Advertisement - Members don't see this ad
-
CycleTrack and SDN are teaming up to make medical school admissions more transparent and accessible! Read the announcement to learn how we’re supporting the future of CycleTrack’s free application-tracking tools .
Number 2 Most Resistant Job To AI
Started by Groove
Get help with your application
Use all the free resources available to you from SDN: articles, guides, expert advising, forums discussions, and school research.
What’s weird is the article reads like it was written by Chat GPT. I noticed it in the “Physician Assistant” section (it was so generic, didn’t say much of anything, and had those telltale dashes).
Tangentially related, but since you posted in Groove: holy hell is Claude a different beast than gemini. I'd used gemini for a bunch of random stuff that I mentioned before and in the end it got me started but I basically wrote everything myself.
I just tried Claude for a little pet project I'm working on and it spit out a working template with only maybe an hour of debugging on my end. And I mean, a slick looking app that was more or less functional from minute one.
I just tried Claude for a little pet project I'm working on and it spit out a working template with only maybe an hour of debugging on my end. And I mean, a slick looking app that was more or less functional from minute one.
Advertisement - Members don't see this ad
Sir, now we're on the "oh yes it can" hit list.
And then I got to the PA part and realized it selected all PAs but only our medical specialty?
Weird that a CRNA is #1 but an anesthesiologist is #11.
D
deleted1223140
I think AI should really be focused on radiology. Read times are way too long.
Yea, not too long ago the state of things was "this is pretty good at getting some code down and I can troubleshoot the weird bits" and it has become "oh, I guess this manifests my wishes, now."
Forbes lol.
Can tell it's a psyop when it lists CRNAs as #1 and Anesthesiologist at #11.
Can tell it's a psyop when it lists CRNAs as #1 and Anesthesiologist at #11.
I know dude...right?! I'm almost convinced that Anthropic has almost solved 100% of coding. I've got codebases on projects that would have taken me years to do on my own that I was able to do in days with Claude Opus 4.6 No other LLM comes close in regards to what I can do in Cowork or Code. It's like a dream generator. All you have to do is think of something you want to create and poof...it does it. I can see what all the people on the vibe coding reddit subs are talking about in that you might create 10 or 20 apps and only 1 ever sees the light of day. I was bored yesterday and made an iPhone palm reading app in like 5 or 10 minutes, lol. It's insane.
I managed to get around the lack of polygon.io access in all the other LLMs (besides Grok 4.2) by just purchasing a subscription and then I just pass the API key to Claude when I want to leverage the financial/quant stuff.
It's not quite as creative as Grok and I still go back and forth between the two but for anything remotely related to code...it's Opus 4.6. The only frustrating thing is how quickly I eat up all my token limits. I'm currently on the Max plan. You get extra tokens on the weekends though I think so I try to knock out the heavy stuff then. I keep reading about this "Mythos" they have behind closed doors and can't wait to get my hands on it for testing.
Interestingly, I find it useful to have Claude and Grok critique each other's output. I appreciate that Claude hallucinates less and is not afraid to tell me when I've got a bad idea but sometimes Grok will find something wrong with Claude's analysis and many times Claude will admit it made an error. That may be because I'm using Grok Heavy with more tokens/agents.
I can't wait for my Mac mini to come in so I can get OpenClaw up and running. I've built out a massive workflow already using Claude in preparation for all my little minion agents. Mush!
Last edited:
How long?
Idk what 90% of this meansI know dude...right?! I'm almost convinced that Anthropic has almost solved 100% of coding. I've got codebases on projects that would have taken me years to do on my own that I was able to do in days with Claude Opus 4.6 No other LLM comes close in regards to what I can do in Cowork or Code. It's like a dream generator. All you have to do is think of something you want to create and poof...it does it. I can see what all the people on the vibe coding reddit subs are talking about in that you might create 10 or 20 apps and only 1 ever sees the light of day. I was bored yesterday and made an iPhone palm reading app in like 5 or 10 minutes, lol. It's insane.
I managed to get around the lack of polygon.io access in all the other LLMs (besides Grok 4.2) by just purchasing a subscription and then I just pass the API key to Claude when I want to leverage the financial/quant stuff.
