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.
Jump in the discussion.
No email address required.
Notes -
Great Another AI Post.
On the front page of Meddit today is a post that basically goes "AI convinced my patient that eating toilet paper helps with constipation." This to my experience appears to be the norm.
In Medicine I've found AI to be of limited utility, because mistakes are not allowed and the required error checking usually nullifies productivity gains.
In Programming it seems to be hot shit because error checking is part of the job and required either way.
Other fields things get more complicated and I have less experience. But the usual people seem excited and the usual people seem frustrated.
So I want to know, for those of you who use AI for work outside of programming (and not to simplify tasks for hobbies/personal life) what are you actually using it for, how is it helping your discipline? I am endlessly fascinated by the immense gap between people who love it and hate it.
Also, for those of you who know a bit more about LLMs, the lack of true error correction seems to be a fundamental problem, is their any hope of this actually being addressed without completely changing away from LLMs?
Chatbots are trained to be obsequious to the user. Even when they push back, there's probably a way you can trick it into affirming whatever insane thing is in your head at the moment. This characteristic is what makes them tools and not oracles. They amplify the user's own quirks and inherit their goals. I try to use oppositional or oblique framings to questions when it's something more complicated than a fact check that I'm querying about. I hope that masking what I'm really after will help it be more objective in its answer, but I can't know for certain if it helps.
I tend to get relatively poor or confrontational results when I ask it questions like that; the AI tends to assume the user is adversarial, and not actually seeking to understand unless pushed back against verbally.
Reading between the lines of analysis remains an effective way to figure out when the bot is wrong or seems to not want to say something, which was quite surprising to me. But then again, how surprised should I truly be, given it only truly "knows" about what humans write down rather than any private or discussed analysis?
More options
Context Copy link
More options
Context Copy link
I find that using AI as a non-coder requires a fundamental change in how you think about things. Imagine you have a generally competent software engineer as your personal servant, and he commands a legion of data-entry interns: what can you now do that you previously couldn't do? Talking about work would be giving away the secret sauce, so I'll give another example. I write poetry, and I have many poet friends. I had AI spin up a database of poetry publishing outlets - journals, magazines, etc. - and create a bot that can a) scrape any given url and, if it publishes poetry, get its info, submission windows, required formatting, etc., b) tell you which outlets are open for submission, which require fees, which pay you, etc., c) answer natural-language questions like "I've written a poem about my faith, what Catholic outlets are currently accepting submissions?". From there, automating formatting your submission and sending it to the journals would be pretty easy, but I chose not to do that out of respect for the editors. There are probably a surprising number of things in your life that can be made easier with a simple personalized CRUD app, but almost nobody is grokking that yet (to be fair, I've spent an inordinate amount of time around software engineers for a wordcel, so I have a lot more pre-loaded intuitions about architecture and planning in this space).
This is a great way of framing it. The AI is even more helpful than this because it's also an extremely useful encyclopedia. I've been using it very effectively to write me relatively simple apps and programs that marginally improve my quality-of-life but which my amateur ass doesn't have the time or skill to write. For example, an ad-free personally catered music-player that I now use daily and is an actual joy to use. I have also been using it extensively to identify trees and plants which has been both very educational, interesting and useful. I, for one, have been impressed. 2020 could never.
As for my work, I don't use or interact with AI at all and hopefully it stays that way until I retire but it probably won't. I'm a truck driver.
More options
Context Copy link
More options
Context Copy link
I work in aged care.
The only significant use I have seen for it at work thus far is listening to and transcribing conversations in meetings, like an automated secretary.
Management also set up a policy bot, so you can ask the bot about the existing policies on such-and-such procedure, and it will search through the intranet and tell you, but thus far it does not seem to be widely used.
