HNHacker News
TopNewBestAskShowJobs

aabajian

1,743 karma · joined December 29, 2010

aabajian aaaaaat gmail.com 697b79
submissionscomments
aabajian··on The thing that brought me joy
AI is making us face the reality of non-physical goods and services. If a product exists solely as data (writing, music, code, drawings, movies, etc.) then the value is in the idea and its distribution, and less on its production.

I'll give some examples:

Novel algorithms: PageRank, BitTorrent, Like buttons, disappearing pictures, etc.

Artistic styles: Cubism, impressionism, Wes Anderson, etc.

The above algorithms are (relatively) straightforward to implement and could be implemented by Claude Code in a matter of hours if not minutes. But, you'd still need a means to distribute them.

Similarly, you can have AI generate an image in any of the above styles (or a combination thereof), but the image won't have intrinsic value unless you can finds a means of (profitable) distribution.

Put another way, there's limited value in being able to master physical tasks (playing piano, typing fast), but fundamental skills that lead to creative innovation will remain important...along with being able to package/market/distribute the AI-implementation of your ideas.

aabajian··on ChatGPT Health is a marketplace, guess who is the product?
As an interventional radiologist, I want it to be easier to see images from outside hospitals. Epic has nearly solved the problem of seeing outside medical records. Yet, I still can't see the images for the CT scan you had from the hospital across the street unless I call the file room and get the images transferred.

I imagine once data sharing is more robust, it would be easier to validate AI models (at least specifically for radiology).

aabajian··on Microsoft drops AI sales targets in half after salespeople miss their quotas
Auto Hotkey doesn't work well for Epic manipulation because Epic runs inside of a Citrix Virtual Machine. You can't just read Window information and navigate that way. You'd have to have some sort of on-screen OCR to detect whether Epic is open, has focus, and is showing the tab that I want to close. Also, the tab itself can't be closed...I'm just clicking on the tab next to it.
aabajian··on Microsoft drops AI sales targets in half after salespeople miss their quotas
Well, it isn't every user. We use a version of Epic called Epic Radiant. It's designed for radiologists. The tab that always opens is the radiologist worklist. The thing is, we don't use that worklist for procedures (I'm an interventional radiologist). So that tab is always there, always opens first, and always shows an empty list. It can't be removed in the Radiant version of Epic.
aabajian··on Microsoft drops AI sales targets in half after salespeople miss their quotas
I think MSFT really needs some validated user stories. How many users want to, "Improve my writing," "Create an image," "Understand what is changed" (e.g. recent edits), or "Visualize my data."?

Those are the four use cases featured by the Microsoft 365 Copilot App (https://m365.cloud.microsoft/).

Conversely, I bet there are a lot of people who want AI to improve things they are already doing repeatedly. For example, I click the same button in Epic every day because Epic can't remove a tab. Maybe Copilot could learn that I do this and just...do it for me? Like, Copilot could watch my daily habits and offer automation for recurring things.

aabajian··on Demand for human radiologists is at an all-time high
Your belief is held by many, many radiologists. One thing I like to highlight is that LLMs and LVMs are much more advanced than any model in the past. In particular, they do not require specific training data to contain a diagnosis. They don't even require specific modality data to make inferences.

Think about how you learned anatomy. You probably looked at Netter drawings or Grey's long before you ever saw a CT or MRI. You probably knew the English word "laceration" before you saw a liver lac. You probably knew what a ground glass bathroom window looked like before the term was used to describe lung findings.

LLMs/LVMs ingest a huge amount of training data, more than humans can appreciate, and learn connections between that data. I can ask these models to render an elephant in outer space with a hematoma on its snout in the style of a CT scan. Surely, there is no such image in the training set, yet the model knows what I want from the enormous number of associations in its network.

Also, the word "finite" has a very specific definition in mathematics. It's a natural human fallacy to equate very large with infinite. And the variation in images is finite. Given a 16-bit, 512 x 512 x 100 slice CT scan, you're looking at 2^16 * 26214400 possible images. Very large, but still finite.