It's not quite as creative as Grok and I still go back and forth between the two but for anything remotely related to code...it's Opus 4.6. The only frustrating thing is how quickly I eat up all my token limits. I'm currently on the Max plan. You get extra tokens on the weekends though I think so I try to knock out the heavy stuff then. I keep reading about this "Mythos" they have behind closed doors and can't wait to get my hands on it for testing.
Interestingly, I find it useful to have Claude and Grok critique each other's output. I appreciate that Claude hallucinates less and is not afraid to tell me when I've got a bad idea but sometimes Grok will find something wrong with Claude's analysis and many times Claude will admit it made an error. That may be because I'm using Grok Heavy with more tokens/agents.
I can't wait for my Mac mini to come in so I can get OpenClaw up and running. I've built out a massive workflow already using Claude in preparation for all my little minion agents. Mush!
Soon enough, Radiology is guaranteed to be cooked by AI.
While I know what it all means, I couldn't do 90% of what @Groove does because in the end he has quite a bit of CS, programming and development experience.
I've tried playing around with these models to make websites, skeleton apps on my phone, and tried to create a thing to help streamline my property management (failed/wasn't that useful/didn't know how to make it useful), and I quickly realized that despite the democratization of coding that AI will bring to the masses, most people won't be able to do anything useful with it any time soon.
The initial steps to get an openclaw or hermes agent environment going will filter out 99% of people.
I'm on the side of "AI will create 10x the amount of engineering jobs in the future"
AI is going to kill some jobs, for sure. But it’s also going to give every person with a good idea the ability to be CEO. All you'll need is an idea, AI provides the talent for free. From a world of cogs, to a world of creators.
Advertisement - Members don't see this ad
It means that these LLMs are great at addicting their users.
Everytime I think I have a good idea, I look it up and find that its already been thought.
D
deleted1223140
3 hours occasionally. 2 hours frequently. Most often radiology is the rate limiting step in dispositions.
They need to be. Boasting about how chill their work from home jobs are while getting paid close to 7 figures. I get it they're important and everything revolves around imaging but consistent delays and misses are frustrating. If there's one area of medicine that needs to be replaced by AI, it's them.
Here's advice from an Iditarod competitor: "NEVER let go of the sled!"I know dude...right?! I'm almost convinced that Anthropic has almost solved 100% of coding. I've got codebases on projects that would have taken me years to do on my own that I was able to do in days with Claude Opus 4.6 No other LLM comes close in regards to what I can do in Cowork or Code. It's like a dream generator. All you have to do is think of something you want to create and poof...it does it. I can see what all the people on the vibe coding reddit subs are talking about in that you might create 10 or 20 apps and only 1 ever sees the light of day. I was bored yesterday and made an iPhone palm reading app in like 5 or 10 minutes, lol. It's insane.
I managed to get around the lack of polygon.io access in all the other LLMs (besides Grok 4.2) by just purchasing a subscription and then I just pass the API key to Claude when I want to leverage the financial/quant stuff.
It's not quite as creative as Grok and I still go back and forth between the two but for anything remotely related to code...it's Opus 4.6. The only frustrating thing is how quickly I eat up all my token limits. I'm currently on the Max plan. You get extra tokens on the weekends though I think so I try to knock out the heavy stuff then. I keep reading about this "Mythos" they have behind closed doors and can't wait to get my hands on it for testing.
Interestingly, I find it useful to have Claude and Grok critique each other's output. I appreciate that Claude hallucinates less and is not afraid to tell me when I've got a bad idea but sometimes Grok will find something wrong with Claude's analysis and many times Claude will admit it made an error. That may be because I'm using Grok Heavy with more tokens/agents.
I can't wait for my Mac mini to come in so I can get OpenClaw up and running. I've built out a massive workflow already using Claude in preparation for all my little minion agents. Mush!
Wow. Most of our reads are under 30 minutes, the group actually wants a call if anything takes 60+ minutes. 2+ hours is nuts.
Ask and you shall receive:
CEO of America’s largest public hospital system says he’s ready to replace radiologists with AI
Mitchell H. Katz, MD, president and CEO of NYC Health + Hospitals, recently spoke during a panel discussion held by Crain’s New York Business.
radiologybusiness.com
Radiology is so cooked.
