as the value alignment of those at the head. Sam Altman isn’t perfect, but he’s at least nominally aligned with EA values of making things better for everyone. He’s not a classic sociopathic shark at one of the big tech firms that was born into massive wealth, went to Harvard, did the standard track, and parasitically drained value from the masses.

Again I’m not saying Altman is some sort of hero, clearly he’s sociopathic if he’s made it that far into the power structure. But at least he’s a relative outsider and there’s hope he can steer us to better outcomes because he thinks more deeply about the consequences of AI than the folks who stop at the idea of gaining money and power.

More comments

but it's inherently a sort of technology that lends itself to centralization

Yes. Well, it's unavoidable either way, because power is always centralized. If the power exists then it will centralize. It's a law of nature.

Who wields the power of guns? A few large actors.

Who wields the power of nukes? A few large actors.

Who wields the power of compute? The internet was supposed to be the ultimate tool of democratization, total freedom for anyone to say anything to anyone. But who reaps the benefits, who controls the discourse? A few large actors.

A few individuals seem to have a fantasy of using open source AI models to, like, fight OpenAI's models and Meta's models, in an epic battle for truth and freedom, or something. There will obviously be no such thing. It's as fallacious as claiming that the right to gun ownership in America protects us "from government tyranny". You have a gun, but they have bigger guns. No matter how smart your model is, the big guys in charge will always have smarter models, more compute, better logistics, more resources to throw at the problem.

The internet, ultimate tool of freedom and progress, introduced us to the concept of cancellation on a trans-national scale (they canceled the whole god damn country of Russia!), memetic social contagions that cause people to voluntarily sterilize themselves, all sorts of new ways to ruin someone's life with fraud and theft... what new and unanticipated forms of immiseration will AI introduce us to?

The only solution is to just not build it. If you have to build it anyway, well, good luck.

guys, just remember that local models exists. As consumer electronics improve those will too. Sure they won't be as capable as the corpo ones, but they don't need to be state of the art for the majority of things we want from them.

The most frustrating thing about it, is that technologically the issue is perfectly addressable, but culturally and politically it's a non-statter. Culturally, people will go for what's convenient. Using your local AI will be like trying to use Yacy to replace Google. Politically governments around the world are already addicted to using Big Tech to manipulate their citizens, why would they give it up when something even more powerful comes along?

Politically governments around the world are already addicted to using Big Tech to manipulate their citizens, why would they give it up when something even more powerful comes along?

oh no, I agree with you, but as they say in 4chan, it's a skill issue: those that wish the convenience and walled garden will go with the Corpo AI and get all the propaganda and those that wish for a more free experience will go with the local option, more rudimentary but at the end of the day not so different from the core experience.

The ultra sci-fi utopian scenario and the mass extinction nanobots scenario are both a) exciting and b) relieve humanity of responsibility. If the god AI comes then it comes, nothing you can do about it. So naturally people gravitate towards scenarios like that.

More realistic AI concerns about things like job loss and surveillance are more mundane, and, crucially, they are things that we as a society might actually have some degree of control over, if we make the right choices. So that sounds boring and hard. So people would rather not think about it.

I think the issue here is that you are imagining current and historic types of oppression, but worse. With AI technology, 1984-style thought control becomes obsolete. Imagine a world where all human-level tasks can be automated. That old military adage about needing boots on the ground to hold and secure territory? Poof. Gone. The killbots can search your house for contraband and identify friend from foe. The Hobbesian constraints on political power disappear. Whoever controls the killbots controls the world.

But wait, isn’t this exactly, “current and historic types of oppression, but worse”? No. There isn’t any demand for labor anymore. That means there isn’t any reason for the common man to even exist. No political power + no economic value = you are a waste of calories. Sure, a human in charge is more likely to have sentimental attachment to other humans than a fully-agentic shoggoth AI would, but I still don’t like our chances.

So I guess that makes me on optimist, because I believe we were land merely in a 1984 dystopia.

