Googles NEW "Med-Gemini" SURPRISES Doctors! (Googles New Medical AI)

Googles NEW "Med-Gemini" SURPRISES Doctors! (Googles New Medical AI)
How To Not Be Replaced By AGI • Life After AGI How To ...
Stay Up To Date With AI Job Market - / @theaigrideconomics
AI Tutorials - / @theaigridtutorials
🐤 Follow Me on Twitter / theaigrid
🌐 Checkout My website - theaigrid.com/
00:22 Intro
00:31 Med Gemini Development
02:34 Fine Tuning
03:46 Previous SOTA
06:18 New Benchmark
08:53 Advanced Reasoning
10:25 Model Architecture
13:46 Video Benchamrks
14:27 Overall Benchmarks
16:55 Benchmark problems
18:45 Dialogue Exmaple
23:42 AMIE Vs Med Gemini
Links From Todays Video:
arxiv.org/pdf/2404.18416
Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.
Was there anything i missed?
(For Business Enquiries) [e-mail]
#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

[@MediaCreators]

It comes down to this: With medical errors ranking third among causes of death in the USA, one might consider consulting GPT-2 rather than any physician.

[@UFOgamers]

So you're telling me 2:00 that most clinician have less than 40% accuracy? That explains a lot.

[@graphguy]

Harvard med school dean meets incoming new meds students and tells them that 50% of what they learn is wrong. Then he goes on to tell them, “that is not the problem. The problem is , we don’t know which 50%.”

[@nefaristo]

Not only doctors though, if you want a wider and more depressing picture about the testing of experts in many fields, try Kahneman's "noise". I stopped liking humans to be in charge of judging me, assessing me and especially diagnosing me. Sorry for medicine students starting now...I think you'll be obsolete before finishing your expensive studies.

[@danielbrockerttravel]

@@nefaristo Honestly though I'm optimistic about how this will help human doctors and other medical personnel. Many of them are overworked and overstressed. I drive for Uber and I would have to have to collect cash, worry about getting robbed, asking for directions and being annoyed by the shitty directions my passengers were giving, trying to navigate. With the Uber app I just follow the app and don't deal in cash. I think that the job satisfaction for doctors may increase in an analogous way.

[@efficiencyvi8369]

Medicine is not an exact science, it's more like educated guessing. Because there are too many factors to consider that could all be meaningless. It's just the way it is.

[@OlayOlayO]

@@nefaristoai in med, will be working like assistants rather than replacing docs entirely, for that we need more years to come.

[@JavierLopez-yq4qu]

They should release an app where you upload your X-rays or other images, and gives you potential diagnostics. That should be easy.

[@user-uc9nu1yn1n]

They have some apps like this already, but remember the medical industry controls a large part of our executive branch and legistlative, so anything that takes away from their profits will be regulated by the FDA which will make it impossible to release or so costly that you will get charged an arm and a leg somehow.... Sad state of America. Drain the swamp.

[@vellocet1]

This exists in the open source community. The models score extremely well compared to human radiologists. What’s coolest is that you can run a medical image (or multiple) through one model, then through another and use custom GPTs to talk about the results. So you can get multiple opinions that you can show do your doctor to give them more information. Your doctor can’t run your images through these models because of privacy isssues, but you can and then provide them to your doctor

[@burdealexandru3179]

​@@vellocet1 so what is the name of the open source app? thanks

[@mr.reverseit5257]

@@vellocet1 can you share the app name please

[@dgpace]

This would be a great help. My example. Having some trouble breathing, guessing probably adult onset asthma but I am no doctor. Went to my gp who saw me in a day or two., Did some tests and sent to the pulmonologist. 4 months later I see the pulmonologist who does some tests and says not me, lets send you to the cardiologist. 6 Months later I get to see the cardiologist who runs some tests that take 4 months to get scheduled to be done (where I am at now, waiting). How about letting me use this and have it issue the tests (I am sure the tests are ones that insurance, both medical and malpractice dictates) so why wait all this time for a human to do all of this. Treat the AI as a PA or an NP and go forward. Would really be a help but would be disruptive.

[@jedigardener]

probably blood clots on the lung from the Covid MRNA vax! -sorry !

[@fractal69420]

I am SURPRISED this isn't SHOCKING me what a BOMBSHELL of a STUNNING realization.

[@vegan-ro9hm]

I see what you did there... :)

[@probably-ansh]

Brudda

[@zerorusher]

Are you the INDUSTRIE?

