HNHacker News
TopNewBestAskShowJobs

sxg

1,702 karma · joined April 29, 2016

undersample.substack.com github.com/sxg
submissionscomments
sxg··on Using AI to negotiate a $195k hospital bill down to $33k
Anecdotally, it's worked out for a number of friends and people on /r/PSLF. There's definitely poor communication around PSLF, but it is a real program.
sxg··on Using AI to negotiate a $195k hospital bill down to $33k
Public service loan forgiveness (PSLF) exists and a huge number of people in medical professions actually take advantage of it. I know of multiple medical students and residents with over $500k in debt that are in the process of having all of their loans forgiven after 10 years in training and a total cost of approximately $75–150k for their entire education. Sure, that's still a decent amount of money, but it's very much worth the ROI.

https://en.wikipedia.org/wiki/Public_Service_Loan_Forgivenes...

sxg··on OpenAI acquires Sky.app
https://www.macstories.net/stories/sky-for-mac-preview/

This is what you need to know.

sxg··on Claude Code: Now in Beta in Zed
I thought the same, but I gave Helix a shot for fun a couple years ago and never looked back. It really does feel better/more ergonomic, but the greatest benefit is that almost everything you need is built in. I spent way too much time fiddling with Vim and NeoVim configs.
sxg··on Apple Intelligence Foundation Language Models Tech Report 2025
Siri is the default and only voice assistant that has access to all the data on your phone. It doesn't matter if I have ChatGPT, Claude, Gemini, or another SOTA model on my iPhone—I can't easily activate them in the car or another handsfree situation or use them with any other app or data on my iPhone.
sxg··on Apple Intelligence Foundation Language Models Tech Report 2025
OP isn't asking how to use Siri to do his contrived task. OP is saying that Siri in 2025 should be able to handle that relatively simple albeit contrived task.
sxg··on Preliminary report into Air India crash released
Now try to design a plane that also lets you rapidly shutoff fuel to both engines in case of fire.
sxg··on Preliminary report into Air India crash released
Studying how humans make errors is a fascinating field. Simply banning someone who's made a "slip" error as hypothesized wouldn't actually reduce the likelihood of this error occurring in the future. These sorts of errors are stochastic and could happen to anyone at any time. Preventing them requires a lot of thought.
sxg··on New York to build one of first U.S. nuclear-power plants in generation
Can you clarify? I thought the idea was to run nuclear 24/7 to provide a steady, base rate of power while solar and wind provide complementary power that can be quickly ramped up or down.
sxg··on Claude 4
Claude names their models based on size/complexity:

- Small: Haiku

- Medium: Sonnet

- Large: Opus

sxg··on Google AI Ultra
I disagree. There are massive fixed costs to developing LLMs that are best amortized over a massive number of users. So there's an incentive to make the cost as cheap as possible and LLMs more accessible to recoup those fixed costs.

Yes, there are also high variable costs involved, so there’s also a floor to how cheap they can get today. However, hardware will continue to get cheaper and more powerful while users can still massively benefit from the current generation of LLMs. So it is possible for these products to become overall cheaper and more accessible using low-end future hardware with current generation LLMs. I think Llama 4 running on a future RTX 7060 in 2029 could be served at a pretty low cost while still providing a ton of value for most users.

sxg··on Bus stops here: Shanghai lets riders design their own routes
> One of the amazing things of the current system, as simple as it is, is that it's predictable and doesn't require coordination.

In many cities, the exact opposite of that has been true in my experience. I’ve waited at bus/train stops only for it to be 20+ min late or never show up multiple times per week. The unpredictability makes it infeasible as a means of transportation to getting to work or anything time sensitive (e.g., sporting event or show downtown). This is a much bigger problem in smaller cities with rudimentary public transit, but I’ve also experienced it in larger cities like Philadelphia.

sxg··on Claude Integrations
Agreed—we're working with different definitions of "research". The deep research products from OpenAI, Google Gemini, and Perplexity seem to be more aligned with my definition of research if that helps you gain more utility from them.
sxg··on Claude Integrations
That's actually not what deep research is for, although you can obviously use it however you like. Your query is just raw data collection—not research. Deep research is about exploring a topic primarily with academic and other high-quality sources. It's a starting point for your own research. Deep research creates a summary report in ~10 min from more sources than you could probably read in a month, and then you can steer the conversation from there. Alternatively, you can just use deep research's sources as a reading list for yourself so you can do your own analysis.
sxg··on Generative AI is not replacing jobs or hurting wages at all, say economists
> Read Paul Tetlock's research about so-called "experts" and their inability to make good forecasts

Do you mean Philip Tetlock? He wrote Superforecasting, which might be what you're referring to?

sxg··on The Seven-Year Rule
Lots of comments here disagreeing with the article and missing the point...sure, the seven-year rule can be taken to a ridiculous extreme, but I think many people suffer from dwelling too much on past mistakes and overcorrecting them. Seven years seems like a roughly reasonable period of time to learn from a serious mistake and then move on with life as an improved person.

