HNHacker News
TopNewBestAskShowJobs

doe_eyes

1,230 karma · joined June 9, 2024

submissionscomments
doe_eyes··on Image Editing with Gaussian Splatting
There's a ton of room for using today's ML techniques to greatly simplify photo editing. The problem is, these are not billion dollar ideas. You're not gonna raise a lot of money at crazy valuations by proposing to build a tool for relighting scenes or removing unwanted to objects from a photo. Especially since there is a good chance that Google, Apple, or Adobe are going to just borrow your idea if it pans out.

On the other hand, you can raise a lot of money if you promise to render an entire industry or an entire class of human labor obsolete.

The end result is that far fewer people are working on ML-based dust or noise removal than on tools that are generating made-up images or videos from scratch.

doe_eyes··on You Better Shop Around: Litigant Characteristics and Supreme Court Support [pdf]
Actual PDF here: https://elizabethalane.com/wp-content/uploads/2024/08/d08a8-...

Note that the summary is somewhat misleading. This is about the public perception of cases, not the likelihood that the SCOTUS will reach a particular decision.

The study also tries to imply that SCOTUS cares about the public perception of its decisions, and it probably does - but this isn't something they prove. Instead, the study deals with the sentiment across focus groups.

doe_eyes··on AI tool cuts unexpected deaths in hospital by 26%, Canadian study finds
But I think that's a wrong way to look at it. Or rather, it posits that we're at a point we truly consider good enough independent of cost.

It's entirely possible that we want better healthcare outcomes - all the historical trends point to that - but that we're more or less out of ideas how to get there on the cheap. This might be a new possibility.

In your model, why do we get improved, costlier insulin if the old thing was good enough? Because we actually want to pay more if it works better, and it doesn't mean we cut something else to make up for it. You just pay more in taxes in a subsidized model, or pay more at the pharmacy with private healthcare. There's a drug manufacturer profit motive in there, but it holds true in the added-cost ML scenario too.

doe_eyes··on AI tool cuts unexpected deaths in hospital by 26%, Canadian study finds
I think you're missing an important part of the equation: it's outcome quality per amount paid. If you could have gotten 20% better results but it would mean tripling the costs of healthcare because we'd need to hire a lot more staff, perhaps we felt that was a bad deal.

If you can get 20% by paying... what, presumably <5% more for some ML tool that double-checks stuff and flags risky stuff... perhaps it's something we want to do.

doe_eyes··on Why wordfreq will not be updated
> I agree in general but the web was already polluted by Google's unwritten SEO rules. Single-sentence paragraphs, multiple keyword repetitions and focus on "indexability" instead of readability, made the web a less than ideal source for such analysis long before LLMs.

Blog spam was generally written by humans. While it sucked for other reasons, it seemed fine for measuring basic word frequencies in human-written text. The frequencies are probably biased in some ways, but this is true for most text. A textbook on carburetor maintenance is going to have the word "carburetor" at way above the baseline. As long as you have a healthy mix of varied books, news articles, and blogs, you're fine.

In contrast, LLM content is just a serpent eating its own tail - you're trying to build a statistical model of word distribution off the output of a (more sophisticated) model of word distribution.

doe_eyes··on Bluesky Reaches 10M Accounts
I think that's a reductionist view. Platforms like Twitter are pretty unique in that unlike most other social media, you don't have well-defined communities organized around topics or locations.

On Reddit or Nextdoor, you join to discuss specific interests, and that serves as "social glue" that lets you make new acquittances along the way. The same goes for HN, by the way. But on Twitter / Bluesky / Threads / Mastodon, you probably don't have anything in common with another user picked at random.

The most common (non-commercial) motivation for joining is probably to just hang out with the people you already know from elsewhere, and then expand your social network very gradually and conservatively, mostly based on who they know. So it doesn't matter that there's 10M strangers over there if none of your buddies use it.

doe_eyes··on How to succeed in MrBeast production (Leaked PDF)
> The whole modern social media / influencer sphere seems like a huge bubble that will pop eventually.

Before teenagers were looking up to YouTubers, they were looking up to TV celebs, musicians, sports players, and so on. You had entire publishing empires built around following such celebs around and reporting on their private lives.

I don't think this is hugely different. The tech has evolved and the formulas have been perfected, but it's still catering to the same obsessions and urges that we had for a good while.

doe_eyes··on Notes on OpenAI's new o1 chain-of-thought models
Alignment started as a fairly nifty idea, but you can't meaningfully test for it. We don't have the tools to understand the internals of an LLM.

So yes, it morphed into the second best thing, brand safety - "don't say racist / anti-vax stuff so that we don't get bad press or get in trouble with the regulators".

doe_eyes··on Cruise ships chopped in half are a license to print money
Yeah, I went on a cruise once, and the whole idea is that you don't spend time in your cabin. It's not a train. You have cinema screens, live performances, restaurants and bars, libraries, quiet lounge areas, pools, and so on.