[@yahanaashaqua]

This means that we already have AGI

[@EEKmusic]

meh the joke is getting old now

[@TokyoNeko8]

Hooray new WeBMD!!!! Hopefully every answer won't also have a small print of "can lead to cancer" on every single diagnose

[@Gardener7]

Or on every medication, possible side effects include... death. 😮

[@DailyTuna]

After the Covid debacle, many found out that a lot of the data had to be recalled because it was compromised so how do we know when this stuff is trained that the data is good?

[@happytape307]

I can picture a bot in future that uses this kind of technology coupled with the ability to see, hear and feel its surroundings, being trained in advance human psychology, identifying an array of human emotions. They will be like GPs in the old days who do house visits, but arriving in autonomous vehicles in record time. They will be able to diagnose and even place online orders for prescribed medication and have it delivered by drone in just a few minutes. On the one arm it will have a Swiss Army Knife style choice of tools in case you need emergency surgery. It will do all this while remaining professional and friendly.

[@Petbrandpodcast]

sans leta hu toh AI grid ki ek aur video aa jati h

[@eltodesukane]

After IBM Deep Blue won a chess match agains Kasparov, and Watson won at Jeopardy, IBM spent a lot of effort to make a medically useful version of it. But nothing came out of it.

[@babbagebrassworks4278]

Depends on what it found, perhaps the causes? Lots of lawsuits if that info gets out?

[@DarklordWarlock]

Ada Health was not listed at all, despite it offering AI based health advice for years already. How does it compare?

[@digletwithn]

10:20 advanced reasoning texsniks

[@markhathaway9456]

The federal government should promote the use of AI for diagnostics and diagnosis, though it should not yet be the final word. It could perhaps be included via the PPACA -- Obamacare. We would see performance changes relative to non-Obamacare states.

[@Mr.moon_vol13]

I just had a chat with chat gbt 4 about my inner ear , so I m on board were can i get Gemini, East york family doctors these days wont even see you . in Toronto so this is way better ....

[@keirapendragon5486]

I wonder how much they've done to minimize unintended biases. For example, if it can use pictures to look for skin cancers example. Or another example of an actual error in reasoning - an AI was able to "accurately" predict cancer in patience because their scans were taken on older machines, and those older machines were located in areas where people were exposed to more carcinogens and thus more likely to Get cancer, and it had literally nothing to do with the actual patients at all.

[@qnaloonline9536]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@paolo.miscia]

This is the "Paradox of Artificial Humanity" at play - 100%

[@qnaloonline9536]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@hansherzog5097]

When is the release?

[@bartoszludera2604]

Is there any posibility to try this model?

[@Figurenow]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@TheDro]

So basically dr Gupta that shkreli had made? He’s got a vet one too.

[@petratilling2521]

Given these results, why isn’t it illegal for doctors to practice medicine anymore?

[@qnaloonline9536]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@michaelschlageter8381]

A docbot With a bloodtesting device and x-ray eyes. A comforting and soothing voice would also be nice. And at least one ear ..to listen... and not only to hear

[@Figurenow]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@RpKaswa]

The app does not make real-world clinical reasoning diagnoses; it is just a theoretical lecture without clinical reasoning. The LLM is successful because of its finite outcomes, but human biology is more complex and has infinite outcomes. There is still a lot to explore for human biology unknown-unknown, so predicting the model based on current knowledge is a data hallucination. I don't agree that it passes the test.

[@qnaloonline9536]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@user-uc9nu1yn1n]

I QA questions for stuff like this all the time, its definitely true many questions are so poorly formed that they have logical holes that intelligent people would not be able to answer correctly. Bar exam is rife with them.

[@Figurenow]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@mrd6869]

Good.maybe it can assist this half ass docters and help some people

[@cfjlkfsjf]

One day everyone will have their own doctor, wont need to wait inside for a doctor or visit the family doctor and it'll know your history you feed it over time. Hell you can have a team of doctors inside your walls that have their own room if you need it with all the tools they need. Or we just lay down on a machine that scans us for what's wrong, then again a 2nd time to fix the problems instantly like the movie Alysium.

[@Figurenow]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@FutureGuy47]

15:49 What? Humans have way less data points than an AI. That's why AIs are way more accurate.

[@Figurenow]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@knexilva]

BAYMAX! BAYMAX! BAYMAX! 🤩🐻‍❄️

[@honkytonk4465]

I want my old style shaman back

[@akzsh]

stop these thumbnails 😭

[@josephboomtv7811]

Holy god! This is unspeakably Shocking

[@honkytonk4465]

Are you traumatized?