The spirit of the article is saying that you should ultimately forgive yourself for making a mistake after a period of time (not always literally seven years).

sxg··on Apple needs a Snow Sequoia
The point of this comment thread is that important Windows settings are scattered throughout many different interfaces beyond just the Settings app, and you can never be sure where to find what you're looking for, which results in a poor user experience. Off the top of my head, you have the Settings app, Control Panel, Device Manager, System Configuration, and Network and Sharing Center.
sxg··on Apple needs a Snow Sequoia
Using search as a UI means you can only find things that you know exist, but there are plenty of important settings that I've only discovered by actually navigating through the UI.
sxg··on Portable MRI Scans Patients for Signs of Stroke
If the patient is having a hemorrhagic stroke, fibrinolytics could cause catastrophic bleeding. As soon as you’ve excluded hemorrhagic stroke (and other hard contraindications), you give fibrinolytics ASAP if you suspect an ischemic stroke.
sxg··on Portable MRI Scans Patients for Signs of Stroke
I’m a neuroradiologist. The article is confused about the stroke workup—the most critical thing is to detect acute hemorrhage, which is much easier, faster, accurate, and cheaper with a CT than an MRI. We already have mobile CT units in deployment.

That’s not to say there isn’t value in mobile MR, but it’s not as the article states.

sxg··on Show HN: I Built a Visual Workflow Automation Platform – FlowRipple
This is a guess based on personal experience and old HN comments, but I think HN boosts potentially promising links based on how quickly they're getting upvoted (i.e., momentum). If the link doesn't keep gaining momentum after the boost, it'll quickly fall off the front page.
sxg··on Privacy Pass Authentication for Kagi Search
I think this supports my point and the OP’s example. Video game makers have figured out how to segment their customers into two groups (former and latter in the OP’s example), and this only works because they’ve made their games extremely cheap or free.

A cheap/free game supercharges network effects to amass players, each of which incrementally adds value to every other player. Most players will never directly pay enough to offset their own cost to the game maker. However, they will create a real community that draws in a small number of whale players who will directly pay for themselves and indirectly pay for all of the free players.

Not so different from the two-sided markets on Facebook and Instagram.

sxg··on Privacy Pass Authentication for Kagi Search
The other part to this is that the internet accelerates network-effects, which you can further supercharge by making your product as cheap as possible or free to the former group in your example.

It’s hard to make money by charging a lot to a small group of people since now you’re dealing with anti-network effects. Doubling the price of a product will likely more than halve your user base.

sxg··on Who exactly needs to get approval from an institutional review board (IRB)?
I’m a radiologist, and I’ve worked on these sorts of research studies before. You absolutely do need an IRB to do deep learning on clinical MRIs. I’ve had to write IRBs just to collect retrospective MRIs. Even for studies that are potentially IRB exempt, going through the exemption request takes hours of work on my part and weeks to months before the IRB grants the exception.

As other commenters mention, the “theoretical risk to a human” encompasses nearly all research. For what I do, any imaging studies that aren’t de-identified pose a theoretical risk to patient privacy. If you try to de-identify images, you learn that this is nearly impossible. Sure, you can try your best to scrub DICOM headers, but these headers are mis-used by vendors, so identifying information can appear almost anywhere. You could delete the headers entirely, but then you lose a lot of metadata that you may need to properly display the images. Further, people contend that you can identify individual peoples’ faces if you 3D-reconstruct CT/MR images, so then you have to expend resources to delete faces from all head/neck/brain imaging. Confirming that this was done properly requires manual review and limits the size of your dataset.

Edit: I think the disagreements here are partly due to institutional differences in IRB requirements and partly due to conflating “IRB exemptions” with the idea of not having to interact with the IRB at all. You always have to interact with the IRB—even just to obtain an exemption. While obtaining IRB approval is a cumbersome process, obtaining IRB exemption is only slightly less cumbersome. I’m sure this varies across institutions, but I’ve been at three different large urban academic centers, and obtaining exemptions has been a multi-month process at all three.

sxg··on Ask HN: Is there a [blazing fast] alternative to homebrew for macOS yet?
For someone whose needs are generally met with homebrew, what are its major issues (aside from speed)?
sxg··on Apple Intelligence is coming to the EU in April 2025
Most people aren’t upgrading from the 15 to a 16, so they’re probably getting more than just a button. You could argue that the only update for the 15 over the 14 was also just a button.
sxg··on Apple wins $250 US jury verdict in patent case over Masimo smartwatches
Masimo's CEO and board have to approve the purchase, but it appears that Masimo isn't interested in being bought.
sxg··on CrowdStrike's impact on aviation
The ad-driven model does count as free, and it's far less likely to deliver better quality news than a subscription service users pay for. The core metric for ad-driven news sites is maximizing views—it doesn't matter how you get views as long as you get them. This means free sites are heavily incentivized to be the first to break a news story even if the details are wrong or sparse. Sure, they'll issue corrections and updates later, but only a small percentage of the initial viewers will ever see these, and there's essentially zero cost for having made the mistake.

The core metric for subscription news sites is minimizing churn. A mistake will cost a subscription site subscribers who have a massive lifetime value. These sites are heavily incentivized to report high quality, accurate news even if they're not the first to break the story.

sxg··on iTerm2 and AI Hype Overload
> These are exactly the people who shouldn't be running code written by a random number generator.

The beauty of technology and the internet is that there's near-zero gatekeeping. Anyone with enough interest and time can learn to build anything without seriously harming anyone. Dismissing LLMs as a random number generator is very clearly an overstatement for the value they're already able to provide. Ideally, how would you suggest a new developer learn to build something such as a REST API in Python?

sxg··on iTerm2 and AI Hype Overload
No worries—and it's clear that it's your personal problem. I think it's always valuable to present a counterpoint in these comment threads so that less-informed readers know the issue isn't totally clear and that the truth likely lies somewhere between your point and mine. I apologize if I came across as argumentative—my goal is to expand the scope of the discussion.
← PreviousPage 2 of 11Next →