For the most part I think nobody knows what it is. All the doctors and RNs I talk to make their own judgements, and find the idea of outsourcing patient welfare decisions to a computer terrifying. Personal care workers and lifestyle workers have no use for it. Residents mostly have no idea what it is. I appear to have been identified by a few co-workers as relatively tech-savvy (you help people log in or troubleshoot a computer problem a few times and suddenly you have a reputation...), so there have been a few cases where I have reluctantly explained to people what ChatGPT is and how you use it. In practice I try not to encourage its use, though.
The most visible use of LLMs or bot-generated content I see at work, as I've commented before, is via YouTube slop. We have large televisions that can access YouTube and sometimes use them to play music, classic movies, church services, the news, or anything else residents might be interested in. Search results are increasingly polluted by AI-gen content, and sometimes YouTube automatically rolls over to them. This is a regular if minor frustration for care and lifestyle workers looking for good shows or playlists.
I probably can't generalise from my field, but if I had to guess, I'd say that for most jobs 'on the ground', which require either interacting with people or interacting with real, physical things, AI/LLMs/bots remain a solution looking for a problem. I know my case that the site manager is quite technophilic and is intrigued by the potential for AI, but thus far the use cases just don't exist.
More options
Context Copy link
Handling contract negotiations with employers. It's far faster and cheaper than lawyers, more straightforward with you, and can draft replacements for contract language you don't like rather than just telling your employer you don't like it and getting back something substantively similar.
More options
Context Copy link
I use it for some regulatory stuff, like filling out forms based on some guideline document and rough notes that i have. You have to iterate a bit and make it ask you questions, but it still saves a fair bit of time.
More options
Context Copy link
I am not sure that if rigorously tested on frontier models which are adequately tuned, that the error rate would be much lower than that of physicians. The difference between medicine and programming is that medicine is a guild which a strong lobby that makes it very hard to innovate. While programmers literally give away their work through things like open source. OpenAI has tried to probe this space (and claims that even going back to GPT-4 their models could beat physicians in some settings, look up "HealthBench") but other companies have not done so yet. Quite unfortunate, given that healthcare spending is about 10-20% of GDP in developed countries.
That Reddit post has literally been made by people whose salary partly depends on the belief that these tools are bad and error-prone, so it can be filed away under "vibes".
Apparently the biggest recent adopter has been law.
"True error correction" doesn't exist for human physicians either. Which is why you pay malpractice insurance.
If AI was able to write prescriptions for me, I would choose Astra over 95% of doctors.
If you really believe this you should consider your biases. A response to a post noting that AI is telling patients to eat TP should not be "more of this please!"
Especially since the LLMs are well known to flatter the biases of their users, which is something doctors are changed not to do due to the negative outcomes associated with that.
The toilet paper thing is the cheapest of the cheap google models. Is it good? No. Is it representative of frontier AI? Also no.
Doctors do that all the time, especially if there are adverse financial incentives at play. The US is going through a multi-decade crisis because of opioid overprescription and people shop around for ADHD stimulants.
I am not saying that LLMs can 1:1 replace physicians. They can't. What they can (probably) replace is the part where if you have a good idea of the symptoms you'd do tests, integrate information, come up with a diagnosis, order further tests and eventually prescribe medicine.
Anthropic and OpenAI definitely think they can replace that part. In an ideal world, I, as an informed consumer, would get the choice to choose between the perhaps inferior AI medicine that costs dollars or pennies or the 10-20% of GDP current medical system.
I hope you can see the difference between a structural problem that consists of some bad apples and institutional incentives, and a process that delivers glazing 100% of the time unless you take significant active steps to avoid it.
More options
Context Copy link
More options
Context Copy link
For some reason people tend to view AI models as a big homogeneous blob. "AI is telling patients to eat TP"... It was a specific model of a certain generation of models with a specific system prompt (with or without user modifications).
I wouldn't be surprised if it were GPT-4o that recommended eating TP. I would be very surprised (astonished, either) if it were GPT-6 Astra.
Research on models' behavior is significantly lagging behind the progress. For example, when I was looking for research on self-correcting behaviors in LLMs today, the latest models they use are a year old models.