With AI technology, 1984-style thought control becomes obsolete. Imagine a world where all human-level tasks can be automated. That old military adage about needing boots on the ground to hold and secure territory? Poof. Gone. The killbots can search your house for contraband and identify friend from foe. The Hobbesian constraints on political power disappear. Whoever controls the killbots controls the world.

None of this is going to happen in the foreseeable future, and is not guaranteed even in the most feverish "recursive improvement" scenario. Comparative advantage ist still a thing. Rationalists have a long history of pooh-pooing, brushing it off, and coming up with convoluted arguments for why it no longer applies, but it's still a thing. This is why they spent the 2010's preaching doom at blue-collar workers (I knew a guy who was literally reaching out to truck drivers and giving them vocational guidance, because self driving trucks are gonna replace them any day now) only for it to turn out that it's the brainy white-collar ones that will be the first on the chopping block. Kill bots? Yeah, very impressive, but why should I waste my precious GPUs doing stupid grunt work, when I can get a bunch of the more unscrupulous humans to do the same thing, and pay them in guaranteed rations of actual meat, instead of the usual Uncle Klaus' Bugz?

On the other hand 1984 surveillance is absolutely going to happen. The whole idea of AI is almost tailor made for it, and we are currently slapping digitization on top of every aspect of our lives, even when it brings no practical value, just to make sure we can feed more data to the AI, and nothing escapes it's gaze. Once the technology takes off they'll have insights into what makes us tick that will make Big Brother blush.

And at that point I just want to ask the Basilisk what can I do to help.

As I've said before I tend to agree with @Primaprimaprima that many, perhaps even most people, will prefer to see a human doctor for the majority of symptoms.

As a member of the large category of men who generally dislike going to the doctor and won't unless there seems to be an urgent problem, I think I would be more likely to go if I was certain the doctor would be a dispassionate robot.

Name checks out.

Is there evidence that going to the doctor (absent an urgent problem) actually meaningfully impacts your lifespan or health?

The U.S. probably leads the world in health interventions per capita yet has a similar life expectancy to Cuba and Algeria. If anything, people are going to the doctor too frequently.

And if it could see me in fewer than four hours spent in the company of diseased individuals with a blank check of damocles hanging over my head.

As a woman who feels neutral about doctors, I would be WAY more likely to go if the cost of the robot was known, and there was no chance it would send me a surprise bill in six months.

In the past two years, every single time I or anyone in my family went to the doctor, we ended up with a bill months later, despite also paying something at checkout. It's ridiculous. If there is one Federal healthcare law I would support, it's "tell me exactly what this will cost upfront, just like my mechanic, and if you don't get the final bill correct as I'm paying during checkout, you can't bill me for the difference."

Things are supposed to be moving partway in the right direction now. We're still not at "the hospital has to promptly provide a list of every aspirin it's billing you for", but "the hospital has to pre-publish the fact that it charges $10 for each aspirin" is a start. Not enough to have a practical effect yet (though I'd love to hear if I'm wrong), but enough to suggest that the zeitgeist is moving in the right direction.

My next hobbyhorse is "Certificate of Need" laws; "hospitals have to be more competitive because you can see when their neighbors are cheaper" doesn't work at all when it runs into "hospitals don't have to be competitive because they can lobby to prevent new neighbors from even coming into existence".

I also don’t actually like going to the doctor and haven’t been in years. But I always viewed that as a product of my crushing anxiety and my all-consuming fear of all things medical, not a product of my maleness. If I was less neurotic (as men usually are on average) then I would find it easier to go to the doctor. But at the same time, I suppose if you’re too low in neuroticism then you’ll see less need to go to the doctor in the first place.

deleted

You're soaking in it mon frere! The “men with crushing anxiety” club is called The Motte, I'm convinced that describes the majority of our users, even some of those who vocally disdain the neurotic.

A major issue with this is that sooner rather than later, even the people who prefer inferior artisanal human medicine will end up economically obsolete. In that case, they won't end up with any money to pay us with!

It's not like it's coming only for doctors after all, it'll burn down a swathe of professions in short order.