Of course, the reality is way, way smaller. As a human, you can't even look at the entire grayscale spectrum. We just say, < -500 Hounsfield units (HU), that's air, -200 < fat < 0, bone/metal > 100, etc. A gifted radiologist can maybe distinguish 100 different tissue types based on the HU. So, instead of 2^16 pixel values, you have...100. That's 100 * 26214400 = 262,440,000 possible CT scans. That's a realistic upper-limit on how many different CT scans there could possibly be. So, let's pre-draft 260 million reports and just pick the one that fits best at inference time. The amount you'd have to change would be miniscule.

aabajian··on Demand for human radiologists is at an all-time high
Well I do enjoy procedures. As for diagnostics, it’s very different when you come from a CS background.

On a basic level, software exists to expedite repetitive human tasks. Diagnostic radiology is an extremely repetitive human task. When I read diagnostics, there’s a voice in the back of my head saying, “I should be writing code to automate this rather than dictating it myself.”

aabajian··on Demand for human radiologists is at an all-time high
4 years undergrad (CS, math, bio)

4 years med school

2 years computer science

6 years of residency (intern year, 4 years of DR, 1 year of IR)

16 years...

aabajian··on Demand for human radiologists is at an all-time high
I'm an interventional radiologist with a master's in computer science. People outside radiology don't get why AI hasn't taken over.

Can AI read diagnostic images better than a radiologist? Almost certainly the answer is (or will be) yes.

Will radiologists be replaced? Almost certainly the answer is no.

Why not? Medical risk. Unless the law changes, a radiologist will have to sign off on each imaging report. So say you have an AI that reads images primarily and writes pristine reports. The bottleneck will still be the time it takes for the radiologist to look at the images and validate the automated report. Today, radiologist read very quickly, with a private practice rads averaging maybe 60-100 studies per day (XRs, ultrasounds, MRIs, CTs, nuclear medicine studies, mammograms, etc). This is near the limit of what a human being can reasonably do. Yes, there will be slight gains at not having to dictate anything, but still having to validate everything takes nearly as much time.

Now, I'm sure there's a cavalier radiologist out htere who would just click "sign, sign, sign..." but you know there's a malpractice attorney just waiting for that lawsuit.

aabajian··on How HubSpot scaled AI adoption
Their founders (Brian Halligan and Dharmesh Shah) are credited with defining "inbound marketing."

Here is the original version of their book: https://www.amazon.com/Inbound-Marketing-Found-Google-Social...

aabajian··on How HubSpot scaled AI adoption
If you're looking for a company to blame for the endless Google results of "top-10 ways of doing X" or "the best new vacuum cleaners review", look no further than HubSpot. Their business model was based on helping small business gain traction by writing a a lot of verbose blog posts. So now when you're looking how to fix a leaking faucet, you first have to read about the history of faucets.
aabajian··on Birth of 86-DOS
The last line is the most interesting, "In October 1980, Microsoft's Paul Allen contacted Seattle Computer Products and expressed interest in reselling 86-DOS. The first version of 86-DOS licensed to Microsoft was 0.3. In July 1981, just a month before IBM PC was announced, 86-DOS was sold to Microsoft and renamed to MS-DOS."

IBM reached out to Microsoft sometime in 1980 about an operating system, so SCP would've had at least 8 months to look into why Microsoft wanted their DOS before selling it entirely to them.

Did Microsoft resell 86-DOS to anybody before changing the name to MS-DOS? Did SCP make any effort find out why Microsoft wanted their DOS?

aabajian··on My AI skeptic friends are all nuts
I've observed that AI excels at creative work, but not exacting sciences. For example, If I say, "Write the code to put the button in the upper right corner," I'll almost certainly get a button in the upper right.

Conversely, if I say, "Put the button in the upper right corner exactly like the existing button in the upper left," I'll get a button in the upper right, but not exactly like the upper left button.

aabajian··on Java at 30: Interview with James Gosling
Just going to remind y'all of this: https://www.joelonsoftware.com/2005/12/29/the-perils-of-java...

I went to a Java school. I remember my operating systems class involved writing simulated OS code in Java (for example, round robin for context switching). The argument was that it would be easier to understand the algorithms if the hardware complexities were minimized. I understand that sentiment, but I don't think Java was the right choice. Python would have accomplished the same task even better (understanding algorithms). I think there was a huge influence from industry to teach college students Java from day one. I had taught myself BASIC and some C back in high school, so it was a bit of a step backwards to learn a high-level language just to do simulated low-level OS programming.

aabajian··on What were the MS-DOS programs that the moricons.dll icons were intended for?
Max nostalgia. When I was ~12 years old I had an 386 PC we got for $5 from the thrift store across from the dump (a bargain for sure even at that time). I had to self-teach myself about DOS, BASIC, Win 3.1, etc. I somehow broke the Win 3.1 installation and all I could find was Win 3.0 on floppy disc. I got the thing reinstalled, but found that the sound drivers didn't work, random programs wouldn't open, and it kept crashing.

