I'm working on MedAngle, the world's first Agentic AI Super App for medical and dental school. You can think of it as literally everything one would need from day one of admission till graduation day as a doctor.
I myself am the first medical doctor and full stack engineer in the history of my country (250 million), graduated as a doctor at age 25, and we have over 100+ users [all of which are medical/dental students and doctors], 10s of billions of seconds studying smarter, hundreds of millions of questions solved, and more.
Our Super App has subsystems including MedGPT, MedAgent, Spaci (our own take on spaced repetition) and much more.
We're bootstrapped, and continuing to scale. If you are in medical school or know someone who is, please reach out!
My gut instinct is to call you a mooncalf, given the lengthy documented history of OpenAI "re-appropriating" things and wanting to profit from it. So here is my honest question, what light did you see them in before and what did you base that on?
Wow, that's a pretty massive decrease in latency, which should unlock some use cases, but the linked web page doesn't exactly make it straightforward to understand the differences between the models.
However, based off my personal experiences with general images models, Google in my opinion is the best for my workflows. Granted, I haven't tried far-east providers yet.
Pretty huge move. Google and their TPUs are looking infinitely more prescient as I think they are on their 7th generation, along with the offshoots it inspired like the LPU and even others, perhaps like Cerebras and their Wafer Scale Engine.
However, based off first impressions, it seems like this is meant for inference side, and not training, which is also an interesting choice.
Training is pretty much a 1x cost, and efficiency there is already on the way down with architectural improvements. Inference though is an ongoing cost which over time takes orders of magnitude more resources, so focusing on making that far more efficient means way greater gains over time.
What makes you think this? With wider adoption the ratio shall shift in favor of inference. And API price is becoming more important than SOTA capability.
Cerebras's Codex Spark 5.3 has been a huge flop. Small context window and old model. But hopefully they can improve so that we can benefit from 1000 tokens/second with GPT 5.5.
> early testing shows that Jalapeño will deliver performance per watt substantially better than current state-of-the-art
We're starting to see what really matters here, and though this is hand wavy the TPU makes similar claims.
I think googles memo about having no moat still stands (see: https://newsletter.semianalysis.com/p/google-we-have-no-moat... if you are unaware). It kind of makes sense that all of this is looking more like 60's to 90's IBM, DEC, Cray, Sun and the hardware race that happened then. History doesn't repeat but it often rhymes and I suspect that these efforts will follow the same trajectory.
To be clear, that is not "Google's memo". It's a memo by a guy who happened to work at Google. There is a diversity of opinions at a company that employs 180,000 people.
I had to check the date after seeing the headline, and again after opening the page. Thought it was April Fools.
Regardless, as a doctor and full stack engineer, I'm looking forward to learning more about their methodologies, their approaches, but I don't think this is going to be displacing MRIs or remotely close, based off the cursory initial glance. If their vision is to be able to provide end users with more actionable data with some kind of "low fidelity" medical imaging data that is somewhere above zero and or standard imaging and high fidelity modalities like CT/MRI, then this could be somewhat interesting.
Not a radiologist and not medical advice. Just my two cents.
False positive rates are extremely important in the medical system as it exists today, where most scans will come without a known baseline and doctors cannot prescribe "biweekly scans for the next 6 weeks to see what changes". If we can achieve the kind of imaging abundance they're imagining (which I don't know how to evaluate based on their short post), I think false positives become much less of an issue, at least in the context of cancer where malignancy is the only problem.
False positives are important because of Bayes theorem. Even a test that’s 99% sensitive in a high incidence population can be indistinguishable from noise in a low incidence population.
If it has a 1% false positive rate but the incidence is 1%, the vast majority of the positives are false. Then you have to deal with the consequences, including invasive procedures for further diagnosis.
If you’re searching for tens or hundreds of low incidence conditions in the general population at a time it’s absolutely worthless because basically every positive is a false positive. At that point save the scan fee, spin a wheel of body parts and go get a biopsy of that.
This is why doctors are confused why companies are offering periodic full body scans in normal people. They only test people who are high risk or symptomatic to confirm a suspected diagnosis. That extra signal is what makes the test useful.
Go down to the medical diagnosis section for a worked example.
Regarding cancers every human has all sorts of weird lumps that are generally meaningless.
In order for this to not be a boondoggle it would have to be spectacularly accurate to a degree previously unheard of. Just from a statistics perspective.
