HNHacker News
TopNewBestAskShowJobs

n8henrie

795 karma · joined May 5, 2014

https://n8henrie.com

F21A6194C9DB9899CD09E24E434B2C14B8C33422

[ my public key: https://keybase.io/n8henrie; my proof: https://keybase.io/n8henrie/sigs/WoQvsZKguJVoMfO7Fxk2mnWhN1Xh0cLZuDGQnxpwhWY ]

submissionscomments
n8henrie··on OpenAI’s o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors
> Even now, I'll take a surgeon that's a complete jerk over a nice surgeon any day, because if they've got that job even as a jerk they've got to be good at their jobs.

This seems like an incredibly poor line of reasoning.

Hospitals are often desperate for surgeons. The poorly mannered ones are often deeply unsatisfied, angry at the grueling lives they've opted into, and the hospitals can't replace them. The market is not exactly at work here.

n8henrie··on OpenAI’s o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors
So why not order your own labs? I'm sure you can think of ways to get your own medications if you are sufficiently convinced that this is the best course of action for your health.
n8henrie··on OpenAI’s o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors
Unfortunately the training data is absolute garbage.

Diagnostic standards in (at least emergency, but I think other specialties) medicine are largely a joke -- ultimately it's often either autopsy or "expert consensus."

We get to bill more for more serious diagnoses. The amount of patients I see with a "stroke" or "heart attack" diagnosis that clearly had no such thing is truly wild.

We can be sued for tens of millions of dollars for missing a serious diagnosis, even if we know an alternative explanation is more likely.

If AI is able to beat an average doctor, it will be due to alleviating perverse incentives. But I can't imagine where we could get training data that would let it be any less of a fountain of garbage than many doctors.

Without a large amount of good training data, how could AI possibly be good at doctoring IRL?

n8henrie··on Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model
I'm still fairly new to local LLMs, spent some time setting up and testing a few Qwen3.6-35B-A3B models yesterday (mlx 4b and 8b, gguf Q4_K_M and Q4_K_XL I think).

Was impressed at how they ran on my 64G M4.

It looks like this new model is slightly "smarter" (based on the tables in TFA) but requires more VRAM. Is that it? The "dense" part being the big deal?

As 27B < 35B, should we expect some quantized models soon that will bring the VRAM requirement down?

n8henrie··on Amazon is discontinuing Kindle for PC on June 30th
I strongly disagree. If it's doing well enough for the owner then it's doing well enough. I don't understand how one can tell someone else that their computer is unacceptably slow for that other individual's personal use.

This is a really unfortunate move by Amazon. My next e-reader will be one that I own (instead of just rent).

Glad that I took the time to jailbreak and pause updates on my 2017 kindle paperwhite while I could.

n8henrie··on Shell Tricks That Make Life Easier (and Save Your Sanity)
Did this many years ago (but with bash) -- life changing is an apt way of saying it.
n8henrie··on Atuin v18.13 – better search, a PTY proxy, and AI for your shell
A ton more conflicting keybindings.

I switched away from tmux a year or so ago due to one crash I kept getting, but thinking of going back. Really miss the simplicity.

n8henrie··on Hammerspoon
Lua in HS is what motivated me to finally learn a lisp (Fennel).
n8henrie··on Espflash – Go CLI/library for flashing ESP8266/ESP32 with no dependencies
Is this related to the popular and longish-standing Rust tool for the same purpose? I expected to see it referenced somewhere in the readme...

<https://github.com/esp-rs/espflash>

n8henrie··on Tuna: A new, modern, modal launcher for macOS
Quicksilver is still around!
n8henrie··on How often do full-body MRIs find cancer?
I can't imagine this taking strong hold in the US unless it shields physicians from legal consequences of false negatives or produces enough false positives to ensure revenue doesn't fall.

I don't see any way that the hospital systems running healthcare in the US would embrace a technology that reduces false positives (income) without decreasing false negatives (risk and lost income) at least as much.

n8henrie··on How often do full-body MRIs find cancer?
But you're not having cancer checked out, you're having a "spot" or a "nodule" or something checked out.

And the person that's making the recommendation on whether or not to check it out may get sued for $10M if they tell you it's probably nothing and they're wrong, but have no harm come to them if they tell you it's worth having some other doctor do a biopsy.

And they might make an extra couple hundred bucks every time you have to come back and see them to follow up on this spot.

And the radiologist interpreting the MRIs... the same perverse incentives regarding how they interpret a "spot."

n8henrie··on How often do full-body MRIs find cancer?
I'd argue that malpractice risk has at least as much negative influence on a physicians judgment.

It's perceived as much less (medico-legally) risky to "do something" (or more often "refer the patient to someone else to do something") than not do something.

n8henrie··on How often do full-body MRIs find cancer?
Like when someone has an anaphylactic reaction to the contrast dye for their CT?

Contrast-induced nephropathy?

Gadolinium accumulation in the brain doesn't sound good for you...

Although I think this argument is usually talking about the risks of the resulting procedures (eg an injury or complication related to a biopsy done for a finding on imaging).

n8henrie··on How often do full-body MRIs find cancer?
Not to mention the malpractice risk and potential for extra income, which -- depending on where you live -- may strongly (if subconsciously) influence your physicians interpretations and recommendations.

Do you think a doctor is more likely to call something "possible cancer" and recommend that you either have a specialist do a biopsy (keeping in mind that many of these will be... hard to reach) or at least have a follow up MRI in 3, 6, 9 months?

Or tell you it's "pretty unlikely to be cancer, I don't think we need to worry about it" and then get sued for 20M when they are wrong about 1 in 100 cases (not to mention missing out on all the potential income from above).

