mlyle

About 40-45% of US petroleum use is burning it in passenger cars.

We're already a net exporter of oil; if we pushed a bigger fraction of cars and other transport uses to electrification we'd be a massive net exporter.

That wouldn't completely insulate us from global energy markets but it'd do a lot.

I think the scary bit is down the road. We can all use AI well because we have the gist of how things work and a well-honed set of technical empathy for "how they probably wrote it."

I'm not sure how you build that sense as well if you're starting out today, though.

It's just memory mapping the peripheral.

For historic reasons (e.g. 32 bit address spaces, plus the need to reserve the space for multiple pci peripherals) it has been a narrow, movable aperture.

Resizable BAR lets the size of the aperture be chosen (which is usually chosen to allow all of VRAM to fit in and be directly accessible).

Moving 1000W of heat generation takes way less than 1000W.

Hyperscaler PUE (which includes cooling, power conversion, etc) is typically 1.10-1.15, so multiply 1.5% by 1.1.

The person you replied to already mentioned cooling and multiplied by 1.67 to cover ancillary uses.

I think mechanical / electronic intuition are important, and they come from tinkering (especially at younger ages).

Sure, being able to do math and detailed analysis are important. But having a good guess on step 1 is important, too, to narrow the search of huge design spaces.

I think we've all experienced different kinds of engineers. The one you trust to do a month-long trade study with deep mathematical reasoning is usually not the one you trust to just know a plausible path forward. And we need both.

I operate a small fleet at my school:

* 1x Prusa XL Gen 1 with 5 toolheads * 1x Prusa Mk4IS * 1x Bambu Labs X1C with multi-material * 2x Voron 0.2 * 1x Sovol Zero

My experience: the typical higher-end printer runs out of the box and runs hundreds of hours (or more) between maintenance beyond wiping the plate down, occasionally greasing the rails, and blowing little plastic bits out of the enclosure. Every year I end up doing a few hour significant maintenance job on about one of the six printers.

But then there's lemons. That Sovol Zero is terrible and has been down more than it's been up, because of something in the extruder assembly. It also came with a defective camera and needed it replaced. I need to decide whether to rebuild all that with many hours of labor or just get rid of it.

I mean, on the one hand, manufacturing has grown near-continually for those 70 years, and its demand for labor has reduced. So, it's not like you're wrong.

At the same time, our production of lower value, commodity-ish things has indeed shrunk rather than grown with population of demand. That manufacturing expertise is more concentrated more on higher value (aerospace, information technology, etc) things and less on fasteners, bulk materials, or extrusions.

We are not vertically integrated in the way that China is.

Is this really true? Or are we counting like the games kids made in a computer lab somewhere.

Or there's this 87% number, but it's based on "no longer a legal way to buy it," not that it's not archived/preserved: https://gamehistory.org/87percent/

That doesn't really have anything to do with what I said.

I know the P/E isn't completely nuts.

It still prices in revenue growth.

It is possible that nv maintains an absolutely dominant market position and total amount spent on GPUs continues to rise.

It's also possible that either of these things doesn't happen.

I don't think anyone expects nvidia to "sink". We're concerned that competitive pressures and possible reductions from sky-high demand could make it really hard for nv to justify or maintain their current valuation.

That is, revenues may be unusually cyclically high.

It's going to be hard for frontier labs to keep spending at this rate without starting to make money. And even if they do, lower cost alternatives will likely impact at least to -some- extent how much of that revenue nv gets.

It's not like you can allow kids to shoot pellet guns out of a car indefinitely in a metro without substantial risk of something going wrong.

We need to (be able to) trust the police to not escalate irresponsibly and to deal with situations with nuance. Declining to escalate to authorities because we're scared of what they may do is not great.

Sure. 99% of teachers don't know yt-dlp or use things like that, though, so you can expect minor disruptions in a lot of rooms.

But, c'mon, we have complete power outages and still address our curriculum.

Believe me, without the internet we can still teach.

We'll be pretty annoyed that we can't project the video that we think scaffolds today's science lesson best or show the approved choreography for the school play.

And our office staff will be annoyed that we suddenly are all running attendance to the main office old-school.

And students will take a few days to adjust to writing down homework in their planners again.

There's an infinite number of possible math problems, but the things that make these open problems worthwhile is they're interesting to people who have worked in related areas.

They're good to give to new mathematicians, and they're good to help humans understand the shape of the problem space and relative difficulty with the tools we have.

Cheesing these problems with LLMs gets rid of both the training benefit and our ability to create good related problems. There's an aesthetic part of this, too, that LLMs do not capture.

Eating is not a weakness. But a strong requirement to eat is a weakness because you don't have the strength to be happy without eating.

Trying to do 4 Osprey flights per day, with refueling, for spartan resupply... is going to be tough. Sure, it's not impossible...

In such a situation I expect you'd have a lot of other stuff competing with food for that lift, too, like spare parts.

When we had crews that had trained for it and recent operational experience with the capability, it was deemed not worth it.

Trying to stand it up in an emergency now is asking for trouble.

You'd have to be in a really dire situation to risk the carrier and aircraft this way.

I have kids.

I also travel, socialize, work, code, play video games, solve puzzles, etc. Love all that shit.