A year ago Terence Tao likened a model he used to a useful if unreliable undergrad. Today he worries about the future of human mathematicians.
More options
Context Copy link
More options
Context Copy link
I think whether the doctors like it or not, their patients (maybe not the geriatric ones, but often they have younger family helping) will be asking AI questions. I will admit to using it for "does this merit going to the doctor?" Sorts of questions, and for "recommend some PT exercises for this nagging issue" (seems a blurry line between sports medicine and personal trainer). I trust the doctors I see, but copays are high enough that I don't go for just any sniffle, and I don't want to pester them with questions outside of appointments.
From using it in other fields, I'd expect it to be good at transcribing verbal notes (I hear doctors hate charting, maybe it could do that interactively with the patient?), and at pulling obscure facts from the training data. I know pharmacists are currently supposed to be checking drug interactions, but adding an extra layer of verification there seems an easy addition. I'd expect it to be good at triage, but I'm not sure exactly what that would look like.
I would not expect mass layoffs in medicine as a result. Demand for medicine is very elastic, and if appointments were cheaper, we'd probably just have more of them. The limiting factor seems to be the 300 seconds you get with a physician, and how many hours they have in a day.
The problem is that patients do this now and it's a disaster and they don't realize it. Everyone who does outpatient medicine is flooded with instances of patients being sure about something and it being wildly off base, useless, and time consuming because the patient has been convinced of some nonsense that we have to walk back. I'm sure some fraction of people are saving themselves time or getting useful information but it's not well represented in what we actually see.
Part of this is because people don't realize that a lot of medical work involves things like dealing with an anxious patient who repeatedly asks the AI until they get the answer they want, assholes, and things like rationing care (which the AI has no facility with).
Believe it or not a lot of visits (especially to the ED) are pointless and we'd love it if people were successfully using AI for triage.
They aren't.
Yeeees, you're probably not wrong, but if you reverse the viewpoint to get the patient's experience for these examples you get:
The doctor repeatedly dismisses your concerns.
The doctor dislikes you.
The doctor agrees you would benefit from something but refuses to give it to you anyway
And when you put it like that you can kind of see why so many people use LLMs. I think this is one of those 'the only way out is through' situations where it would be far more useful to improve the availability of testing while also training the AI to fluidly guide people through quick home diagnostics, so they can get solid evidence. The alternative seems to be sticking with the current situation of, "pay the doctor $500 dollars for twenty minutes of thought and hope desperately he/she doesn't think you're a whiny crybaby or too convinced of your story'.
You have to keep in mind that a relatively small fraction of the population (often with relevant diagnosis such as personality disorders, anxiety, illness anxiety disorder) are super utilizers of services, inappropriate users of services (such as misuse of mychart physician messaging) and are disproportionately the people complaining online (as well as misusing AI tools).
You see another version of this in the anti-psychiatry movement where people who can't recognize they have an illness because of the illness (ex: schizophrenia) are complaining about being forced into the hospital and take medication when that is the correct thing to do.
As for rationing the simplest example of this is as a good example of over utilization that many people fall into - suffering sometimes is part of life. If you have a viral illness you need to wait it out and take care of yourself most of the time, the doctor will not be able to do anything and rarely is it worth confirming which URI it is. Antibiotics will not do anything useful.
This does not stop people from emphatically making demands.
I can believe that. I guess what concerns me is that institutionally the same machinery has to apply to me and to Nutty Alice, and there seems to be a strong presumption that the role of a doctor is to constrain the patient and that a patient researching their own condition is essentially misbehaving. From there it's only a hop, skip and a jump to "people shouldn't be able to get medical advice from AI, and doctors should ignore it when they do".
Whereas instead we could work hard to make AI better at obtaining clinical histories and to make feedback loops with reality via diagnostics faster and cheaper. Maybe cutting Nutty Alice off if she exceeds some kind of monthly limit.
