HN user
Sort of, but isn't the focus on prompts a bit myopic? The huge difference between earlier GPTs and ChatGPT was RLHF, which not only makes it better at following prompts, but also enforces a lot of hidden dogma. It certainly influences how ChatGPT talks about climate change or AI risks, for example.
This is actually a very lucid way to frame this. People love to complain that "Google is useless now," but it's pretty clearly not the case if you look at how most people use search.
What they usually mean is "nobody can find my interesting hobby projects and I can't find theirs." And that definitely tracks. As a person who poured a lot of energy into completely free, non-commercial educational content, it grinds my gears that there are 2-3 pages of derivative blogspam peppered with affiliate links - and increasingly, LLM-generated drivel - ahead of me.
What I think we get wrong is demanding that others fix it for us, though. Yeah, it's the cool part of the internet, but it's a commercially insignificant one. What the article is trying to do - pick a specific practical solution and lead by example - is probably better. Even if it's a rehash of what we tried in the pre-Google days.
Huh? Car maintenance is a rational, physical necessity. I don't need to compliment my car for it to start on a cold day. I'd like it to stay this way.
Having to be unconditionally nice to computers is extremely creepy in part because it conditions us to be submissive - or else.
This is such a weird topic in the ham community. The reason this restriction exists has nothing to do with the retro-justifications used by the community.
For a long time, the US government genuinely feared that ham radio would be used for espionage. It had listening stations across the nation to monitor all communications. It flat out shut down the entire service (!) during WWII. And it came up with the idea that you have to communicate in the open, and that no form of obfuscation or encryption is permissible.
And then hams came up with this roundabout explanation that actually, it's good that you can't have privacy. No matter that it holds back a hobby that is by all usage metrics dying, and that there are many countries where encryption is allowed and doesn't lead to any terrible outcomes.
Privacy is useful in hobby uses. Maybe you want to talk to your spouse without a nosy neighbor listening. Maybe you want to periodically beacon your GPS location without the whole world knowing. There are so many cool things you can do, and there is spectrum that is... quite frankly, largely dead right now, and if you don't encourage new uses, it will be reclaimed by the government.
It's literally the playbook of every single business. Database vendors and graphics card manufacturers tout studies that show their product performs better. Pharmaceutical companies pay researchers who conduct studies that show their stuff works. If Safeway wants a zoning variance to open a new store, they will pay for an environmental impact study that says it's fine.
I'm not saying this to convince you should trust this study. But I think it's important to recognize it's absolutely happening everywhere, not just in the industries we don't like. Most of the research we read was paid for, and an overwhelming majority of it reaches the conclusion that aligns with the views of the researchers or of whoever is footing the bill.
Yes, it probably applies to the most qualified hires, so probably folks who are already around $1M total comp where they are.
But the most important point is, you don't get that money now, and maybe never. The most likely scenario where you get that kind of a payday is if they go public and if the pre-IPO pie-in-the-sky valuation holds for a good while.
It might. But it's far more of a gamble than your FAANG stock.
The work culture isn't putting kids in front of computer screens and discouraging outdoor activities.
It's not even something you can control as a parent. My kid, 16 years old, walks to school on his own - but he's always back right after classes. And it's not that he's an outcast. That's just how they roll in the SF Bay Area. Other kids go home too, or they are shuttled by their parents to some organized after-school activities.
You're not describing a position taken by sane libertarians, though. Their argument is different: that if the government didn't regulate so much, customers would depend on the merchant's reputation, possibly backed by independent testing done by private sector institutions. A modern-day parallel would be electrical safety. In the US, this is largely handled by private organizations such as the UL. You can buy a non-UL extension cord or a toaster if you want.
And look, I'm not arguing that this is a better solution. But I think it makes sense to attack the strongest version of that argument, not the weakest one.
I'd go a bit further with this claim. Most of what's being done in this space is about inventing new retro aesthetics, not about faithfully approximating how things worked in the 1980s and 1990s. For example, color TVs of that era didn't really have pronounced scanlines. They also didn't have thick, lightly-colored, reflective bezels.
I get that it looks cool and makes old games more aesthetically pleasing. But the reality is that we liked these visuals back then because we had much lower standards, not because CRTs had some magical properties that made the games look awesome.
