Allusion
eevee đŠdown with Whatever https://eev.ee/blog/2025/07/03/the-rise-of-whatever/
sristow
Luna Lactea
Risotto Bias@eevee :0 you don't like closing tab autocomplete?
Like <div>|</div>
Or func(|)
Or var strings =[]{"|"}
@eevee or tab auto indenting in Python files :0
@risottobias i have autoindent, that's fine
Ălbaum@eevee Same. Since I started programming before these features existed (itâs not a flex, itâs context,) I took on the habit of typing ()â whenever I write a function call, for example. Itâs as fast as the autoclose, but most importantly, it (as in I) only does that when I want it to. I can type a single ( without fighting the system.
@risottobias
@risottobias
@eevee Recently, I managed to configure VS Code to show suggestions (function names and arguments only, no Copilot bullshit) but never complete them unless I explicitly ask it to. Before I managed that, I had to disable the whole feature.
I didnât want suggestions to save on typing. Programming should be 90% thinking, 10% typing, (and 40% drinking coffee if thatâs your thing.) I use them to save on remembering and looking up function signatures. Donât need an LLM for that.
@risottobias
mirabilos@risottobias @eevee thatâs even worse!
@risottobias no it drives me absolutely up the wall. the html version is even worse because i've gotten spurious </div> in places that don't even make sense
@eevee it would be nicer if it detected an unbalanced set of divs. some editors when you type ) will automatically recognize that it autocompleted one for you and skip over it instead, some are definitely more primative and just go )) instead :/
definitely something in an editor to tune exactly how you want,
e.g., autoclose brackets, braces, tabs, under what circumstances instead a list or class, etc
auto-bracket lisp :O
@risottobias when i've encountered this feature it usually understands when i type the closing ), but if i go BACK to edit code later â which happens, like, a lot â then it gets hopelessly confused because it just has no way of knowing what i'm trying to do
and </div> is more than one characters so it doesn't understand when i try to type that myself
@eevee oh 100% infinite unbalancing vs coloring brackets and going "we already have a matching pair"
LPđžDont let the night be over@eevee @risottobias I usually have it on but it's definitely a borderline one as to if it helps or hinders in a given moment
But, crucially, 1) it's based on a largely predictable algorithm, so it's rarely *surprising* and I know when I'm going to have to dance around pressing *more* keys to correct it, and 2) it's nobody's pet Whatever to pump, so I can just...turn it off, granularly from other IDE smarts, and won't even get a "turn it back on? [yes!] [remind me later]" dialogue tomorrow.
Emmy, Gendermancer@eevee @risottobias I'm not a programmer. I barely even write code, or script, or whatever. But when I do... That shit drives me up the wall. Glad to finally see someone else talking about it. đ
I'm about 2/3 of the way through reading, btw, and it's fantastic.
raphael-proust@risottobias @eevee yeah same i dislike it, i think it falls in the same category as autocorrect for me
yes maybe if you build the habit it can be faster most of the time⊠but âfastâ is just the one metric that's easy to collect for something more important: âunfrustratingâ. like a drunk looking for their keys near the lamppost because that's where it's easier to look, so does the developer try to optimise the number of keystrokes it takes to complete some specific scenario (instead of doing lengthy qualitative research into uses).
sure auto-close and auto-correct and the like are faster for some specific tasks, and if you get used to it you tend to write so as to stay within these specific tasks or you just accept the frustrating cases that keep coming up because you've built up the muscle memory for the fast cases and at this point it's more frustrating to not use this feature. but often it's worth trying a new tool even though you need to climb that learning curve.
too many typos bc of the small keys on your phone? maybe give a go at alternative keyboards like 8vim and flickboard.
@risottobias @eevee yes, fuck that antifeature
Wolf480pl@eevee
> I feel like I get a little dumber every time I accidentally start reading it [LLM output]
so much this!!