At least in the US, the incentives here are grossly misaligned.

n8henrie··on Rivian R2: Electric Mid-Size SUV
Agreed! But I have to say, lane centering and adaptive cruise control have been amazing, coming from a previous vehicle with neither.
n8henrie··on Anki ownership transferred to AnkiHub
Amazing. A few nerdy med students and I started a student group for tech back in 2009 or so, a big part early on was preaching Mnemosyne (another SRS app), with shared decks syncing over a free Dropbox account.

Later used Repetitions (iOS / Mac / web) for the steps, EM boards, and informatics boards most recently.

Only within the last year finally tried Anki -- and this time for language.

n8henrie··on Embassy: Modern embedded framework, using Rust and async
Yes, I have several esphome devices. Very easy to use. Unfortunately the project is not written in a language that I know, and I strongly prefer projects in languages that are familiar to me.
n8henrie··on Embassy: Modern embedded framework, using Rust and async
I never learned any C based languages, so it's been a challenge, but I've enjoyed learning the basics of bare-metal no_std rust on the esp32c3, with esp-rs and its support for embassy to help me get started!

My learning project -- using mqtt for HomeAssistant integration: <https://github.com/n8henrie/esp32c3-rust-mqtt>

n8henrie··on Embassy: Modern embedded framework, using Rust and async
I don't believe so. But the esp32c3 is dirt cheap and runs embassy projects well!
n8henrie··on Jeffgeerling.com has been migrated to Hugo
Tried to migrate to Hugo from Jekyll multiple times and have bounced off every time. Don't really know Go very well, but better than Ruby, and used this as justification -- since dealing with Jekyll updates was sometimes a headache (I use GitHub Pages for free hosting and let them build things for me when I push updates).

Instead I eventually just created an environment in nix that had compatible dependency versions to what GitHub uses and have been pleased since.

n8henrie··on Going immutable on macOS, using Nix-Darwin
After a decade of homebrew, a few years ago I got tired of their very grumpy maintainers and switched to nix-darwin + home-manager. I've been overall fairly happy, and for tinkerers would recommend giving it a shot. Admittedly I bounced off my first try a year before that.

A few of my favorite parts, which I see less represented in this thread so far:

- I simultaneously jumped into nixos on several Linux machines (starting with a few Pis for experimentation), and maintaining all of my systems with a single flake and mostly shared code is a dream come true. - no more convoluted dotfile syncing, most of my scripts and aliases and bash config and binaries all sync together - cross-building linux from Darwin -- including integration tests in a vm -- works surprisingly well, this is mostly just nix but nix-darwin has helpers that make this easier - writing system services (launchd) on my Mac then converting them to a headless Linux machine (systemd) is generally very straightforward - prefixing my path with GNU coreutils works well and saves me from many e.g. `sed` quirks, I get expected behavior across OSes - this was always a sore spot in homebrew, either dealing with the `g` prefix on everything (eg `gsed`) or dealing with intermittent breakages when stuff depends on the BSD behavior - I was also able to put nix-darwin on my wife's MacBook and greatly simplify admin / maintenance tasks I do for her - finally, the nix crew is just thirsty for help and contributions, particularly the darwin crowd; I feel like my (minor, occasional) contributions are celebrated, differences of opinion are met with an open mind, it is in general a far departure from the relative hostility of homebrew - on the down side, I have spent far more of my limited time helping contribute to nixpkgs / nix-darwin / home-manager

n8henrie··on Going immutable on macOS, using Nix-Darwin
I installed on macos14? Maybe 13? And have had minimal trouble since.

Comparable to the amount of breakage I regularly had with homebrew.

n8henrie··on Scientists unlock brain's natural clean-up system for new treatments for stroke
Citation?
n8henrie··on Show HN: Per-instance TSP Solver with No Pre-training (1.66% gap on d1291)
Thanks. Was wondering if this was about my federal thrift savings plan.
n8henrie··on BehindTheMedspeak: A Spinal Tap
Yes, the possibility of severe and prolonged headaches are part of my consent for this procedure. That said, I'm usually only performing the procedure to help exclude (or confirm) a medical condition with risk of permanent disability or death, so it can be a tough decision at times.
n8henrie··on BehindTheMedspeak: A Spinal Tap
Love that you had a good experience. I perform these procedures semi-regularly, and in some cases they can be painful (even with lidocaine). Most people tolerate them very well though, I usually compare it to the pain of an IV stick, which most people have already tolerated, but which can also cause some people a surprising amount of distress.
n8henrie··on A brief history of Time Machine (2024)
Use Time Machine over SMB to a ZFS-backed drive. Works great, even remotely. See my longer comment elsewhere in this thread.
n8henrie··on A brief history of Time Machine (2024)
For the last several years I've very happily used it over SMB to ZFS (with autosnaps) for this very reason, and wrote an AppleScript to automatically "verify" it every week or so.

Once or twice a year it gives a verify error (i imagine this is because a plug gets pulled halfway through a backup on one side or the other), and I just have to go find the last verified date, zfs rollback, and then re-verify. Afterwards it picks up where I left off, and the historical backups are preserved.

Wish it didn't require this extra effort in the first place, but much better than having to nuke and pave every time.

Even better, it's working great over Tailscale so I can even use it remotely. Only big hiccup I ran into was figuring out some ZFS setting about quota vs refquota (something like that) to have the Time Machine's (artificial) space limit match the ZFS quota so that Time Machine would prune the oldest backups appropriately (otherwise the ZFS snapshots took up an unpredictable amount of space and Time Machine would unexpectedly get out of space errors before hitting its space limit).

n8henrie··on Narco-sub carrying 1.7 tonnes of cocaine seized in Atlantic
To cocaine? By "lose a kid" I assume you mean death and not some euphemism for addiction.
← PreviousPage 2 of 20Next →