More broadly, and with what I hope is appropriate tact, I guess I see this as another skirmish in the long-running war over whether doctors should be respected servants or priests. It's simultaneously true that doctors have genuine expertise and it's often sensible "not to [flatter their patients' biases] due to the negative outcomes associated with that", while also being true that from the patient perspective doctors' anti-sycophancy training seems to be dialled so high that the relationship feels antagonistic. The default response to patients with their own opinions seems to be "no" rather than "yes, but".
To put it another way,
is true but doesn't seem to me to be striking quite the right attitude for a relationship that I conceive of as client and service provider.
A lawyer who tells you, "mate, it's not worth suing, even if you win the damages won't be worth it" or a plumber who tells you, "plastic pipes sweat a bit in the cold, it doesn't mean there's an issue" is worth their weight in gold. A lawyer who tells you, "all landlords steal your deposit, come back when you've got real problems" or a plumber who says, "pipes do that, call me when there's an actual leak"... not so much.
With respect to viral illnesses, I might have a better opinion of anti-virals than the doctor does, or I might suspect it's actually something else and want a test to check even if the doctor thinks I'm being silly, or I might just be really really interested in URI strains. In that scenario I think it's more productive to give me access to cheap good advice and make my wishes easier to fulfil than to litigate whether or not I'm permitted to have them. If Nutty Alice needs special accommodation I'd rather we tried to lay down some rules for her specifically than constrain everyone.
Caveats: some issues genuinely need gatekeeping, such as coordination problems like antibiotics or medical interventions with genuine health risks. I'm talking about a change in the relationship during diagnosis and exploratory treatment, not opening up the medical cabinet to all comers.
Caveats the second: I have experience with the British private system but not the American one (which might be quite different).
Intuition pump: debugging is usually the slowest and most annoying part of fixing any problem. Making each debugging attempt cost hundreds of dollars and take 20 minutes max and have a 25% chance of being refused outright is very inefficient.
This is a good comment - but we are down some staff at work and I know by the time I swim out of this it will be hard to motivate myself to circle back, I didn't want to leave you totally hanging though.
Although one specific thing, a common problem is people having a viral illness and demanding antibiotics (which at best would do nothing). Anti-virals generally being useless is a separate kettle of fish.
Appreciate the thought! Good luck at work.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Yeah, I try to be very careful to avoid confirmation bias, and only for relatively minor things. I have to do that fairly frequently with LLMs in other contexts, so it's not an unfamiliar skill to me. A couple extra strength exercises (which make sense to me given my knowledge of the anatomy) isn't likely to hurt anything.
But I doubt most patients are avoiding it. Frankly, I'm probably not completely either.
Yeah that's fair, we definitely have many posters here who would use a tool like that and be sensible about.
We also have even more people who would do similar stuff and be intensely foolish, and have no awareness of that fact.
Unfortunately the productivity gains from the former are probably outweighed by the damages from the latter.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Haven't you been looking up symptoms and successfully diagnosing yourself via the internet for at least a decade, though? I'm not sure what the AI is really adding on this front except taking WebMD out as the middleman. It's still the case that I figure out what is wrong with me myself, then go pay some doctor to write the prescription that I know I need and that they agree I need. Paying a doctor to figure out what is wrong with me, mostly because my main question is usually whether I want to spend my time/money on a doctor, has seemed crazy for a long time.
My experience is that Dr Google isn't great at differential diagnosis, and absolutely sucks at "What are the potential diagnoses I should be thinking about - either because they are likely or because they are moderately likely but serious - as a Bayesian inference from this set of symptoms?"
More options
Context Copy link
In my experience, the AI can go an extra step beyond what WebMD could previously. Sure, there was a list of possible causes for your symptoms, but now it'll ask you followup questions to narrow it down, or suggest simple diagnostics ("does it hurt when you...").