@eevee
> [The attitude about LLMs] feels like the same attitude that happened with Bitcoin, the same smug nose-wrinkling contempt. Bitcoin is the future. Itâll replace the dollar by 2020. Youâre gonna be left behind.
> What do programmers get out of this?
I think you've struck something deeper, not tied to vested interest.
Isn't this similar to the smug "I use Arch btw" ?
I think I've seen this much earlier, in non-IT context, but I can't recall an example...
religion? đ€
@eevee
Also you wrote that the LLM hype is rooted in the idea that "doing things is worthless".
To me that sounds awfully like burnout / depression / learned helplessness.
So maybe that's where this comes from? Maybe people have been doing tedious work without seeing any purpose or meaning in it, they got burned out, and now they want a machine to do that job for them...
> What are you all even writing that so much of it consists of generic slop?
IMO many jobs are like that :(
Kiloku@wolf480pl @eevee My lead dev who uses LLMs to "help" him (I've caught code with mistakes I know he wouldn't have made if he wrote it himself) tends to work until like 3 AM every now and then. I guess if you feel the need to overwork you also start resenting the work
Also, pressure to get the thing to work no matter what is such a killjoy.
It can be external, but it can also be internal - when you know a lot of people rely on you, they don't need to tell you, you just know it.
It makes it seem like the result is all that matters, and then the process itself is no longer enjoyable.
@wolf480pl @eevee read up on what TESCREAL is (and the names behind that are the techbros you recognise from the AI crap). Religion is not too far offâŠ
Arne Martin@eevee This was extremely good đ
IrgndSonDepp@eevee Your post is đ„!!!
I'm still holding out hope that the bubble will burst sooo hard before my kids get into school. My head is close to falling off from all the head shaking whenever I read about the shit that's going on schools and universities with LLMs.
I hate the "whatever machine" with the passion of a thousand burning suns. đŹ
đ Emerald@eevee I had the strangest experience entering a comment on your post, using firefox for android on a pixel 2, using a really old version of (now Microsoft) Swiftkey (I'm afraid to update anything b/c of business-driven regressions like llm feature tumors)
entering text at the front reversed it character by character & reset the cursor to before the text, unless I typed too fast (I did the first line very slowly w/ much do over). entering text elsewhere or newline caused strange duplications.
@lioness haha i just saw it. unfortunately it's disqus which is third-party so i have no control over it whatsoever
@eevee "I mean, I get it. I was trying to do something that had never been done before. LLMs are fine at things that appear a zillion times in their training data â in fact, this is probably a big part of the trick, because the things that appear more often in their training data are the things people are more likely to ask about in general and thus the things people are more likely to ask an LLM."
I've tried to explain this sometimes and people don't seem to get why this makes it useless đ
Joel Kin
Jocelynephiliac 
@eevee read the entire thing without skimming, fantastic essay that really nails everything Iâve hated about this tech. Bravo
lime!@eevee the devaluing of creation really is the crux of it, because that also stops adoption of these things for whatever stuff there actually is they could be useful for by saying "they can do everything for you". we'll likely not see actual uses for these things until the entire current cycle dies down, and at that point they will likely be so constrained that they're not recognisable as 'ai' tools.
like, i sometimes do single-pass image-to-image runs on digital paste-up artworks i try to do, so i have something to paint over without having to repeatedly hunt for the right colour combinations for shadows and such, which i'm really bad at. in that case it acts like a photoshop filter. that's where i see this entire thing going once the hype dies.
@lime image-to-image still seems pretty weird since aiui it's going through /two/ massive generalist databases so you have basically no idea what will come out the other side (or, of course, how much it will resemble someone else's extant artwork, which seems to be a much bigger problem with imagegen.)
@eevee more like one-and-a-half; since the process basically "de-blurs" an image from noise, i can add random noise to my own image and tell the model "okay we're on step 15 of 20, deblur the rest".
the fact that it's basically impossible to avoid plagiarism is why i only do these for myself, as experiments. i don't want to put more slop on the web.