I learned years later that there are a huge number of changes between 3.1 and 3.0. The biggest being support for more memory and multimedia extensions. The latter was the first time I learned what a dynamic link library (DLL) was and that took me down the rabbit hole of C++.

aabajian··on AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms
From the paper, "Notably, for multiplying two 4 × 4 matrices, applying the algorithm of Strassen recursively results in an algorithm with 49 multiplications, which works over any field...AlphaEvolve is the first method to find an algorithm to multiply two 4 × 4 complex-valued matrices using 48 multiplications."

If you do naive matrix multiplication, you get a sense that you're doing similar work multiple times, but it's hard to quantify just what that duplicated work entails. Compare it to, for example, calculating the size of the union of two sets:

Total size = size(A) + size(B) - size(intersection(A, B))

You have to take out that extra intersection amount because you've counted it twice. What if you could avoid counting it twice in the first place? That's easy, you just iterate over each set once, keeping track of the elements you've already seen.

Strassen's algorithm keeps track of calculations that are needed later on. It's all reminiscent of dynamic programming.

What I find interesting is that it seems the extra savings requires complex values. There must be something going on in the complex plane that is again over-counting with the naive approach.

aabajian··on Why Pale Blue Dot generates feelings of cosmic insignificance
The other existential question I've asked is, "How do I know any of you are conscious?" It may be in fact, that I am the only one who is conscious.
aabajian··on An image of an archeologist adventurer who wears a hat and uses a bullwhip
I'm curious, is it the AI that should be blamed or the prompter who asks to generate something based on copyright?

For all of the examples, I knew what image to expect before seeing it. I think it's the user who is at fault for requesting a copyrighted image, not the LLM for generating it. The LLM is giving (nearly) exactly what the user expects.

aabajian··on Microsoft's 1986 IPO
People play up Bill Gates’ connections, sure, but next to no 16-year-olds could program BASIC in assembly.

Microsoft’s success was as much to do with them being a programming languages company first. DOS, Windows 3.1, and even Windows 95 shipped with an interpreter and compiler for their BASIC and C, respectively. This empowered developers to use and write code for the OS out-of-the-box.

aabajian··on Nobody cares
I agree 100% with the comment about the McDonald's food ordering system. It is legitimately the worst of all the major fast-food chains. It is slow, ad-heavy, and there are certain things that you cannot order.
aabajian··on Calculating the norm of a complex number
Does the problem boil down to selecting the real part of a complex number? Correct me if I'm wrong, but the conjugate of z is merely 2*Re(z) - z. For example, the conjugate of a + bi is 2*a - (a + bi) = a - bi.
aabajian··on What is theoretical computer science?
My master's degree in computer science technically says, "Theoretical computer science" as the specialization. I was a mathematics undergrad, and generally enjoy doing proofs, albeit I'm not the best at them.

With that said, I regard TCS as the study of discretized computation at scale. Or, more colloquially, what is the fastest way to do X, N times? This could be sorting, searching, indexing, drawing, tracking, balancing, storing, accessing, saving, reading, writing, whatever. The distinguishing characteristic is complexity analysis in time and space.

As a counterexample, I think cryptographic methods such as AES and RSA are much more pure math / number theory than TCS.

aabajian··on Bridge to Nowhere
Neurosurgeons are extremely smart, hardworking, and (usually) amicable physicians. They are also in short supply and can determine whether an entire health system makes or loses money.

Neurosurgery is the hardest residency, and the hardest physician career. However, this difficulty is in-part due to the bravado of the neurosurgeons themselves. During neurosurgery residency, you have to learn to be a brain and spine surgeon, undertake two years of research, learn how to do endovascular interventions, and (of course) manage all of the pre-/post- operative care including clinic. It is not uncommon for a single PGY-2 neurosurgery resident to follow 100+ inpatients. Suffice to say neurosurgeons are overburdened, yet they themselves have fought against reducing training hours.