Plus, there are sightseeing opportunities on land, and the neat thing about cruise ships is that they dock where the action is. Airports are always on the outskirts; ports tend to be situated pretty centrally in most cities.

Honestly, it's probably the nicest way to travel to faraway places, short of a private jet. It's not for everyone, but it's not a dystopian experience. The ships carry insane numbers of passengers, but they are also pretty darn spacious.

doe_eyes··on Cruise ships chopped in half are a license to print money
For the most part, it doesn't matter how long it takes to pay this back. They borrow the money, pay installments, and still end up with a 10-15% profit per passenger from day one (and that's after deducting deprecation).

But because they live on credit, they were pretty badly hosed during COVID-19.

doe_eyes··on Google says replacing C/C++ in firmware with Rust is easy
Inertia is doing things because you're used to doing them. This is pragmatism: until the arrival of mainstream Rust-based OSes, there's probably no escaping C/C++.
doe_eyes··on Google says replacing C/C++ in firmware with Rust is easy
C is 52 years old; C++ is nearly 40. That's more than a generation. Yet, I'd bet that there are more up-and-coming developers learning C/C++ than Rust today.

The problem isn't just personal preference or inertia. It's that all mainstream OSes, libraries, and system software are written in C/C++. You can't exactly ignore this and not learn the language if you want to work in this space.

Of course, there are many developers who don't work with system-level stuff, and they may be perfectly fine just knowing Java or JavaScript, but Rust is not positioning itself as a direct competitor to that.

doe_eyes··on Death Rates by Vehicle Model
The highest-ranking car is Ram 3500 crew cab long bed. It's over 14 foot long and weighs around 8,000 lbs. It's closer to colliding with a bus than with a sedan.

But it's also not a truck people buy as a suburban fashion statement. They usually go for smaller pickups, like F-150 with a short bed.

doe_eyes··on Ilya Sutskever's SSI Inc raises $1B
No, sometimes it's about an IPO. In which case, if we're being cynical, the exit is funded by your 401k.

But yeah, VCs generally aren't about building profitable companies, because there's more profit to be made - and sooner - if you bootstrap and sell.

doe_eyes··on Solar will get too cheap to connect to the power grid
Why is it supposed to be an investment? If it's sufficiently cheap, it becomes something that you buy because it's useful to you, not because the government is promising you subsidies for life.

Also, the other driving force here is the cost of grid-delivered energy. California is sure working hard on incentivizing solar + storage that way...

doe_eyes··on American Radio Relay League paid $1M ransom payment
Their insurer paid most of it.
doe_eyes··on Is My Blue Your Blue?
I suspect it tests your monitor and monitor calibration as much as your color perception. In particular, sRGB displays have a pretty severely limited green gamut. If you have a wide-gamut display, the test is probably gonna appear different.

But another problem is with displaying the colors essentially full-window, which is going to be nearly-full-screen for many users. When we're staring at a screen with a particular tint, our eyes quickly do "auto white balance" that skews the results. It's the mechanism behind a bunch of optical illusions.

To address that last problem, I think the color display area should be much smaller, or you should be shown all hues at once and asked to position a cut-off point.

doe_eyes··on Buy, Borrow, Die – Explained
Nah. Or rather, I don't doubt that some billionaires do, but they're not the target audience. First, not many billionaires spend time on Reddit. Second, if you're Warren Buffett and have obscene net worth but minimal living expenses, there's really no point. It doesn't matter if you're spending .1% or .125%. If you do borrow against your assets, it's usually to invest, not pay the bills.

People preoccupied with strategies like that are the ones who are burning through their net worth at a much higher rate, so the difference in tax burden actually makes a difference. They're usually just wealthy enough to have a "wealth advisor". Doctors, successful techies, business owners, etc.

doe_eyes··on Buy, Borrow, Die – Explained
This kind of an explanation overlooks the obvious issue: you're exposing yourself to asset valuation risk. We don't tax unrealized gains for that exact reason.

Let's say you have $10M in in index funds. You don't want to cash out and pay capital gains, so you get a $5M credit line with your stock as a collateral. Then, there's a market crash, your collateral is all of sudden worth just $4M, and you have a bank knocking on your door - but you already spent $5M on a McMansion. Now what?

You can opt for "risk-free" assets, but then, if they're truly risk-free, they're probably not appreciating in a way that would make this tax gamble worth your time.

doe_eyes··on Judges rule Big Tech's free ride on Section 230 is over
I think you're looking for the kind of precision that just doesn't exist in the legal system. It will almost certainly hinge on intent and the extent to which your actions actually stifle legitimate speech.

I imagine that getting rid of spam wouldn't meet the bar, and neither would enforcing that conversations are on-topic. But if you're removing and demoting posts because they express views you disagree with, you're implicitly endorsing the opinions expressed in the posts you allow to stay up, and therefore are exercising editorial control.