I see the problem a bit differently. Perhaps a society could function in a model like that. But we decided it's too burdensome and we ceded this responsibility to the government. In such a world, it is paradoxically more dangerous to come across some under-regulated niche, because our default assumption is that the government took care of the risk. It doesn't even cross your mind that you should be asking about lead in your turmeric.
This is sort of what happened with welfare too. For a long time, we depended on private charities to take care of the less fortunate. We decided the system sucked, so we established a government-operated safety net. But in this reality, it can be worse if you slip through the cracks of government programs. People around you by and large no longer think it's their duty to help.
Anyway, I wouldn't write it off as extremism. It's just we need to pick an option and stick to it. In a "nanny-state" world, you can't decide that you're not going to regulate food safety anymore and hope that the market will sort it out.
Well, HN isn't a forum with a well-defined "expert" scope. It's a link aggregator for nerdy news. If you get bored with Rust, you talk about ChatGPT. Or CPU design. Or biotech. Almost anything goes.
I think this makes it immune to the patterns described in the article, although there are other ways for such communities to die. The decline of Slashdot is a cautionary tale.
In practice, it's a tiny market. CRTs are bulky, unsightly... and for the most part, really not as amazing as claimed. Plus, they are difficult to ship. Most of them end up in the landfill.
Meanwhile, antique phones, typewriters, cameras, and so on are popular enough that people make replicas: https://www.ebay.com/itm/266280860021
That's gonna be tough with phonetics across several thousand languages... not to mention regional slang and other subculture stuff. I think the surest bet is just to avoid short names. If it's four or five letters, the likelihood of a collision is high. Stuff like Josephine, Gabriella, or Nathaniel is probably pretty safe.
Most of these are commonly shortened and then you're back in the danger zone, but then, at least you have the option of reverting to the long form without jumping through any legal hoops.
None of this is a problem in any objective sense. It's just that if your goal is to use one name across two languages, it's not exactly what you get in this scenario.
Stuff like that doesn't bother me at all, but I bumped into quite a few immigrants who had strong preferences one way or the other.
This is actually a pretty interesting problem and the website doesn't do it full justice.
Do you want the same spelling? That's easy, but the pronunciation is quite often completely different. A good example is Jules in French vs English. In this scenario, you're effectively going by two differently-pronounced names in all face-to-face interactions, not that different from the folks from China or India who are adopting "Westernized" names abroad. The only perk is that you might not have to spell it out over the phone.
Do you want the same pronunciation? This is also fairly easy in many languages, but the spelling is likely to differ. An example of this might be Hannah versus Hana (English / Czech). This option makes verbal communications easy, but may confuse people who are trying to read your name out loud or to write it down - so any interactions with customer service are going to be mildly annoying.
Do you want both? For most languages, the list will be extremely short, perhaps half a dozen names such as "Anna". If you don't fall in love in one of these options, tough luck.
There is also a softer version of this goal: have a name that isn't native in the second language, but that is easy to spell and pronounce. For most people, this is probably the best compromise. It lets you keep your national identity, doesn't limit your choices too much, and minimizes friction.
Right. It's not even that they need to create a market. These things are already selling on eBay for hundreds of dollars. I guess it's a fairly common decor element for offices and homes. The same goes for old typewriters.
There are obsolete electronics that would be a lot harder to sell, for example CRT TVs - but phones are basically free money.
The desire to disrupt almost always arises out of ignorance. I don't mean that in a bad way. It's just that if you know the complex reasoning and all the institutional baggage that explains why hotels, banks, or old-school tech companies operate in a particular way, it's hard to say "let's blow it all up."
On the flip side, if you can explain it away as "they just don't get it and I do," it's a lot easier to act. And the thing is, sometimes, you get good results. Sometimes, the old way of doing business is just a matter of inertia, and the justifications used by others turn out to be bad.
But about just as often, you end up reinventing the wheel or re-learning the lessons that others learned before.
Few people write blog posts to say "I worked at Foo Corp and it was okay." There's a strong bias toward horror stories. There's also a HN selection bias: a story about FAANG is a lot more likely to be upvoted than a story about some small startup.
I have friends in all the big tech companies and I don't think any of them is fundamentally better. They all have their idiosyncrasies, they all have good teams and terrible ones, they all have a ton of red tape.