Patients can generally expect to spend less than 30 minutes speaking to their surgeon before/after their surgery.

For those unaware, you can become a spine surgeon via either the neurosurgery or orthopedic surgery routes. Just my simple opinion, but I think spine surgery should be its own residency. Brains should be left to the neurosurgeons and extremities to the orthopedists.

aabajian··on Launch HN: Modern Realty (YC S24) – AI Real Estate Agent for Home Buyers
There are no websites that let you make offers on houses by yourself. If you want to make an offer on a car, a boat, even land, there are ways to do it without an agent. Real estate (save land) seems the lone exception where you have to involve an agent. If you try to make an offer without an agent, the seller's agent becomes your agent and takes the fee. While they don't list their prices, I bet there's some people who'd pay the standard 3% to be able to control the process themselves.
aabajian··on Launch HN: Modern Realty (YC S24) – AI Real Estate Agent for Home Buyers
I've purchased two condos and one house in the Seattle area. My first condo I bought remotely and my agent toured a couple of places for me. Agent was very valuable in this case as I was in medical school trying to buy a place for residency.

Second condo I found myself and my agent merely facilitated the transaction. She hired a home inspector who missed a couple things. Not saying I would've caught them with my own inspector, but when your agent is chummy with an inspector, you can't really call them out on their mistakes.

When it came to looking for houses, I saw a few with my agent, but the one I ended up buying I found myself on Redfin. I went to an open house myself and just told my agent I wanted to make an offer.

Of note, my agent got paid the least for the first condo (because it was the cheapest), but she was most valuable in that transaction. The other two transactions just required the agent to fill out the paperwork, yet they got paid much more. There has to be some relationship between the amount of work done and the cost of the buyer's agent, not just a percentage of the sale price.

aabajian··on The Palletrone is a robotic hovercart for moving stuff anywhere
You've taken it further than I was even thinking. There are a lot of potential applications. A micro-drone swarm could aid in construction by moving building materials into place. They could grab a beer from the fridge for you. The only requirement would be proximity to the recharging station (meaning they couldn't, for example, walk your dog).
aabajian··on Launch HN: Modern Realty (YC S24) – AI Real Estate Agent for Home Buyers
I personally know Raffi, he was my roommate in college. If ever there was a CEO you wanted to back (or someone to help you buy a home), he is it. Some highlights:

-Not from a wealthy background, he got where he is due to hard work.

-Paid his share of the rent by trading stocks.

-He bought a used SUV off of Craigslist and negotiated the price down by almost 90%, afterwards the seller smiled and said something akin to "the balls on this guy!"

-Double majored in electrical engineering and economics in college.

-Worked at the US Patent Office reviewing video card patents, well before cryptocurrency, bitcoin mining, or AI.

-He built his own desk using woodworking skills.

-Studied contract law and has closed some multi-million dollar deals for other startups. He quit this high-paying, stable job to start Modern Realty.

-Has his own number of AirBnb properties.

Raffi has the very unusual combination of street smarts, book smarts and salesmanship.

aabajian··on The Palletrone is a robotic hovercart for moving stuff anywhere
Why not make these much, much smaller, and have the battery swap be part of the design? Connect 100 bee-sized platforms together in a grid. Any one of which could be the "battery swapper" bee and keep the rest of the flock afloat?
aabajian··on Show HN: Put this touch sensor on a robot and learn super precise tasks
I get the application being a touch sensor, but the real breakthrough is the embedding of magnetic particles with parallel orientations in a flexible medium. A few immediate questions:

-The company that makes Magnequench presumably knows their particles can be embedded in other materials. What's the most common use case for these types of particles? Are they typically embedded in fixed (non-flexible) solid medium or liquids?

-Is it necessary to use Dragon Skin? The idea of mixing magnetic particles in a semi-solid medium is intriguing to me. Like putting particles into Jello or Silly Putty. Could I then apply an external magnetic field and have the particles deform/shape the medium they are in?

aabajian··on Artists score major win in copyright case against AI art generators
IANAL. Is it legal to create derivatives of copyright work and then post them on public online forums? For example, I can certainly write, "Mickey Mouse got food poisoning from his Big Mac." But, if I ask an AI generator to "Make a picture of Mickey Mouse getting food poison at McDonald's", could I post the resulting picture?
← PreviousPage 2 of 10Next →