foldr

The quoted section makes sense to me. It’s like video decoding: the more you can do in hardware, and the fewer user mode / kernel mode switches, the less CPU you use and the less power you consume.

Like the sibling, I can't really agree with this. Google is now quite good at answering general queries and doing simple calculations, which is a big step up from its capabilities a few years ago. For example, you can ask

> What is the equivalent focal length on APS-C of 80mm on full frame?

and get the correct answer. It's not good compared to frontier AI models, but that's to be expected, given that you don't pay for it.

That's a bit of a false dichotomy, as you could pay ~$200 to get them assembled by one of the Chinese PCBA services.

I’ve found that PCBWay will do these kinds of checks, and I’ve had no problem communicating with them in English for basic stuff (contrast sibling comment about JLCPCB). Anyway, English <-> Chinese machine translation is getting so good that I can’t see this being much of a blocker for long.

A lot of it is moving small objects precisely from point A to point B while integrating expert domain knowledge and general awareness of what’s going on. This stuff can be automated in principle, but with diminishing returns after a point.

Agree with this. I used to do my own assembly almost as a point of pride, but the Chinese shops will do it so cheap now that you are really just torturing yourself to no purpose. (It’s fun the first few times, but gets old pretty quick.)

That should never cause a crash. You don't let one train go onto a section of track unless you know that there's no other train on that section of track. If train A has failed to move when instructed, then train B waits for it to move.

Aside from running local LLMs (which is tight on 32GB anyway), what on Earth are people doing that requires 32GB as a bare minimum? I have 24GB on my current MacBook Pro and never find myself limited by it. My work MacBook Pro has 16GB, and that is occasionally a bit tight. However, that's largely because the stingy 512GB SSD quickly fills up, leaving no room for swap, not because I really need more RAM as such.

That said, I don't believe that SSDs can feel pain, so I am quite happy for the OS to thrash them as much as necessary. The fear of SSD failure is massively overblown IMO.

In terms of performance, the difference is active cooling. That will matter to people who are regularly doing sustained CPU or GPU intensive workloads.

I think this is drawing the wrong conclusion. Humans can write assembly code too. The downside for a human is a greater time investment. For an AI agent, it's greater token usage. So, sure, you can get your agent to write assembly code directly, but just as for a human coder, all you're achieving is expending more resources for little or no benefit. You're burning tokens on problems (e.g. register allocation, stack manipulation) that can be solved more reliably with vastly less computation.

> the high taxes of Europe

Taxes in the UK are quite low by European standards. The UK's tax revenue to GDP ratio is similar to Canada's (either a bit lower or a bit higher, depending on year/source).

I used to switch between a MacBook Pro with a 120Hz display and a MacBook Air with a 60Hz display and never noticed any difference. I think I could probably have noticed if I'd put them side by side and compared the feel of the scrolling, but why would I do that?

There's a minority of people who really do notice stuff like this, which is fine. The problem is that for some reason, they sometimes have difficulty accepting that they're a minority.

As people have already pointed out to you, it can't be evidence even of that because both candidates asked for extended time.