I think the main problem is that they built their reputation by claiming they invented a new way of doing business and a new way of treating their employees. They were supposed to be the antidote to the corporate culture at Microsoft, IBM, Sun, and whatnot. But ultimately, they converged on pretty much the same.
Most large tech companies essentially give a pass to new hires on their first eval cycle. Depending on your start date, you might be either outright ineligible for a rating, or the rating defaults to "meeting expectations," because the expectations for the first 1-2 months are for you to just learn the stack and get to know the team. If you hit the ground running, good for you, but your manager might face an uphill battle to justify anything other than "meets."
For the next 1-2 months, you can probably keep making excuses. If your manager is paying close attention to new hires, they might object. But many managers are overstretched. Between office politics, planning, and all the ongoing "problem cases," they might simply not have enough cycles to watch your output real close.
Even after your manager is fed up, it takes time to fire you. In part to avoid legal risks, HR typically wants to see a written plan first, giving you about three months to prove yourself. If you do nothing, that's usually the end of the road. But if you lift a finger and earn a passing grade, the timer essentially restarts. In fact, you're now your manager's success story, and they might be reluctant to admit they were wrong.
And that's on average teams. If you end up on a dysfunctional team or on a project in turmoil, you might not even have to pretend. There's just no one who is close enough to your role and still cares about the result.
I think this is a weird take. He found his groove, had a strong popular following, and enriched the lives of two generations of gamers. This is more of a positive legacy than most of us can hope for cranking out JavaScript for advertising companies throughout our prime years.
And as others note, he has hobbies and moderately successful side projects outside this envelope.
Folks here are probably reading too much into the "boot sector" angle. This project, like many others on HN, is best understood as "doing something in a constrained way because it's a fun challenge." As to why 512 bytes, the honest answer is "because it's a round number with some vague retro connotations."
There was never a constraint on x86 that you had to fit any real functionality into a single disk sector. All the bootsector code ever did was setting up the registers and calling BIOS to load more data from disk - and the only reason you had this done in stages was so that BIOS wouldn't have to know the particulars of your OS, such as where to find the kernel or what address to put it at.
For what it's worth, the pre-boot BIOS environment was always quite featured, offering text and VGA graphics, disk and keyboard handling, and so forth. In fact, when you look at something like MS-DOS, it was actually the BIOS-side code that was doing a lot of the heavy lifting. MS-DOS was halfway between a shell and a "real" OS as it is understood today.
Nowadays, with UEFI, we essentially decided to make the initial code executed by the CPU an operating system in itself, with filesystem support and so forth - and the notion of a 512 byte first-stage bootloader is largely gone.
They were also on the forefront of the "blockchain revolution" several years back.
There's worthwhile research coming out of IBM once in a blue moon, but by and large, they seem to be chasing PR trends.
A single retail account is of no consequence to a national bank that has tens of millions of customers. It doesn't matter if you're spending $10 or $50,000 a month. They care about you as an individual about as much as Google cares about a single Gmail account. If an algorithm flags you as potential trouble, off you go.
If you want better service, there are two ways. One is to be an ultra high net worth individual. But the threshold for that is pretty high - from what I recall, real private banking nowadays starts around $5M to 10M in assets with a single institution. The term "millionaire" doesn't mean as much as it did in the 1990s.
The other option is to go with a small local bank or a credit union. If you're in a CU with 10,000 members, it's a lot easier to resolve problems or discuss unique circumstances.
How does this interact with the R&D tax credit? Historically, companies actively sought to classify a portion of software engineering expenses as research, because you received pretty generous tax rebates for that in addition to this naturally offsetting your income:
https://www.adp.com/resources/articles-and-insights/articles...
This looks like an attempt to reduce that second part without touching the first, right? So effectively an administrative action to reduce the R&D tax credit passed by the Congress a while back?
1. I’ll only be a millionaire because of inflation over my lifetime.
First, statistically speaking, you might only have a couple of years between statutory retirement age and some sort of an "old-age" illness that will diminish your ability to pursue hobbies, travel, and so on. All the stuff you've been putting off for later... you might not be able to do, and you will regret that.
Retirement savings will allow you to retire earlier, and they will allow you to maintain higher "velocity" after - because many of these things we're putting off do cost money to pursue.