[@J.Christo]

That’s really cool

[@Figurenow]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@Technoizm]

Are you Franklin Hatchet?

[@Figurenow]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@user-zs8lp3lg3j]

When you flirt with danger & get a pleasant outcome. That means your optimism is next to none. Efficacy is not the question. You should just get out the way. The test was introduced by Turing in his 1950 paper "Computing Machinery and Intelligence" while working at the University of Manchester.[5] It opens with the words: "I propose to consider the question, 'Can machines think?'" Because "thinking" is difficult to define, Turing chooses to "replace the question by another, which is closely related to it and is expressed in relatively unambiguous words."[6] Turing describes the new form of the problem in terms of a three-person game called the "imitation game", in which an interrogator asks questions of a man and a woman in another room in order to determine the correct sex of the two players. Turing's new question is: "Are there imaginable digital computers which would do well in the imitation game?"[2] This question, Turing believed, was one that could actually be answered.

[@mvfc7637]

this is how the regime will get around the #competencycrisis

[@Figurenow]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@user-yl7kl7sl1g]

Learn to code, doctors.

[@qnaloonline9536]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@PaweAdamowicz1981]

Will it see the differece, betwen races and genres? Or it will be the same as last time, with asian ssmans?

[@elsavelaz]

@18:00 so out of thousands of doctors some special 3 docs decided how good the questions are- is there data on diversity in that committee? Asking because jm a female minority in stem in AI for now 8+ years, simultaneous Spanish bilingual with a masters in education and 9 years experience with that. When you have simultaneous bilinguals in Korean or Spanish, it’s easier to understand what sentence compositions lose their essence because of how those 2 languages compose the noun in a sentence. When I went back to school for computer stuff, having that knowledge helped me understand why school had been so difficult prior, and probably provides indication on how to be more inclusive

[@Greg-xi8yx]

I’ve noticed that every comment that you leave on this channel makes sure to make reference to your not particularly impressive accolades. We’re just not that interested in you no matter how often you try and force it. 🤣🤡

[@petratilling2521]

What this tells me is how very terrible doctors are.

[@hbluisfilipe]

❤❤❤❤❤❤❤❤

[@wkardita]

Why isn't anyone uses ai to solve crime, instead of this?

[@honkytonk4465]

Which of those two problem has the higher priority?And why not do both?

[@goodie2shoes]

shock3d

[@forgotten_world]

People will die, because professionals don't understand the underpinnings of AI, and the risks involved about its results.

[@aaronhhill]

You do understand that the AI the general public use and the AI that the scientific community use are not the same. Right?

[@forgotten_world]

@@aaronhhill It is being developed as a tool for professionals, physicians, pharmacists and other, what researchers use is not the topic of this video.

[@aaronhhill]

@@forgotten_world You misunderstand me. The AI being developed for professional use is trained on data for that specific profession. The AI that is being deployed for use by the medical and scientific community is vastly different than the AI the general public are using. When you say "people will die" you sound like you don't fully understand the difference between the two. I could be wrong. However, it's not as if the AI will immediately be deployed and doctors will no longer be necessary. The AI will be a useful guide, and is capable of noticing things that human doctors may overlook. The AI would not be given full control to diagnose and treat patients. So, your statement lacks understanding.

[@bdown]

People are already dying whats ur point?

[@forgotten_world]

@@aaronhhill Data is not intelligence, the problem with AI is that current systems are clueless about the real world, trained data is and always will be limited regarding health possibilities, until the day we could understand consciousness and replicate it into AI systems. The data is very specific and not holistic in nature, issues will arise when health professionals bring overconfidence to the systems results, not doing appropriate checks, what will result in more medical errors. The problem is not the AI systems themselves, those are important tools - the problem is the marketing over it, the greed, and the overconfidence of not properly trained health professionals. This is exactly what plagues current allopathic treatments, AI will be promoted as another panacea.

[@Simplicity4711]

Great content, but you say "of course" a lot, even where it doesn't make much sense.

[@qnaloonline9536]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@Behzad-yw5ox]

2

[@antaka503]

first

[@MomoTrader809]

But it’s Google so every diagnosis is woke 😂 “your pain is due to a diagnosis of trans…here’s your hormones.”

[@qnaloonline9536]

GPT-5 Releasing soon.... Found An Amazing video of GPT-5 On"Insights With Om"

[@olehoiii]

*Refuses to help white males