As we gain more data, might we be able to find patterns in that data that we now cannot see? I'm not only thinking of these regular scans but combining it with other data sources, like maybe regular, more complete blood panels, Apple Watch data, whatever we can get our hands on. Maybe we can find data points that together have a lower false-positive rate, like lump plus increased nightly body temperature plus weight loss.
As a person experiencing UV sensitive skin, I’ve had multiple wheel-spin biopsies which turned out benign as expected, and at least once a year I find a weird looking spot I have take pictures of and promise to monitor for a bit. I don’t think there’s any reason this kind of stuff couldn’t be extended to other cancers if non-invasive next steps were available.
If you’re UV sensitive and at a higher risk then you’re already in a high incidence population making the tests valuable statistically speaking. That test is wildly more accurate for you than it would be for me, and even still you’ve been the unfortunate recipient of many false positives. There’s no reason for me or most people to do that since practically 99% or more of the positive tests would be wrong.
Biopsies are expensive, waste time, hospital resources and carry risks of infection and scarring that do not net out positively for people who aren’t in your risk group.
Getting a totally random positive doesn’t put you into a higher incidence category so whatever follow up test you take will be just as inaccurate as the first one.
The reason to avoid them is the tests would be a waste of time, statistically, and expose you to a bad risk-reward profile.
If you knew apriori 99% of the positive tests are false positive why are you taking the test?
It’s literally just math. Sometimes the right thing for you on average is to do nothing, which feels bad, but it’s still the right thing to do.
> Curing cancer is one of the only things I’d take a pay cut to do.
Send an email to this head-and-neck oncologist's lab. I saw a talk he gave at a Chicago-area national lab on open-source models for identifying malignancies in scanned pathology slides, and was smitten.
Remember, commercialization isn't the goal. They don't need to make a profit, as a company, they just need to get people to invest in their company and not get charged with fraud for something along the way.
I read the site and it seemed pretty clear? It's a 3d, transparent, high res image of your whole body reconstructed from the wave data from a large number of high frequency ultrasound scans. But it's also a high end spa in San Francisco that softly scans your body. Then, you uh, do as you want with the data (presumably show it to your doctor, who will be perhaps bemused)?
Sometimes I cannot wrap my mind around how funny people can be and how utterly humorless HN has become with its downvoting. @dang this is valid for the community too!
I guess the key quality of the best humour is that everyone can see the humour in it; at least appreciate the joke even if they didn't agree with the underlying point. At best it makes them rethink what they previously thought about the topic. Much of what is offered as humour online (and also mainstream television) is theatrical rage, which generates laughs and applause (a.k.a. 'clapter') from people who already shared the rage, but to those not already on side it just falls flat. That seems to me what's going on here.
They've done multiple "evaluations" by third parties, but still, it seems that they aren't being fully transparent. I think the approach is quite interesting and novel, but this feels like deja vu.
I get why they aren't disclosing all the details, but it seems more hype-train-esque to me for this moment. I don't disagree that this could be big.
I've always wondered what that would be like. A fleet of 50 relatively modern flagship smartphones, wiped and retrofitted software wise to act as a homogeneous server, running ubuntu or centos or fedora, something like that.
Deals like these are quite rare, would be nice to know more about how the funding worked, but for customers, what are they doing? Total silence from Groq for a very long time now.
I'm working on MedAngle, the world's first Agentic AI Super App for premed, medical, and dental schools and recent graduates - young doctors.
MedAngle is literally everything one could need, personalized to their curriculum across 4-6 years of medical school. Quizzes, videos, notes, flashcards, reminders, scheduling, performance, search, and more.
Our Super App is comprised of MedGPT + MedAgent + Spaci (futuristic spaced repetition), which serve as layers over our massive collection of features such as the Smart Suite, Learning Library, Clinical Corner, Tested Tools and more.
100k+ users, 10s of billions of seconds spent studying smarter, invite only. Bootstrapped, growing nicely. I lead a team of top medical students and doctors.
I myself am the first medical doctor and full stack engineer in the history of my country (250 million), graduated as a doctor at age 25, and we have over 100+ users [all of which are medical/dental students and doctors], 10s of billions of seconds studying smarter, hundreds of millions of questions solved, and more.
Our Super App has subsystems including MedGPT, MedAgent, Spaci (our own take on spaced repetition) and much more.
We're bootstrapped, and continuing to scale. If you are in medical school or know someone who is, please reach out!
https://medangle.com
reply