diarish

[kaberett]

Thing that is nice: I am mostly, currently, getting to horn practice o'clock after dinner, starting around eight, which is to say, just about as bat o'clock rolls around. I get to sit in the study and make squawking music and watch A Bat happen. It is lovely.

Another thing that is nice: I have spent the afternoon pivoting from "ugh but I don't WANT to read this paper from 1997" (the year I finally learned how to tie a shoelaces, via a lot of concerted effort) to "as a geologist I get to gleefully declare that I am CORRECT about the fundamental nature of the world and it's even ACCURATE (and now I've done that maths I suppose I'd better read the paper)".

I'm still see-sawing wildly between solid trust in my own models and towering impostor syndrome, of course, but it's so satisfying that every time I go "oh heck I didn't consider this constraint--" my beautiful model just. satisfies it. without even needing any massaging. Almost as if I'm right, or something.

Comments

[alexseanchai]

woohoo!

[vass]

I'm still see-sawing wildly between solid trust in my own models and towering imposter syndrome, of course, but it's so satisfying that every time I go "oh heck I didn't consider this constraint--" my beautiful model just. satisfies it. without even needing any massaging. Almost as if I'm right, or something.

This is so happy-making to read.

[simont]

but it's so satisfying that every time I go "oh heck I didn't consider this constraint--" my beautiful model just. satisfies it. without even needing any massaging. Almost as if I'm right, or something.

\o/

There's a similar sort of feeling I get with certain kinds of software – typically the kind that's trying to solve a complicated problem. A program like that goes through all the phases you'd expect: unwritten, partly written, written but not even handling the simplest test case correctly, gradual testing on more and more complicated handwritten or random test cases while bug-fixing, first deployment 'live' (whatever that means in a given case) and beginning to find the bugs your deliberate testing was too narrowly focused to look for, gradual stabilisation in live use. And in parallel with the program itself becoming better, my trust in it gradually increases.

All of that is what you'd expect, of course. But the interesting part (I think) is that somewhere near the end of this process, I often find that the final step towards really trusting the program consists of me going 'wait, that's odd, I haven't had to fix a bug in crunge-splorter in a while.'

The point when the last bug actually gets fixed (or rather, the last of the bugs that I'm in practice going to encounter in my normal use of the thing) is entirely unremarkable, because that bug looks just like the previous 53 and doesn't have a big sign on it saying "LAST ONE". So when I fix it, I have no idea that's what I've done. And then I remain in my previous untrusting state of 'I expect this contraption to fail again any minute now' for days, before suddenly realising '... wait, it hasn't been failing!'. Especially the moment of realisation often comes because I throw some unusual input at it and it doesn't hiccup, and I think 'ooh, I expected that to crash ... and come to think of it nothing else has crashed in a while.'

And that tends to be the moment where I flip over into believing that my program works, to the extent that now if it disagrees with someone else's result or fails to interoperate with someone else's supposedly compatible program, I might look for a fault in the other one first.

(A good example, I suppose, is puzzle generation. The test of trust is: if I think this puzzle has no solution or two solutions, do I look for the bug in the code, or the mistake in my play?)

[sfred]

Hurray for bats. D saw an owl out of our bathroom window last night. I missed it. It feels like I've seen fewer bats than usual this year, but maybe it's just a bit early for them this far north.

[silveradept]

Almost as if you have done reputable and solid science and it stands up well to the things that will test it. How odd.