Just as importantly, though, savings buy you peace of mind. You're no longer as stressed about day-to-day challenges, like conflicts at work, an uncertain economic outlook, or a leaky roof.
Plus, on a typical tech salary, you can certainly hit 1-2 million in retirement savings in today's money, and there are investment vehicles that offset inflation with relatively modest risk.
2. Social Security is less than a decade from insolvency,
People have been saying this for the past four decades. But even if it does go insolvent, do you expect government retirement benefits to suddenly become a lot more comfortable than they currently are?
3. Saving for retirement is a luxury reserved for the folks who don’t live paycheck to paycheck. For those who do, would their savings even be enough to matter?
If you live paycheck-to-paycheck, you can't save, by definition. The question is whether you need to live paycheck-to-paycheck. For most people on HN, the answer is a resounding "no."
4. Most pensions and retirement savings accounts use the stock market to beat inflation. Maybe it’s just me, but I am extremely bearish on the stock market for the next four decades.
You don't need to "beat" inflation. You just need to match it. People invest into risky stocks because they want their savings to grow faster than inflation, but you certainly don't have to. You can buy corporate or municipal bonds, you can bet on commodities, you can buy real estate, you can go for collectibles, you can buy gold bars if you want to. You can invest in a friend's business.
Finally, the whole talk of "the market" is a distraction. Unless you're buying index funds, you can choose specific companies that you're not bearish on. They might not beat market returns in a good year, but if your goal is to have confidence in your choices, it's a sound plan.
I would add: don't do that in KiCad. Although they added the functionality, it's clear that they never really QA'd it. It's profoundly broken. Many simple shapes are impossible to achieve, adjustments are painful and often require redrawing the entire track, selection boxes are not calculated correctly so you end up accidentally selecting curved tracks that are inches away...
If you want to design boards with curved traces, paying several hundred bucks for DipTrace is almost certainly a better route.
Are you looking at the same thing? There clearly are decoupling caps not far from the chip. There's no rule to say you need to use 10 x 100 nF capacitors in 0603. They are more cost-efficient, but they're more work to solder by hand.
The DC-DC module also seems like a sensible choice for hand assembly. These things cost about a buck or two these days.
This board is clearly meant to display the electronic components in plain view and be easy to solder at home.
This is really not about an engineer keeping a bootleg model in your basement. It's about the barrier for entry for commercial products. Or the ability to curate improved open-source implementations in the long haul, for that matter - as past a certain scale, this entails creating a non-profit of some sort to pay your bills.
Plus, while it's definitely the case that with sustained interest, old data tends to linger around... the moment the interest wanes, it's gone. I've been on the internet for a while and there are so many hobby sites, forums, and software projects from the early days that are simply gone for good (and not on archive.org).
Why? The ChatGPT interface is not integrated with an editor and not really tailored for writing. I'd wager that most of the current Grammarly users would rather pay for a tailored tool than a chat window that excels at schooling you and refusing to do anything that doesn't align with OpenAI's sprawling brand safety rules.
Of course, GPT-4 can be molded into an editing-centric companion, but I bet Grammarly is already working on that, and they might end up paying OpenAI for the technology.
Brand recognition means that it's a lot easier for Grammarly to build that product than for a random third party to break through.
There's a lot of people with strong opinions whether these detectors can or cannot work, but I implore you to do the experiment I just did. Go to the top 10 you find in a Google search and paste a sufficiently long sample of your prose. Not a random HN comment - at least 200-400 words of normal, coherent text.
It's a game of cat and mouse in the sense that you can build LLMs specifically optimized for evading the current crop of detectors, but in my testing, they work pretty darn well in the general case. While they might not reliably pick on all LLM text, and while there's sometimes a couple of words in human-generated writing that causes them to output a low but non-zero probability of LLM content, they do not rate human-generated text as "99% AI". Especially not across multiple writing samples.
The most likely story here, I suspect, is that the person leaned on LLMs for commissioned writing and is now trying to save face. The secrecy of the models works both ways, right? And frankly - how often do you see people in HN, or people who do commissioned writing, admit in private that they're using ChatGPT? It's cropping up all over the place.
Note that I'm not commenting on the ethics, fairness, or transparency of tools like that. I'm just saying they work far better than you might be suspecting.