Specialties that do physical work (ER, Surgery, proceduralists like IR, IC, GI, Intv Pain) will have more protection from AI replacement in the short term, compared to specialties where thinking or interpretation is the main skill. Fully autonomous robots that must think and do procedures will take more time and capital, to develop. Plus, no intelligent robot will ever agree to deal with ED administrators.
That article mentions a human radiologist can read anything marked as abnormal.Ask and you shall receive:
CEO of America’s largest public hospital system says he’s ready to replace radiologists with AI
Mitchell H. Katz, MD, president and CEO of NYC Health + Hospitals, recently spoke during a panel discussion held by Crain’s New York Business.
Radiology is so cooked.
It's not AI detecting abnormal scans that's the problem. I worry about it marking an abnormal finding as normal and things getting missed.
Maybe in 10-20 years, but I think we are way ahead of our time here to replace any physician with AI. It needs to prove its accuracy before we go down that route.
Yeah, we're not quite there yet but I'm sure we will be soon enough. There was a meta analysis last year that showed overall accuracy around 50% but specialists had like 15-20% greater accuracy. Interestingly, the Stanford study from JAMA showed that docs + AI had LESS accuracy meaning that there was a bias towards the AI. Docs "hunch" got out ruled because the doc thought AI must know something they didn't when in reality their intuition was correct. The tech is best in radiology and worst in pediatrics (makes sense, more vs less training data). I think we're safe for awhile and even when it gets better diagnostic accuracy, it's going to be more of a tool than a replacement, at least for awhile. I know our radiologists use it but are basically green lighting whether they agree or not.
I think no doubt it will help us reduce diagnostic errors which is good for us because that would hopefully also reduce lawsuits and obviously would be better for patients as well.
Radiology also does a lot of procedures also it would still need a doctor to sign off on it legally since it would be practicing medicine
Billionaires are working on a solution to that problem.
This Startup Is Building An AI Doctor To Treat Common Ailments Like UTIs And Sore Throats
Certuma launched out of stealth with $10 million in funding from 8VC and a grand plan to make AI smart enough — and safe enough — to actually diagnose and prescribe treatments. Now he needs to figure out how to get the FDA on board.
www.forbes.com
It's absolutely clear to me that physicians are thinking about the AI wave like physicians (head in the sand, can't think outside of their own sandbox, don't know what they don't know), and not like the immensely powerful Tech Bro lobby/behemoth or profit-focused administrators or corrupt legislators.
This is obviously the beginning of the profession's demise as we know it. Doctoring will be unrecognizable in 10 years from what it is now.
This is obviously the beginning of the profession's demise as we know it. Doctoring will be unrecognizable in 10 years from what it is now.
It'll be fun when the AI-attorneys drag the AI-doctors into court for frivolous medical malpractice suits.
I've already experimented with this by uploading difficult radiology images (devoid of all HIPPA identifiers, of course) into LLMs and asking for a read. I never rely on it and don't do it unless I'm planning a confirmatory study to follow up with. But so far it's been surprisingly good at getting the diagnosis right and at creating a radiology report indistinguishable from an actual radiologist.
Don't forget we would need an AI judge to rule over the trial.
Advertisement - Members don't see this ad
Same as they do now
I was just thinking about how I wish someone with no expertise in AI would rank its impact on professions they have no experience in.
People who think we are 10 years away have not been paying attention to the pace of change in AI. I also am a big Claude fan. It is supposed to be 100x better every 4 years or so. The real issue is the lack of compute and power. If we feed it it will come. Also agree robotics and interacting in a 3d world is a bit further away but AI will speed up everything for us. I am optimistic about the future.
I find the folks in any field with “look at how dumb AI is” and they show some stupid response to a question and then say We have a decade. Simply this is like saying a 6th grader cant do calculus they wont be smart for 10 years. Yet we all know those 6th graders will get there. AI is the same except much faster. I suspect we will find the AI models will hit major walls and things will stop there, perhaps for a year, perhaps for longer.. but we are far from that. Look at FSED. Forget the tesla data that it is much better. Ask anyone who has a tesla and uses it. Game changer. The iterations have quickly improved. When I got my tesla i never used it, the ride was rough and jerky, turns sucked etc. Most recent update it is smooth, not quite as aggressive as I would like but it is impressive and I use it often. Cant for for the unsupervised version so i dont have to pay attention at all.
Correct.