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
How are you, or the public at large, supposed to use Astra (or any other sufficiently powerful model), which must be used in the cloud unless you have a 50k dollar setup at home? Do you trust these companies with very sensitive personal info?
(Though, having asked that, it comes to mind that we don't actually have privacy in regular doctor's officers either, due to phones and other devices recording all audio at all times)
More options
Context Copy link
Make that 99% of doctors and I'd agree. Smash the cartels!
More options
Context Copy link
More options
Context Copy link
Well I've been using it for proofreading/editing tasks at work, creating scripts for certain tedious tasks. Not perfect but pretty helpful, better than a human especially in cost-efficiency and speed. AIs can spot things, tiny grammatical issues with quotes or repetition that humans miss. Or for Excel tasks. Or deep research for that matter, compares well to most journalism.
I don't think error correction is a big problem for AIs, at least in non-critical domains. They will make them smarter and more hardworking and cheaper, so we can just run it over and over again.
Also Blender, it's pretty decent there, albeit with inferior visual sense and requiring iteration for good results.
(Where I say AIs, I mean frontier models, not free/cheap models).
They are good at this sort of thing, but I find it hilarious that I also have trouble with models flummoxing tool calls (a JSON payload) with unterminated strings and braces. I had it try to fetch
about:blankrecently.This isn't free-tier models, this is the large, recent open weight models. But most of the time it does it correctly.
More options
Context Copy link
More options
Context Copy link
This is going to be similar to self-driving cars where it's impossible to reduce the failure rate to zero, but the robots will outperfor humans by so much in aggregate that we'd be foolish not to use them.
Doctors are in general one of the highest IQ professions and are often quite practical as well, working as close to the public as they do. But I think most of us have had the experience of dealing with doctors who are midwits, lazy, or checked out. I have myself received extremely bad medical advice from doctors on at least 2 occassions.
If I had to choose between Astra and a replacement level telemedicine doctor who couldn't write prescriptions or examine me, I would choose Astra every time.
In 15 minutes, I was able to use O-3 (an ancient model at this point) to correctly diagnose my father's medical condition which mystified multiple doctors. It really was pretty obvious in the end, but that's where the laziness comes in (compassion fatigue more generously).
For the things I use AI for, hallucinations have become rarer and raret. I almost never encounter them anymore. I think we'll get there for medicine at some point too, but I do acknowledge that AI's are really bad at physical and spacial reasoning at this juncture.
More options
Context Copy link
I wrote a basic MCP so it can pull data from my iotawatt (electricity usage); and then look at historical usage on a circuit that powers the electric baseboard heaters in my bonus room to see the actual BTU used to heat the space. Then based on those numbers identify ducted and ductless minisplit candidates to heat the space.
More options
Context Copy link
I entered a bunch of medical data to Astra and asked it to recommend some protocols for me. It told me to stop taking my B1 vitamin (it helps with my neuropathy but ChatGPT says there’s no proof it does that), ask my doctor to reduce my thyroid prescription (!), and make sure I eat 100g of protein a day. Yeah it’s really not there yet.
On the other hand I regularly consult ChatGPT while cooking. It’s very handy at proposing substitutions or modifying a recipe. (“Hi ChatGPT please give me a recipe like pudding choumer but with these two ripe bananas I have lying around”. “I dont have milk how about yogurt.”)
I would say AI replaces Google searches and Wikipedia half the time now. It’s very good at complex but discrete questions I couldn’t otherwise get at without a lot of searching. (“Hi ChatGPT I just read this article about pilot pay in the WSJ. Give me a sense of the amount of jobs in maintaining the airline industry, grouped by pay, first in America, then comparing to Canada Europe and Asia.”)
AI is very good for language practice and basically eliminates the need for me to have a workbook. I tell it what grammatical concepts I’m working on, it designs infinite exercises, I don’t have to flip to the back of the book, I can ask for clarification, I can ask for harder or simpler longer or shorter.
