a continuous cutting motion

there's also a big shift from "you have to do this work to earn arbitrary marks and better yourself - you are paying for this and it's for your benefit" to "you have to do this work because it will be useful - we are paying you to do this for our benefit"

i thrive on the latter

cult: you must not eat pineapples after 8am you: thats really stupid cult: wow i cant believe you care about eating pineapples after 8am THAT much lol.. mr pineapple over here.. lol you will literally die on this hill? lmoa? is there a name for this type of “attempted insanity transferal”

honestly it’s really fucked for everyone involved that the internet (inter alia) is in this messed-up place about trans women such that they can’t ever just be mildly annoying.

anything they do has to be trumpeted as Unimpeachably Valid and Beautiful as a counter to the prevailing tendency to casually brutally Cancel and make a huge fucking Moral Case about everything; it’s fucking exhausting.

hey just a quick heads up, I checked online and saw that some of you are still reblogging my posts

yikes thanks for letting me know

not helping

My programming habits must be very different from some people's, if the popularity of Github Copilot is anything to go on.

Maybe I need to give Copilot more of a chance? But the one time I tried to use it, it was so disruptive that I simply could not get anything done until I turned it off.

When I'm writing code, typing and deciding-what-to-type-next are all mixed together. Before I know exactly what a line ought to contain, I'll type out a subset of it that I feel relatively confident in, and that helps me concentrate on solving for the remaining piece.

Like, if I'm writing a list comprehension, I might first type out

and only then double back and figure out what the expression at the start should be. Or, if I'm writing a complicated condition, I might type out a relatively obvious part of it

and then think about the remaining parts with this text securely in place.

It feels helpful to do this, like I'm narrowing down a search space. I type the obvious parts quickly and readily, and then pause, to think about the less-obvious parts.

But with Copilot on, the act of stopping gives Copilot a trigger to suggest a completion. Right at the moment where I need to sit and think carefully, it jumps in and messes up my carefully arranged thinking space!

I guess if its suggestions were always right, this wouldn't be a problem? Like, if it just filled in the "less obvious parts" for me, and I could just read them and verify that they're correct.

But it's nowhere near able to do that.

The main problem isn't even with its level of coding ability, but with the fact that it doesn't know what I'm doing, and I have no way to tell it. It sees the code I already have, but it doesn't know about the new stuff I'm trying to put there -- the feature I'm working on, or the bug I'm fixing.

It would be more useful if it could automate the "obvious parts" -- the parts where it currently does nothing, because I'm typing quickly. That is, if it would pre-fill stuff like

Sometimes it does in fact do stuff like the second one... but only sometimes. And sometimes, even when it's guessing a short piece like `x is not None`, its guess is still wrong.

And then the other half of the time, it's Leeroy-Jenkinsing way out ahead of me, writing entire nested blocks of useless, irrelevant program logic. All because I typed `if`, or something, and then paused for a moment.

Update: I find Copilot a lot more helpful these days. I tend to have it turned on most of the time when I'm coding for work, now.

What made the difference? Presumably, some combination of

Subjectively, it feels like #1 is doing most of the work, with a bit of help from #3.

The rate of bad, irrelevant suggestions has gone down dramatically. The suggestions are now much more grounded in the real program logic of the codebase as a whole, rather than feeling like shallow autocomplete based on linguistic features of the nearest code block.

----

In the OP, I complained about the tool "messing up my thinking space." Why is this less of a problem than it used to be?

I'm not totally sure. Some of it is surely #2, that I've gotten used to the tool.

But it's also #1, again: the new suggestions are usually close enough to what I want that I can simply accept them immediately, then treat the resulting code as "thinking space." Even when the code isn't correct, I still save keystrokes; it has all the right symbols, and something like the right structure.

It's like the first few steps of my thinking process have been automated. "Obvious" code that only takes a few thinking-steps is correct on arrival; less-obvious code arrives in a "naive" form, the result of only 1-2 thinking-steps when more were required.

But that's okay. I get the first few steps for free, and only have to finish out the remainder.

----

These improvements might correspond to the model improvement that was announced on June 29:

There was a splashier new-model announcement in February, but I made the OP in April, so apparently the February model still wasn't good enough for me.

Separately, I've gotten access to the closed beta of their newer Chat feature, which IIRC uses a newer/better model even if you aren't making use of the chat functionality. But I never actually use that version of Copilot, because it only works in VS Code. So it can't account for the improvements I'm seeing. (I used VS Code in the final days of nostalgebraist-autoresponder, and tried the Chat model a bit in that context, but at work I'm very locked in to Pycharm.)

having flashbacks of looking at another bsd devs screen years ago at a hackathon during our libssl rewrite after…whatever that horrible libssl vuln was..

literally not even recognizable as C at first. they created their own weird byzantine convoluted sludge language with preprocessor macros and, my god, after closer inspection the sins they committed were unspeakable. we aren’t talking like #IFNDEF level shenanigans. every preprocessor sin committed at every level severity and in every combination possible. a permutation-complete reference work on how to go to hell.

it would be like building the wallet and grommace bed/breakfast contraption in real life, and then trying it for the first time, dislocating several vertebrae from the bed spring, breaking both legs falling through the floor, hitting your forehead on the table & concussing you, and then getting blasted with superheated porridge that has also started a fire. you yell “help me grommit!!” but it is a normal real life dog.