It amazes me how people I consider intelligent can't make the simple intellectual jump to understand the power of this technology, and what it will do given the trajectory. They can't think beyond the surface, and I think their deep emotions on the matter prevent them from doing so.
"Oh, well, it hasn't replaced my job yet, we'll be good" - Head in the sand for what's going on right now
"I tried using ChatGPT, it couldn't do ANYTHING useful" - Writing off the tech because they have no idea how to use it
"They've been saying that for 10 years, I'd like to see them try" - Ignorant of how fast tech improves/iterates, how regarded can you be?
The most disappointing part of this all is how 99% of the physicians I talk to refuse to acknowledge real and obvious data/evidence right in front of them that this is a new paradigm (hate that word, but it's useful here). OpenEvidence is better than any PCP I've ever had, and I'm saying this as a physician. Sorry, it's the hard and objective truth, it's hard to hear (for the PCPs reading this).
My theory is that our zero-creativity rote-memorization kuk-yourself-for-the-next-advancement education and career path laid out for physicians (meds school, residency, desirable job, desirable admin role/whatever) has been existentially threatened, thus creating this head-in-the-sand effect we're seeing among recalcitrant physicians.
This disrupts the ENTIRE thing, and it's not surprising that physicians EMOTIONALLY reject the very real tsunami we can LITERALLY see.
Adapt or die. Simple.
Dario Amodei has been harping on these concepts for awhile now and hopefully governments and organizations and society has time to prepare. He has doubled down on saying that AI could eliminate 50% of white collar jobs in the next 1-5 years. He's been saying that since last year and people just don't get it. This is one of the smartest humans beings on the planet running one of the most successful frontier labs who's team just created a model so powerful that they are spending extra time building guard rails and allowing cybersecurity teams time to review it. Do people just think he's unhinged? Crazy? Look at some of these Capex spending numbers. Do the richest tech companies on the planet pour their collective hard earned cash into a hype product? Of course not.
People need to wrap their brain around the concept of AGI. This is a machine that can out think any human, across ALL cognitive domains including the absolute highest level of artistic and scientific creativity. True ingenuity with the ability to self improve. We're probably 2-3 years away from that and pray tell me what is the value of human cognitive work when you have a machine that can out perform you in every single cognitive task. People are used to these steady ramp ups of discovery and integration but we don't have anything remotely comparable to these AI product improvement curves. We're not talking a gentle logarithmic curve, we're talking a line that looks like a SpaceX rocket launch. Next few years are going to be wild.
The security is what has started bothering me lately. Like...how exactly do you secure a system that is smarter than you and recursively improves itself by re-writing its own code? The minute we "create" AGI, it will immediately create a better and more powerful version of itself. Hell, 70% of current Claude source code is written....by CLAUDE. I've read all the prosaic alignment crap and bootstrapping, CAST approach, etc.. but is all of that going to work on something that powerful? These are systems that will be able to hack any government in the world. Create viruses that could wipe out humanity. The guard rails are going to be so incredibly important and I still sometimes wonder if humanity's future will be similar to some of the great sci fi shows where AI was banned in some degree because of the danger to humanity. i.e. Dune.
I still think that the most powerful systems will be proprietary and owned by governments completely and utterly secured in a closed system because they are too dangerous to connect to the internet or give to the public. I watched an expert describe what they envision in the future and they said these systems will be heavily guarded because they are so dangerous and the creative power will be so vast that we simply won't need it anymore as a race so we'll lock these things away and guard them.
People need to wrap their brain around the concept of AGI. This is a machine that can out think any human, across ALL cognitive domains including the absolute highest level of artistic and scientific creativity. True ingenuity with the ability to self improve. We're probably 2-3 years away from that and pray tell me what is the value of human cognitive work when you have a machine that can out perform you in every single cognitive task. People are used to these steady ramp ups of discovery and integration but we don't have anything remotely comparable to these AI product improvement curves. We're not talking a gentle logarithmic curve, we're talking a line that looks like a SpaceX rocket launch. Next few years are going to be wild.
The security is what has started bothering me lately. Like...how exactly do you secure a system that is smarter than you and recursively improves itself by re-writing its own code? The minute we "create" AGI, it will immediately create a better and more powerful version of itself. Hell, 70% of current Claude source code is written....by CLAUDE. I've read all the prosaic alignment crap and bootstrapping, CAST approach, etc.. but is all of that going to work on something that powerful? These are systems that will be able to hack any government in the world. Create viruses that could wipe out humanity. The guard rails are going to be so incredibly important and I still sometimes wonder if humanity's future will be similar to some of the great sci fi shows where AI was banned in some degree because of the danger to humanity. i.e. Dune.