More options
Context Copy link
Out of curiosity, pulled up the post:
https://old.reddit.com/r/medicine/comments/1wmbn87/a_pt_showed_me_proof_that_eating_toilet_paper/
"Google AI" so, presumably, the search model. Even there, I'm unable to reproduce. The closest I can get is by searching "why does eating toilet paper cause you to poop more," but even there, it includes paragraphs and paragraphs of verbiage saying it's unsafe and dangerous.
I think that they should probably add some classifier that categorizes queries into regular users and idiots, and then for the idiots generate a text that cannot possibly be selectively taken out of context or misinterpreted.
Mostly for programming, but at this point there's a bunch of other, mostly administrative things it's useful for. From the past week:
Identifying work I've done, for reporting upwards
Juggling bugs/issues
Creating the skeleton of design docs (with the caveat I have it create the outline, headings, text plugs, and tables/graphs/diagrams, without writing the actual text content, which remains terrible)
Extracting action items from emails, meeting notes, chat messages
Random formatting things for spreadsheets
Creating recipient lists for emails and lists of attendees for meetings (along with figuring out scheduling for the meeting)
Researching existing documentation and design docs to describe how a system works today, and verifying the actual system implementation works as described
Crawling the calendars of coworkers, to see if there are meetings I should be invited to but haven't been.
Gantt charts
Skimming that link:
More options
Context Copy link
More options
Context Copy link
In programming, you can just check if something works before deploying, or roll back to a previous version. In medicine, that involves dead bodies.
More options
Context Copy link
LLMs will always be probabilistic, by their nature. They will get smarter, and thus make errors less frequently, and there are other techniques like Mixture of Experts, and various schemes which are basically "AIs checking other AIs," but no, they will never be 100% error-free. The question is whether we will accept the risk when, say, they make errors no more often than human doctors do? Self-driving cars already on most benchmarks drive better than most humans, but one fatality as a result of a self-driving car sets the industry back months as they have to make sure the car won't make that particular mistake again.
You have to have someone to blame / sue when it goes wrong. You can't blame the computer.
More options
Context Copy link
While it's technically true that you can't hit 100%, the error rate can (in theory) be made exponentially small. If you just run a query N times, and (this is important) each query has an independent error rate that's <50%, then taking the consensus result will get you extremely high accuracy, very very quickly approaching 100% even for decently-sized N. So the practical answer is that, yes, we should be able to make the error rate vanish, as long as we're willing to pay a small multiplier on cost. Lots of serious users of LLMs already do this, of course; it's not a new idea. But your average Google search probably doesn't.
But this doesn't apply if the error rate isn't independent, of course. If there's a class of queries for which LLMs will inevitably confidently state the same incorrect answer, then we won't be able to improve on that. (Humans have classes of queries like this too.) It's a bit more difficult to predict whether this will always be an issue, especially if we mix different models together.
Ah yes the monte carlo approach.
More options
Context Copy link
More options
Context Copy link
This was my first thought, has this gone anywhere?
It's being used a lot in software for agentic workflows. I don't know of any rigorously benchmarked use cases that shows them reducing error rates, but there is lots of anecdotal evidence. (Some people swear you get more rigorous fact and error-checking if you tell Claude "This was generated by ChatGPT; check its work" or vice versa, but I haven't seen that myself.) The problem of course is that you're using a probabilistic machine to judge another probabilistic machine, so are you reducing error rates or just watching them converge on shared biases?
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I find ai can be pretty good with medical advice, but you need to get it in the right zone with the right buzzwords like "clinical guidance", "case reports", "differential diagnosis", "treatment indicated", and "patient is presenting to the emergeny room with".
If you just start off with like "my tummy hurts" then you're gonna get really bad results, because it gets the ai in the mode of reciting pop health advice.
The problem is the error rate (for professional use) and misuse with accompanying over certainty (for amateur use).
That's been an issue with doctor google for forever though- webmd says it's cancer!
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link