i just saw (reblogged by a celebrity blog whom i generally admire) a truly asinine "infographic" about how it might make us queasy but killing an oil exec would totally make an orders of magnitude larger negative carbon footprint than planting trees or whatever meticulously calculated from some bullshit numbers plucked fresh from their ass, and it puts me in mind of nothing quite so much as this little nugget of black comedy gold from the nincompoops over at huffpost in 2016. "uhhh sorry but the numbers dont lie, i used the microsoft excel calculator functions meaning this is basically a rock solid mathematical theorem shone into my brain directly by the stars up in platos heaven, ill admit smth in me rebels against the more repellent aspects [viz., respectively: adventurist assassinations, and tolerating the thought of hillary clinton in the oval office] of my political fantasies being 1000% vindicated here but The Numbers Don't Lie"

id mind it less if they just did not put on this song and dance about how reluctant they are to embrace the shocking conclusion of the ironclad logically impeccable argument. its tiresome in the same way as so much TF smut: we all know youre into it, if youre not going to give it the veneer of any actual realism or plausibility can you at least cut with the act that youre being dragged into the end state against your will?

hold up ✋ are you really busting out that "i agree with my own beliefs" rhetoric⁉️🤨

you can believe whatever you want, as long as you're not arrogant enough to think your beliefs are true

i remember arguing with my grandpa, when i told him i was taking a psychology class in college. He said it was bullshit, and I said maybe the freudian stuff they taught when you were there. But now it's the immortal science of cognitive psychology, they do experiments. I concede that argument to him, for what it's worth now that he's dead.

@mourning-ted said:

what are they?

off the top of my head

- operant conditioning: really it's psychology's win but if you put an animal in a box you sure can train it. every conclusion people have drawn though? super dumb, doesn't come close to explaining most things

- attention: if i give you two tasks you'll do worse than if i give you one task. the only thing people have tried to conclude from this truly shocking result is that you can't truly multitask, which obviously isn't true because the brain is doing more than one thing all the time. we can't even adequately work out what can be done in parallel because people get better at balancing too quickly

- working memory: if you have random people memorize "discrete" "things" they'll get around 7 (actually it's a different number i think? but this isn't my area anymore). does this mean we have 7 slots of working memory? probably not! you can memorize lots of other stuff and get around it, plus also there are techniques that let you memorize shitloads of numbers anyway. but if you do numbers with unpreped undergrads, you'll get the same result as every other experiment doing numbers with unpreped undergrads.

for all these you have a similar thing going on- the actual specific experiment is replicable, but there's nothing that generalizes. it's a parody of stamp collecting. like how us political polling replicates "roughly 50/50" every year

I can't laugh at this too hard because honestly molecular biology doesn't feel that different a lot of the time. Don't get me wrong there are obviously a lot of deep insights, but I'm currently working in a metabolomics lab that's technically a translational medicine lab, and the feeling I get from the research going on is a lot of "What's useful isn't true and what's true isn't useful".

We identified how NADH gets into mitochondria! That's really cool! Not really sure what to do with that knowledge though.

Have finally reached Alexander the Great in my reading... Frankly a boring individual apart from being one of history's greatest generals

well it isn’t like he had a lot of time to do anything else, what with the whole ‘spending almost his whole adult life on campaign and dying young’ thing; at what point in the life of alexander would you want him to have learned how to play the trumpet or invent bifocals or whatever

That’s not a defense, that’s an explanation of why he is in fact boring

isn't that a similar critique to anyone who's spent their life being a domain expert in some field, like Roger Federer say? the nasty little secret that biopics won't tell you is they spend their whole lives obsessively focused on their niche interest!

Yet another reason Magnus Carlsen is fucked, the man is the best chess player ever and also a very good soccer player and poker player who spends his non-chess time fucking around

most knowledge isn't scientific. Like even in, let's say, manufacturing penicillin, most of the knowledge involved is not the scientific knowledge the molecular structure of penicillin and how it acts on bacteria. It's stuff like manufacturing best practices, which I've heard are notoriously hard to transfer even from one of a company's factories to another. Even most of a scientist's knowledge that they use in their work is not itself scientific knowledge.

Scientific knowledge is

This isn't defining, and in particular the "general" thing means I'm only talking about specific sciences. ANy historical knowledge, like the mass of the rock that killed the dinosaurs, is not really "general" in this way, but of course it really is scientific. All I'm really trying to say here is, within this narrow definition of scientific knowledge, while it does include much of physics chemistry and biology, it's still a tiny fraction of knowledge in general.

It is a particularly easy kind of knowledge to acquire (since it's written down) and useful kind of knowledge to have (since it applies to many instances, it may apply in some case relevant to you), but it's still a small fraction.

And--ok, here's something that's come up before. I'm very skeptical of psychology and sociology. Yet I think people can understand themselves and their friends (the subject matter of psychology), or their office politics or friend group drama (subject matter of sociology). So what am I skeptical of? Claims of explicit, general knowledge of these things. There's not much you can say to me that will apply to my friends, if you don't know them. And I think it's important to recognize when you're talking in the sociological mode, invoking a principle of how people behave in general. Because it's probably wrong.

Anyway. THis is my attempt at a version of SCott's "Extreme Rationality: It's Not That Great" but for my own version of "rationality". I think science is cool, and I even think that the kind of "general science" knowledge that people get in school is valuable and worth learning even if you don't plan to be a specialist. Like, you don't need to know enough quantum mechanics to solve the hydrogen wavefunction, but I think it's attainable and worth it to know enough quantum mecahanics to understand why radio waves can't give you cancer the way UV rays do. But. It doesn't go that far. It's my thing, it's worth sharing, and it's what I have that's valuable to share, but it's just one little part of life.

Fuck with me

Source

these two are locked in eternal combat

Just found out my facebook birding group is public because my cousin (a lawyer who is not into birds) casually said to me “saw you couldn’t identify a willet the other day… pretty embarrassing”