@eevee but yes, it does still go through clip and the feature database, and still requires a prompt. it just skips most of the actual generation.
sometimes it can be fun to just give the model shapes and get it to find things in them, cloud gazing style, like the google thing that saw dogs everywhere.
@lime noise on an existing image is interesting at least. that's certainly more directed than "give me Whatever"
@eevee it's Guided Whatever. so it's basically the same thing as all the chatbots everyone is getting, but for images.
tangent:
other than doing art experiments just for me, i'm currently studying llms at work so i can stay employable, and going through the code of these systems is depressing. like, i don't have the math brain to actually decipher the vector calculus at the base of the system, but i can churn through all the structure of the system, and... there is none. no organisation, no security consideration, no nothing. it's all research code that escaped.
in no other production ready system have i encountered a function named essentially "load_local_file" which, when called with a string argument representing the path of a local file, does a http call to a proprietary service with an api key embedded in the library, that basically only _looks at the file name_ and says "ok here is 5 GB of extra files for that one file".
deploying these systems in secure contexts (which i know people do) mus be a nightmare.
Filip@eevee
The fact that you have no idea what will come out the other side actually seems to be part of the appeal â as @davidgerard recently wrote, it's like playing a slot machine: https://pivot-to-ai.com/2025/06/05/generative-ai-runs-on-gambling-addiction-just-one-more-prompt-bro/
@lime
The fact that you have no idea what will come out the other side actually seems to be part of the appeal â as @davidgerard recently wrote, it's like playing a slot machine: https://pivot-to-ai.com/2025/06/05/generative-ai-runs-on-gambling-addiction-just-one-more-prompt-bro/
@lime
Philipp
Phil Giammattei@eevee this is so good, thank you for writing it
Bel Tamtu
fraggle@eevee agree so much with everything you said here, thank you. You call it Whatever, I call it bullshit. We're living in the Age of Bullshit. I've come to believe it's the defining characteristic of the time we're living in, and AI is only one aspect of it but it's certainly becoming a big part
das-gBullshit makes
the flowers grow
& that's beautiful.
AI slop can't do that.
Hazelnoot
Damon L. Wakes@eevee It's so good to see someone else address this. It's been driving me nuts. I mean, how much effort does it save to throw in an end bracket for me if I then have to hit an arrow key to skip past the thing ~even if it's in the right place~?
Rémi Dupré@DamonWakes @eevee hm, maybe it helps with keeping a parsable code while typing so the syntax coloring and linters don't panic?
@remidupre @eevee That would make a certain amount of sense, but if that's the reason I'd vastly prefer syntax highlighting to hold off updating while I'm actively typing. I want my tools to draw attention to errors, not try and prevent them from ever existing even for an instant.
Phil Ashby 
đ”@DamonWakes @remidupre @eevee
This, I think the tooling is trying to "shift left" too far (into my head perhaps?). I like the integration of language servers and linters into Vim, they don't run until I drop out of insert mode, or at least until I hit <CR>.
This, I think the tooling is trying to "shift left" too far (into my head perhaps?). I like the integration of language servers and linters into Vim, they don't run until I drop out of insert mode, or at least until I hit <CR>.
Yuna
Omega 
QuARC-08380@eevee '(v0160 (TANG (ST "I die inside whenever I have to close brackets manually (colour coded brackets make it a bit easier, I guess)" )
(ST "that said I did have to write this post without either of those" `)))
(ST "that said I did have to write this post without either of those" `)))
(v0160 (EXT (ST "(for context, watch the old SICP lectures where they spend way too much time closing brackets)" ))) 
Niku
Pavel Machek@eevee I used LLM to translate into czech and get summary. Should I feel bad? :-). There are good uses for LLM, and yes, that includes "startrek", but someone will have to do that. And yes, I tried to start (but did not get far, so far).