I think the lesson here is: either keep your communities small so that you can comfortably reason about the content that's up there, or don't play the thought police. The only weird aspect of this is that you have courts saying one thing, but then the government breathing down your neck and demanding that you go after misinformation.

doe_eyes··on Electrostatic motors reach the macro scale
The sustainability angle is a bit suspect, although of course it's smart of them to make that pitch. Many large motors don't use permanent magnets, and even for the ones that do, neodymium isn't all that rare and is mined in the US. Most of our copper comes from friendly sources too (US, Chile, Peru, Mexico, Australia). When we talk about sustainability for EVs, I think the main concern is batteries, not motors.

But the most interesting (and problematic!) part of their design is the use of a dielectric liquid to increase field strength. They don't give any specifics, but reliability and weight issues aside, I'd imagine that drag-related losses would get significant at high RPM. Maybe the point is to go slow?

doe_eyes··on The semantic web is now widely adopted
The main problem is that the incentive for well-intentioned people to add detailed and accurate metadata is much lower than the incentive for SEO dudes to abuse the system if the metadata is used for anything of consequence. There's a reason why search engines that trusted website metadata went extinct.

That's the whole benefit of using LLMs for categorization: they work for you, not for the SEO guy... well, prompt injection tricks aside.

doe_eyes··on The bizarre siege behind Stockholm Syndrome
Above all, people often stay in abusive relationships because of what they think is a rational analysis of the trade-offs. There are concerns about financial stability, uprooting your children's lives, losing friends, etc.

In reality, the calculation tends to underestimate the risk of staying and overestimate the risk of leaving. But this is not unique to the victims of domestic abuse. It's the same reason people stick to bad jobs, depressing but non-abusive relationships, etc.

doe_eyes··on Parents outraged at Snoo after smart bassinet company charges fee to rock crib
On the cutting edge: I've seen techies using ChatGPT to make up and read bedtime stories to toddlers, or to tutor them during the day, so that the parent can stay focused on their career. I don't know that it's harmful, but it seems messed up. I wonder if a kid raised this way will end up tripping LLM text detectors for life...

The lower-tech equivalent of this is a toddler being raised by a tablet or a smartphone. Not necessarily because the content is harmful, but because it's neglect - just like leaving a kid in front of a TV back in my day.

doe_eyes··on Parents outraged at Snoo after smart bassinet company charges fee to rock crib
Some infants cry a lot. Not every instance is a magical bonding opportunity, especially since you still need to get stuff done.

There are high-tech child-rearing trends that I suspect might be doing harm, but a rocking bassinet probably isn't one.

doe_eyes··on AI stole my job and my work, and the boss didn't know – or care
The problem is that many of what you describe as "terrible" jobs are just a low-overhead way for creative folks to pay their bills. You don't make a living writing timeless poems or painting masterpieces, at least not usually. You make a living doing copyediting or painting commissioned artwork - and then work on achieving greatness in your free time.

AI doesn't have to be good to destroy this income. This is making it harder for creativity to thrive. Relatively few budding artists can count on private patronage or government sponsorship, and both of these come with plenty of strings attached. And pie-in-the-sky ideas like universal income are not going to wish the problem away.

doe_eyes··on Does Reasoning Emerge? Probabilities of Causation in Large Language Models
> Maybe: there is no measurable difference between 'real' reasoning, and 'fake' reasoning.

The point is, it's easier to teach an LLM to fake it than to make it - for example, they get good at answering questions that overlap with their training data set long before they start generalizing.

So on some epistemological level, your point is worth pondering; but more simply, it actually matters if an LLM has learned to game a benchmark vs approximate human cognition. If it's the former, it might fail in weird ways when we least expect it.

doe_eyes··on Does Reasoning Emerge? Probabilities of Causation in Large Language Models
This is proposed as a way to measure "true" reasoning by asking a certain type of trick questions, but I don't quite see how this could be a basis of a sustainable benchmark.

If this gets attention, the next generation of LLMs will be trained on this paper, and then fine-tuned by using this exact form of questions to appear strong on this benchmark, and... we're back to square one.

doe_eyes··on Nationalpublicdata.com hack exposes a nation's data
Nothing is stopping wealthy and populous states, such as California, from rolling out their own smartcard IDs. This doesn't need to happen on the federal level.

The problem is that harmonized, mandatory IDs are "problematic" to the left side of the political spectrum too. Undocumented immigration, voting-related wedge issues, etc.

doe_eyes··on Nationalpublicdata.com hack exposes a nation's data
This comes up regularly, but... is it actually true that SSN still functions as a sole authentication mechanism? I don't recall a single instance in the past decade where I could open a bank account, apply for credit, or do anything else of consequence based on an SSN alone. Well, maybe open an utility account, but that's hardly consequential, they just use your SSN to decide if you're a deadbeat who needs to make a deposit before they hook you up.

In other cases, there's always either some secondary document check, or you need to pass some online quiz about your credit history. These quizzes are not bulletproof, but require the attacker to have a more comprehensive dossier on you.

← PreviousPage 2 of 6Next →