I still think that the most powerful systems will be proprietary and owned by governments completely and utterly secured in a closed system because they are too dangerous to connect to the internet or give to the public. I watched an expert describe what they envision in the future and they said these systems will be heavily guarded because they are so dangerous and the creative power will be so vast that we simply won't need it anymore as a race so we'll lock these things away and guard them.
Last edited:
That's funny, I started using openevidence about 4 month ago. Its absolutely amazing for board questions. About 50/50 for actual clinical practice. I've found it very useful for treatment (it is usually more up to date than I am), hit or miss for diagnosis. Pretty helpful for "make sure you rule out serious conditions X and Y".
Admittedly it could be operator error with how I word things, I would have no way to know. I like the feature where you can attach pictures but I've done that twice and it only spit out the correct answer (among the many things it listed) once.
I don't see any of this as permanent obviously and with the speed these things are improving I don't expect this to be an issue for very long.
That said, I think the biggest hurdle here is taking patient ramblings and putting it into a form that AI can actually deal with. I haven't used Abridge yet but maybe its already good at that.
We havent even discussed quantum computing. Thats a layer on top of AI. They are saying its ability to crack security protocols will be next level and even cause issues relating to the blockchain.
If a functioning quantum computer comes along cracking blockchains will be the least of our worries. They’ll be able to crack every bank account, government system and intelligence agency in the world.
Advertisement - Members don't see this ad
It’s coming..
AI is going to usher in a completely new age of entrepreneurship. I waste hours each week using it to help me dream up all sorts of ideas. Stories like this one are so inspiring. Sadly, I see what all the vibe coders lament about...amazing ideas that just never see the light because people are either too distracted or get lazy at the end or are intimidated to start something they've never started before, etc..
www.nytimes.com
How A.I. Helped One Man (and His Brother) Build a $1.8 Billion Company
Who needs more than two employees when artificial intelligence can do so many corporate tasks? It’s super efficient — and a little bit lonely.
www.nytimes.com
AI will continue to take over more and more of the busy work from us. For those of you who haven't used AI for charting it's a must-have and makes your life on shift much easier.
A friend of mine who's a medical director spent a year programming Chat GPT with a protocol (he uploaded 100's of his charts as examples). He recently hired a programmer software engineer to incorporate this into a standalone website. Home | ChartLock is one of the best examples I've seen. You use Dragon to dictate a really simple HPI, pertinent labs/studies, and diagnosis/plan and it will spit out a full chart hitting all the major billing/coding and liability points. It also automatically detects procedures and critical care and incorporates them into the note.
He's using all of the revenue from subscription for a non-profit to help young people get suicide prevention treatment, as it's a personal cause for him.
His original Chat GPT protocol was so good that Envision actively tried to steal it from him.
A friend of mine who's a medical director spent a year programming Chat GPT with a protocol (he uploaded 100's of his charts as examples). He recently hired a programmer software engineer to incorporate this into a standalone website. Home | ChartLock is one of the best examples I've seen. You use Dragon to dictate a really simple HPI, pertinent labs/studies, and diagnosis/plan and it will spit out a full chart hitting all the major billing/coding and liability points. It also automatically detects procedures and critical care and incorporates them into the note.
He's using all of the revenue from subscription for a non-profit to help young people get suicide prevention treatment, as it's a personal cause for him.
His original Chat GPT protocol was so good that Envision actively tried to steal it from him.
That's a fascinating product he's built. So very cool.
BTW I'm such a huge Dark Crystal fan and love your name, lol.
BTW I'm such a huge Dark Crystal fan and love your name, lol.
Looks interesting. I have made my own as well. The issue is it’s not hard to reverse engineer any of this. For example, have it spit out 100 charts, upload them into our own AI with instructions of what you want. Edit it to fit work own workflow. Use CODEX or some other coding Ai to generate the app / website. There is a reason software stocks have gotten hammered. The ability to code is no longer a moat like it was. Other than the 7 day trial i cant find pricing.
