It's not just "cheap labor", it's also quality, which is absolutely still a problem with "India IT" (both abroad and domestically).

You absolutely can replace H1B-quality work with an AI - at less cost, time, and effort.

The common theme is: they have one competent developer who's a complete slave driver and he instructs an army of incompetents what to do. The incompetents are paid laborer wages and the competent guy (who may be a $300k/year prior SV employee who wants to live in India) is making a comparable ransom vs his local compatriots.

This pattern is ubiquitous with small development shops. "We have a team in India who does great work!" The work isn't great, it isn't fast, but it mostly works, and cheaper than a US team.

However, 1 midlevel US worker can exceed the quality and velocity with agentic development - and the result is maintainable.

It's almost as if nobody's actually read Asimov's books.

The whole point of the "3 laws" wasn't some moral maxim. They were:

1) a clearly, narrowly defined specification, 2) designed to offer a base level constitution which was 3) tautological.

All then in his books then went about telling stories about how those 3 laws were inexplicably and invariably violated. They weren't "oops, here's a minor edge case" violations, either, but foundational problems with broad implications.

Asimov was a science realist. He realized the limits of things and that perfection wasn't a possibility. Arguably, he had more in common with the philosophy of someone like Ted Kazinsky than Arthur C Clark. He would realize today, were he still alive, that something like the 3 laws wouldn't have a meaningful impact on AI based on how their foundational technology works: far more chaotic and nondeterministic than a positronic brain.

I'd have fired her. This is institutionally damaging levels of incompetence, for (apparently!) "clicks".

I don't fundamentally understand the question.

POST/EUFI has fundamentally taken the same amount of time, and has gotten a lot shorter since the days of 640K being enough for everyone. The difference between one board and another is largely negligible at a couple of seconds difference. I'd say it's probably less than 5 seconds on average, maybe 2-3 seconds to the beep, for anything I've owned in the past 20 years. That's perhaps faster than it used to be, due to the advent of SSDs.

The biggest improvement I've seen in boot times themselves has been with the advent of SSDs. Frankly, it hasn't been that impactful, because I don't reboot unless I have to (and have switched almost entirely to Mac for the past decade for anything that matters - booting to Windows for games isn't important anymore, the games all suck).

Where you'll see variance is when you add different, additional ROMs, or you're booting a server. I've seen time to post take as much as 10+ minutes (or so it seems) on various server hardware - IBM has always been a horrible offender in this regard. Adding a storage card with its own BIOS (like LSI MPTSAS cards) take a long time. Supermicro systems take a mercifully shorter time. I'm sure video cards and other add-on cards have an impact on this on laptops and personal PCs, the system has to initiate the devices and every device takes a little more time.

Unless you've suffered through rebooting an IBM system accidentally during production hours, you don't know pain. If you've never had to diagnose such hardware, standing in the hot aisle as you repeatedly power cycle the thing, you don't know suffering.

This is why you shouldn't write anything in javascript or python, anymore.

It's not sustainable because the languages are not stable. The only thing that makes it tenable is being able to throw LLMs at it to keep up.

And at that point, just use a more stable, performant, useful language like go, C/C++, or rust.

The usability of Linux as a desktop has markedly degraded in the past decade, largely due to the efforts of the Wayland team. It keeps getting more bloated and less useful. It's a travesty.

Where do you think Apple makes its products, Wonderland?

Meanwhile, there are quite a few verifiable good coders - eg. Eric S. Raymond, I'm sure you know of him - who are doing quite well with 'vibe coding' and producing quality code.

As someone who's spent most of his career reviewing others code related to performance and race conditions in C and Java, the biggest issues I've found with code produced by frontier models is that you still need to know what you're doing to tell the model what to do. The code itself is as good as any I've reviewed, and often better.

My son is literally building / has built a productivity/objective-oriented time management tool for himself with vibe coding. He's a kid. It works and he's using it daily. I dare say it's not a lie nor a hallucination.

The biggest problem with this is that 12TF cards are expensive, but more importantly, the ones with enough memory for the models to be much more than a curiosity (say, 32GB+) are extremely expensive, and frankly, not terribly available.

The state of the technology needs to improve a lot for general local inference. Yes, you can do a lot with smaller specialized models (things like whisper) or do minor things with free models like lfm2.5 or mistral but these aren't even in the same ballpark

So if you need a thinking/reasoning model for one task, and you need (say) a coding specialist model like qwen2-coder-next, you're looking at needing frontier models (reasoning) and a lot of memory both. It's the reasoning that's particularly expensive to pull off, and still effectively impossible for anything short of enterprise scale.

That's probably because you're going from a cryptic and niche language with precious little documentation to a verbose and often poorly implemented language. People who like R aren't known for writing good docs or uncomplicated code. lol

You will get far better results going from a properly documented (as in, code comments) codebase to go.

Thanks for that.

I know a lot of older guys my age who've had good luck with bp157 but it's got a pretty steep price. This seems perhaps cheaper.

There isn't a meaningful difference in the industrial requirements for making DDR6 vs DDR4 that are being impacted by AI. The real crunch is in the wafers, which HBM uses as well. That's what's driving up the prices.

To add to this, the tech which would be used to fund the automation is currently making the vices and tech used to placate the UBI masses entirely unaffordable.

Yang has been consistently wrong - seemingly intentionally so - due to his personal ideals:

- Grossly understating the cost of UBI
- Strongly opposes automation due to job loss which was actually caused by offshoring and deindustrialization
- Claimed that Amazon pays zero in taxes (which, while at least directionally correct, also false).
- Saying UBI would create jobs (zero basis in what has happened, just feels)
- Consistently overstates how many jobs have been lost due to automation, because UBI

AI is no different. It fits his narrative and how he sees the world. Put an elephant's ass in front of a man who's only ever seen a horse, and he'll think it's a fat horse.

I actually do have shoulder problems, from a combination of a motorcycle accident and regular jujitsu. They told me I needed to reconstruct the shoulder. I put it off, and started combat sports instead... my shoulder no longer hurts (generally) and I've got (mostly) full range of motion and strength.

In contrast, I suffered with lower back pain for years, and it got to the point where I was immobilized. MRI scans showed nothing.

The problem in most cases is usually muscular-skeletal interaction. Not the bones, not the ligaments, not the muscles - but how they all interact.

Most modern medical science seems to be largely oriented towards the Monty Python orientation: if their machine goes ping, it's a confirmation. They're not very good at multiple possible causes and because they're so specialized they're not trained in thinking big picture. It makes them absolutely horrible at making informed decisions.

But hospitals like to print money, so they're more than happy to do a rotator surgery or throw a couple screws in the back of someone who's sedentary or muscularly unbalanced and get them addicted to pain pill and a lifetime of being unable to move. Repeat customers are good customers.