I'd be sad to "miss" some novels, or to never see the Ocarina remake, etc. Or just the time I spend piddling around with things.

But I think if I was about to die and an advance bought me a few months of moderate quality life, it'd mostly be about seeing my kids cross milestones. I don't imagine I'd spend nearly as much time cranking through those things.

Finding an answer isn't always an objective, though. I agree it often has been too much in baseline schooling.

First, there's a certain amount of baseline knowledge we'd like students to possess. Without a certain prerequisite amount of underlying information committed to memory, it gets far more difficult to achieve fluency in a topic.

But more than that, there's other skills we're trying to build: frustration tolerance, processing contradictory information, disciplined problem solving. You only really develop these skills through productive struggle. If you find a way to shortcut the productive struggle, students truggle.

> but tutors have always made people better students.

Sure. Bloom showed us that students taught with a combination of tutorial and mastery methods, one-on-one, outperform students in a normal classroom by roughly 2 sigma.

The paradox has always been-- why hasn't technology unlocked these gains for students in normal classrooms? If we could boost everyone's performance by this amount, it would be huge for society-- but society can't afford to teach everyone with tutorial methods.

Since the 1970s, we've invested in edtech towards trying to make this happen, but most of it has actually had net-negative effects as best as we can measure. AI, so far, looks to be much worse.

I think part of the answer is that a big part of what makes a conventional classroom work are social pressures. So far, it looks like AI (and edtech in general) does more to dismantle conventional pedagogy and to break down the social fabric of the classroom, than it has improved differentiation or unlocked this tutorial effect more broadly.

The time where "turns" are really great is when a whole lot of what you're doing is a phase accumulator.

I have added tooling so that I can get code listings into / out of the textbook that I'm writing so that I can easily and make sure that they work and produce correct code.

And I'm doing it in LaTeX so I have templates doing some work.

But it's sure nice to just be able to type `make` and see a pdf pop out and green unit tests on the code 3-4 minutes later.

(I have also standardized TODO things to make them easily searchable and show up in a distinctive style in the manuscript).

> It'll be a hard sell for critical infrastructure of any kind

Remember, critical infrastructure attached SCADA systems to the internet and has had trouble removing them all despite 30 years of warnings and audits.

There's definite advantages to having a LLM in the loop. The best we can hope for is that it's architected in a way that it fails safe and damages are limited, so we get some of the advantages but less of the downside.

The way I read it was that he expects large industrial systems with AI to have safeguards to "fail closed."

Which: it seems like we're already getting to the point of capability vs. unexpected behavior that having the ability to "trip" a complex system may be an important design criterion.

I mean, there's plenty of harm these systems can do besides trying to turn us all into paperclips.

Yup. Often one rod fully in is no big deal -- it has been analyzed and you keep operating. But no one has certified to run the reactor with a few rods fully in.

The control systems are designed to err on the side of dropping the rods -- a failure to insert a rod is a lot worse than inserting one accidentally. So now they figure out why this happened while the reactor is in standby.

No one knows the reasons yet.

Some rods dropped. Something went wrong with the rod control system. Maybe an errant trip, maybe something mechanically wore, etc.

When a few rods dropped "accidentally" into the reactor, humans pushed the button to shut it down.

Now they figure that out and then turn the reactor back on.

> Meaning, we should be living a life where there is no manufactured societal stress we need to escape.

I don't think that at all. I think struggle and making stuff better and working hard and feeling ourselves grow is great.

And I wish I had something routine like his picking of berries, because sometimes just feeling existence and our place in it... feels really good. I'm envious of the people who find a routine that does this well for them, while I mostly find it by happenstance.

And I think you do better at the other stuff when you go pick berries.

> We should all be picking berries all the time

I don't think that he wants to pick berries all the time. Or be without his family all the time (even though he prefers being on his own with the berries).

I envy the guy. I wouldn't want to pick berries all the time. But having a routine source of release and tuning out like that which happens to commune with nature and use one's hands...

I was building workbenches for my students last week. A lot of the time was spent outside breaking down pallets and staging things. It was too hot, but I saw the trees sway, the few puffy clouds move, the lizards dart through my path, the squirrels squabble. The few times the wind suddenly picked up felt like heaven. My body ached in new ways. My fingers got into a rhythm and I watched them do things.

I could easily get a job doing work like that. No! I want to be with my students, and I want to work on my hard cerebral projects. But there is a joy to being in nature and to do things with your hands, and there is a rhythm to doing time consuming work, and when we make time for it, it feels really good in a primal way.

> Braking distances exist for AI-driven vehicles just as much as human-driven ones.

Reaction times differ, though, and abilities to blend deceleration and turning and awareness of what else is around.

(Waymo's tendency to be a good defensive driver and bleed a few miles per hour at anything odd help too... though it is worth noting that if Waymo was 1MPH slower, that 6MPH collision wouldn't have happened at all).

Dismissive is appropriate-- "By the time the confirmation renders, the request and its replies will have crossed North America six times"

35ms times 6 is 210ms-- so just about physically possible with current infrastructure. But: this leaves no time for any of the hosts to do anything, or the wifi in the cafe, or whatever.