The data center is on fire. The crowd starts chanting "the roof is on fire, let the motherfuckers burn". And some idiots keep throwing stuff onto the fire to feed it.

This post is about LLMs and climate change.

I was at Dynamo Open Air a few times in the 90s, but looking at the lineups, I mostly recall '96 and '99.

https://en.wikipedia.org/wiki/Dynamo_Open_Air

I think it must have been 1999 then that the party tent caught fire. The crowd started chanting the then relatively new Fire Water Burn chorus from the Bloodhound Gang. I left, but friends told me that some stoned metalheads thought it was funny to throw floor boards on the fire to feed it.

At any rate, that's a vivid mental image these days:

This is also a contender.

Edit: people comment it's slop. I didn't know it was. I don't post this to advertise anything.

For me, the source is a person near and dear to me, who sent the image in recognition of a part of me. For that, it doesn't matter if it's slop or handcrafted.

Is it disappointing it's slop? Yes. But it's also not what this thread (see OP) is about.

Contemplating a new profile pic.

It's not as if there is no law in space. That'd be a misconception.

But I'm fairly sure where e.g. GDPR or similar are concerned, you can always argue that a geostationary satellite above Europe *is* in Europe. Meanwhile nobody knows what you're actually doing there, or if the data is processed on a geostationary satellite above the, say, USA. Or even on any satellite.

Perfect deniability.

Was talking to a friend in the space sector today, and we turned towards how laughable orbital data centers are.

Then I thought: yes, *unless* your goal is to move data processing to a place with questionable jurisdiction and no way for any government to cut a pipe.

I've gone rogue and broken into a supermarket, but you really can't help it, I'm the future!

I used to be a little proud of our German words for everything. I like "fremdschämen", for example, which expresses embarrassment you feel at someone *elses* actions.

But now the world uses "cringe", and I mean, it's not exactly the same, but it sort of works, I guess?

Was wondering why I have a mastodon database of several gigs, and realize it's all posts the instance pulled.

Would be nice to be able to configure for how long to keep those, not just the ones you write yourself.

/me sighs

It's-a mushroom time!

I want to label the post vegan, but I know some vegans avoid mushrooms as well. 🤷

Of course, this was my little self-reward side project to inject into my more serious TODO list at intervals, that was supposed to keep me motivated.

Now I'm overwhelmed and dread doing anything at all.

*sigh*

I'll give it an hour, then serious TODO list it is. Has to be.

I wanted a quick dopamine hit, and thought "maybe I can give some thought to this very tightly scoped little side project".

Fatal mistake.

As I do these days, I first look around for papers that may say something on the subject. Found one immediately, which was neat, and referenced another one. That other one led to on to *another topic* I find fascinating, which overlaps with some previous project *and* a third toy project.

And now I want to do all of them, and realize I won't.

Thinking of archiving. I wonder if there's an affordable CNC type machine that can carve stone tablets. Might be worth writing some things down.

Mood.

Since everybody is posting about their model escaping its sandbox, I regret to tell you that was a daily occurrence for my kids when they were younger. It's not as cool as you think it is, particularly when they track sand all the way into your bed.

These were Windows 98 days. Nvidea still had competition.

I turned on the machine. It was slow to boot. After the desktop started was the slowest part. I noticed a tray icon popping up, then another. Meanwhile the VRML background loaded, and I saw a star field. Another few system tray icons flickered, and I was momentarily distracted by a 3D model of the USS Enterprise 1701-D flying towards the front of the screen.

Well, I say "flying". I saw a slide show. But it was fascinating ...

2/n

In the late 90s, I was working in a computer shop. On good days, I'd build a computer to spec from parts. Bad days, I'd open one up, hoover a bucket of nicotine dust out of the case, close it, and would have "fixed" the "it inexplicably shuts down a few minutes after I turn it on" problem.

Anyway, there's one customer I'll never forget. Well, I never laid eyes on the guy. I'll never forget the problem he had. He'd bought a top-of-the-line machine, then brought it back because it was slow.

1/n

... firefox code base. What I can say is that the answer to that always depends. There is usually a way to guesstimate a reply based on the historical data of the project.

But what's *also* clear here IMHO is that the time to ask this question is *when a big change* comes.

It's always too late to ask this when several such big changes have been haphazardly flanged together.

But the question that is being asked instead is about benefit, not about TCO.

Now a big change rolls around. Making that change fast is going to take C1 time. Making that change such that the code base remains reasonably clean will take C2 time, where C2 > C1.

What we're really asking is this:

Is (N * T * F) - (N * T) larger or smaller than (C2 - C1)? That is, is the cost of doing things well higher than the difference in TCO that stems from doing things well vs. doing things fast?

Hey, look, I can put this in business-y terms after all.

I can't answer this for the...

... none, we can ignore them. This is also going to be the vast majority of incremental changes.

So if we're asking about incremental changes related to these subsystems, you can probably handwave the incremental cost of such a change on the grown codebase to be twice that of a clean code base. Any more realistic factor F is fine, really, I'm pulling the "twice" out of my arse^Wexperience.

So we have N * T * F for the cost of those changes on the grown code base vs. N * T for the clean one.