John Salvatier

It sounds like you don’t understand that engineering is compromises.

As soon as you involve radiation, you have some severe compromises to make, and the choices of elements you can use are limited.

Sodium is not chosen because of its dangers. Sodium was chosen despite the dangers, because the engineers very very carefully looked at other options and the other options were worse for the needs of their design. https://xkcd.com/793/ and http://johnsalvatier.org/blog/2017/reality-has-a-surprising-amount-of-detail seem relevant.

The following is an explanation of the Chernobyl accident, which in part was due to using water as a coolant:

Efforts to increase the power to the level originally planned for the test were frustrated by a combination of xenon poisoning, reduced coolant void and graphite cooldown. Many of the control rods were withdrawn to compensate for these effects, resulting in a violation of the minimum operating reactivity margin (ORM, see Positive void coefficient section in the information page on RBMK Reactors) by 01:00 – although the operators may not have known this. At 01:03, the reactor was stabilised at about 200 MWt and it was decided that the test would be carried out at this power level. Calculations performed after the accident showed that the ORM at 01:22:30 was equal to eight manual control rods. The minimum permissible ORM stipulated in the operating procedures was 15 rods. The test commenced at 01:23:04; the turbine stop valves were closed and…

> We expect our machines to behave in predictable ways.

I expect LLMs to produce randomly varying output. Maybe it's the thousands of hours I spent doing monte carlo simulations for my PhD.

> This is one of the best arguments against using LLMs I've seen.

> It reduces to the classic argument- at the point where you've described a problem and solution in sufficient detail to be confident in the results, you've invented a programming language.

I'm not an LLM true believer, but I use codex for various small tasks and it often (not always) does a thoroughly decent job. Yesterday I gave it a pretty vague request to set up a new Home Assistant dashboard and it handled it just fine--I told it what I wanted to see but it figured out itself which helper variables it would need to set up to realize that vision and wrote all the config for it.

I probably could have done it in 15 minutes if I was familiar with Home Assistant's yaml configuration schema and all, but I'm not so it probably would have taken me closer to an hour. Asking codex took me 30 seconds and it did just fine.

I am skeptical that LLM's are going to kill all white collar jobs or whatever anytime soon. Not being able to truly learn things is an issue. Reality has a surprising amount of detail[1], and while codex does well at things like writing Home Assistant configs and setting up a Minecraft server, where there are thousands of examples online of how to do it, when I've asked it to do some more esoteric things…

You may say that as a joke, but looking back at my education, this is in part what I did later, after uni, when I thought about things to learn next. In my mind, I started explaining stuff in depth, and found the many many many holes in my apparently quite superficial school knowledge.

I think this would be excellent as a once-in-a-while exercise to teach the complexity of it all, after we went the other way to abstract the hell out of everything to make it more palatable. A reality check, so to speak, and as a check for everyone about the many holes in their understanding. Basically, in support of something that has been posted here more than once, this blog post: http://johnsalvatier.org/blog/2017/reality-has-a-surprising-amount-of-detail

In the last few years I read a lot of easily digestible progression fantasy on RoyalRoad. One frequent theme is rebirth or isekai on another world, one that's usually much more primitive. Often the main character starts bringing earth technology ideas into the new world. And every single time it is obvious that the authors very, VERY severely the complexity and difficulty of even the smallest thing that we take for granted. My favorite blog post describing an extremely simple product, and how large and sophisticated its supply chain is: https://medium.com/@kevin_ashton/what-coke-contains-221d449929ef

To me, this shows that some more awareness of how much complexity there is in things might be valuable.

Being more aware of the…

> If the law is broken, fix it? The whole "spirit of the law" thing is what leads us to political attacks being made with laws.

Reality doesn't work that way. In fact, the whole reason for the ballooning of regulations everywhere is an attempt to make things more about "the letter" than "the spirit".

Reality is fucking complicated, as every programmer who ever tried to model or simulate anything from real life should know (and if one didn't have that experience, I recommend reading [0], recently discussed on HN). "Letter of law" is like code - in attempt to capture all the relevant details, it gets fractal in handling corner cases to corner cases. But, the best thing about humans is that our intelligence allows us to comprehend the intent behind something even without understanding all the relevant details and corner cases. This is what "spirit of law" exploits, and why it's a great tool at reducing complexity.

Having everything run on "letter of law" is practically impossible today, and would definitely not be practically usable with laws in textual form.

(This is, incidentally, why "smart contracts as laws" are stupid - computers can't understand intent, so you're enshrining some hard rules that will always be buggy and full of loopholes.)

--

[0] - http://johnsalvatier.org/blog/2017/reality-has-a-surprising-amount-of-detail

Cross-post: http://johnsalvatier.org/blog/2017/reality-has-a-surprising-amount-of-detail

Let’s consider some of these details. Something I think people don’t consider is that there are a non-trivial number of land animal species that migrate across the border. They have no concept of nation-states, and for some, our xenophobic electorate is basically passing a death sentence. The wall would devastate several important ecosystems, especially in the Rio Grande Valley.

Walls also need to be maintained, like any other piece of infrastructure. Have you ever visited the great state of Michigan? We can’t even handle roads. How will the federal government, which doesn’t have the best track record with maintaining infrastructure, deal with a 2,000+ mile-long wall?

Another issue is surveillance. Who will watch the entire thing? Who will make sure someone isn’t chipping away at the other side with a spoon or trying to dig under it? You can’t watch thousands of miles of wall very easily.

There are all kinds of issues with this (oh, funding is a big one too), and this just scratches the surface. If you Google “Trump border wall problems”, plenty of articles have already been written that go into much more detail.

Be fascinated with it. The normal advice is "do what you love", but quite often you can hate a lot of things. You can still be fascinated with something you hate.

You can look up Pirsig's Brick - if a city is boring, look for a specific building, and write about the building, starting with the top left brick. Reality has a surprising amount of detail: http://johnsalvatier.org/blog/2017/reality-has-a-surprising-amount-of-detail

Most of those little details are quite interesting. The big picture may be interesting too. It's the stuff in between that's boring. Try to narrow down your scope until you find an interesting thing to start with.

But besides boredom, it could be fear. We have a fight, flight, freeze response. The physiological response is exactly the same; it's the mental response that changes according to the situation. What this means is that we can just mentally swap out the flight response with fight.

If you find yourself procrastinating, fight for your goals! Make that goal important, to yourself, your society, your values. Stand your ground and move forward.