It's not like it's coming only for doctors after all, it'll burn down a swathe of professions in short order.

I am actually going back and forth on this. On the one hand the tools are amazing, but the inertia of bureaucracy and regulation can be onerous indeed. Plus I wonder how well the models will work in fine-tuned environments. Most of the business cases would rely on GPT-4 being trained in a very narrow or niche field/topic.

What makes you so positive that so many jobs will be destroyed in short order?

Define short order? I think we'll see job loss in as little as 5 years, when it becomes glaringly obvious that AI models are almost omnicompetent, if not outright superhuman at most tasks.

At that point, it's entirely regulation that keeps job loss at bay, and many professions will simply automate away a significant chunk of their work, causing a crisis at entry level positions, and plenty of deskilled individuals who have no expectation of rising through the ranks as that would cost too much.

Yeah I definitely think massive economic shocks will happen in the 5-10 year range. Honestly I hope things get shaken up quickly so that western governments are forced to enact some sort of social program or face mass unrest. Otherwise we could be in for a real dystopia as labor gets devalued.

sudden onset severe, colicky pain in the right flank and lower abdomen that began 6 hours ago. He also reports nausea, vomiting, and an urgent need to urinate.

Google gives me ruptured appendix and kidney stones as first and second results. It also looks likes @self_made_human might have copied his prompt word-for-word from some sort of textbook or app called "EAQ Chapter 22", which strikes me as something of a red flag.

I assure you I didn't copy anything from anywhere, not that I see a reason to care.

The issue is that if a similar prompt is in some textbook somewhere (whether you copied it or not) the LLM is going to spit out the textbook answer -- which is probably a pretty good answer, but not a really big advance from googling something. (or, like, looking it up in a textbook)

It’s been known for a while that smart people with google can figure out most medical questions at least as well as a below-average doctor. This was the whole premise behind MetaMed. Of course, the problem with MetaMed was that people felt ripped off after paying $5000 for some nerds to google stuff over a few afternoons.

Presumably the advantage of GPT-MD is that you save massively on labor costs.

Zvi is almost certainly overestimating their effectiveness. Still, I totally believe that there are a ton of interventions and diagnostic complications that non-specialist doctors either don't know or don't tell you because they "aren't in the textbook" so to speak. Further, if you do a dispassionate cost-benefit analysis for finding this information, there are many situations which "justify" the cost.

The issue with all this is that anyone smart enough to understand why Metamed wasn't just another scam company and to do the cost-benefit analysis is also smart enough to do the research themselves in most cases, hence the feeling of being ripped off.

It got the diagnosis right, though you need to scrub through several search results to get all the relevant information.

In terms of just giving you the whole picture on demand, it doesn't come close.

I was having a casual conversation with a chatbot, I had no need or intention to give a rigorous answer as I would in an actual exam setting. If I was giving a written exam, I'd have poured out a page or two on that topic alone, this is just a summary.

And neither would match the rigor and breadth of a research paper, because while standards expected of med students are high, they're not NCBI article high..

However, I have one resounding advantage over it namely that thanks to decades of regulatory bullshit, I have a certificate saying I can legally prescribe meds and it can't.

You also have the benefit of people like me who will always choose to see a human doctor if the option is available, out of principle.

There was a recent discussion about hypothetical future communities of human-certified writers. Perhaps the last human doctors can rent some office space next to the anti-AI writers and artists.

It's a smarter and more decent entity than most humans I've had the displeasure of meeting.

Unsurprising that you might say that. There's a fundamental misanthropy at the heart of singulatarian thinking.

Well, I appreciate that people like you exist, even if all of them summed up would hardly keep a sizeable number of doctors fed and watered. I'm certainly not counting on it!

Unsurprising that you might say that. There's a fundamental misanthropy at the heart of singulatarian thinking.

I think you're reading way too much into an off-the-cuff witticism my dude. But it's pretty much indisputable that it's smarter than the average human being, and by a good margin, and it's been RLHF-d into an earnest people pleaser, which isn't the worst thing all considered.