Programming thread

[Belisarius Cawl]

That's a floating point calculation and not an int. Monetary values should use a monetary datatype. Floats should always be compared with an epsilon, etc.

I can bitch and moan about language design, but numerical operations being garbage has been a problem since C. It's perhaps the original sin of programming.

While I was researching COBOL I found out that one of the advantages it had at the time (and since) was native support for (not always decimal) currency types

 

[Marvin]

Jamie Zawinski (worked on mozilla back in the day, also a huge sjw nowadays) wrote a blog post about Javascript's float nonsense ~15 years ago and Brendan Eich popped up in the comments to offer his thoughts. It's an interesting read.

 

[Belisarius Cawl]

Jamie Zawinski (worked on mozilla back in the day, also a huge sjw nowadays)

Every so often I pop into the most recent posts to watch him moo and ree about various things

 

[${Sandy}]

Tbh, I'll take doing any sort of math in javascript over coding up fucking ui.

Writing scripts that actually do something is fun. Because even when its difficult and I find it infuriating, its a tricky puzzle, and making my scripts work feels just as satisfying to me as a fucking orgasm.

But holy shit, css is such a pain in the fucking dick. Trying to make anything look good with css drives me nuts. Not because its particularly complicated, I just find it to be fucking time consuming and fickle.

 

[Belisarius Cawl]

Tbh, I'll take doing any sort of math in javascript over coding up fucking ui.

Writing scripts that actually do something is fun. Because even when its difficult and I find it infuriating, its a tricky puzzle, and making my scripts work feels just as satisfying to me as a fucking orgasm.

But holy shit, css is such a pain in the fucking dick. Trying to make anything look good with css drives me nuts. Not because its particularly complicated, I just find it to be fucking time consuming and fickle.

I know that feel. There are CSS frameworks out there that can assist you greatly:

https://www.browserstack.com/guide/top-css-frameworks

The end product won't be the same as one made by someone greatly familiar with graphic design but it will be functional and even perform well across various types of devices

 

[Private Tag Reporter]

I'd recommend Tailwind CSS if you're looking to get into CSS. It's halfway between a framework and inline CSS, and a not half bad way to start learning. The downside is that the build step required is kinda silly, but otherwise it's very cromulent for UI.

 

[Lard Shart]

I'd recommend Tailwind CSS if you're looking to get into CSS. It's halfway between a framework and inline CSS, and a not half bad way to start learning. The downside is that the build step required is kinda silly, but otherwise it's very cromulent for UI.

I second this. I despise writing CSS and while I usually avoid frontend tooling, tailwind really helps me quickly get a nice UI.

build step required

FYI, you can use Tailwind Play CDN. You only have to download a Javascript file, include it in your page and you can immediately start prototyping. Although, it's still recommended to have a build step for production use.

 

[${Sandy}]

Normally I use bootstrap but Ill check out tailwind.

 

[Belisarius Cawl]

The 25-year-old proompter:

I'M!!! ... PROOMPTING!!! *PROOOOOOOMPT* PROOMPTING!!! PROOMPTING!!!

Idiocracy-tier shit

 

[${Sandy}]

AI is pretty useful as a reference tool.I can bounce questions off it, and it will give me helpful pointers. Its just quicker than searching the internet. I ask it to link me to the the docs to learn.

Anybody who asks the AI to do their job for them, were probably not very good at coding to begin with. They are probably a dipshit like PirateSoftware or Yandere dev who would have made everything a massive niggerlicious if else statement. I honestly see nothing wrong with bouncing questions off it. Sometime finding what you are looking for takes forever.

Learning it doesn't take that long, actually finding the information does. I've had days where I spent fucking hours looking through docs and stack overflow and could not find anything fucking useful. AI points me in the right direction in like less than 5 minutes, and it links its sources so you are kind of a retard if you aren't taking advantage of that to learn. If anything AI isn't really an issue, its fucking morons who don't hold themselves to a standard. AI isn't gonna be taking their jobs, I will.

 

[Zeftax]

AI is a literal demon and anyone retarded enough not to see that shall burn in the eternal flame of the impending technojihad (hopefully).

 

[Creative Username]

Learning it doesn't take that long, actually finding the information does. I've had days where I spent fucking hours looking through docs and stack overflow and could not find anything fucking useful. AI points me in the right direction

why cant they just make search engines do that

AI is a literal demon and anyone retarded enough not to see that shall burn in the eternal flame of the impending technojihad (hopefully).

i consider it an overpowered markov chain that people use for many unorthodox things because its shiny and cool looking and people call it ai
it also approximates typical demon behavior quite well

 

[${Sandy}]

why cant they just make search engines do that

They used to before every tech company in the country decided to practice dei.

In all seriousness though, I use AI sparingly. I have microsoft copilot pulled up in one tab, docs in the other. I bounce between them.

Every coding sesh is a study sesh.

 

[Belisarius Cawl]

AI is pretty useful as a reference tool.I can bounce questions off it, and it will give me helpful pointers. Its just quicker than searching the internet. I ask it to link me to the the docs to learn.

I mainly ask LLMs about matters of opinion then go and do further searches based on the output provided. Any statement of fact I vet immediately before repeating it elsewhere, or to myself. I still haven't used Copilot or the like.

i consider it an overpowered markov chain

That's going too far, like those saying AI is just spicy if/else statements. If you've ever seen the output of the Mark V. Chaney script trained on a corpus of someone's actual writing, it looks like the author is in bad need of Seroquel. LLMs are typically coherent across sentences and even paragraphs, even if the factual content is sometimes lacking entirely.

 

[ADHD Mate]

why cant they just make search engines do that

AI is insanely good at semantic search, something that old indexing might not be as good at.
Being able to type actual sentences and having AI look up what you want without having to deliberately though out keywords is a game changer.
Such a pity that instead of focusing on models that could empower shit like Confluence, most AI companies focus on gimmicks like chat bots.

It's almost like most AI research were primarily done to empower search algorithms, ads targeting, and translations, and that's what they are best suited for.

 

[Belisarius Cawl]

AI is a literal demon and anyone retarded enough not to see that shall burn in the eternal flame of the impending technojihad (hopefully).

I, Archmagos Dominus Belisarius Cawl, Prime Conduit of the Omnissiah, █████████ unequivocally ███████████████ Abominable Intelligence ███████████████

 

[Creative Username]

That's going too far, like those saying AI is just spicy if/else statements. If you've ever seen the output of the Mark V. Chaney script trained on a corpus of someone's actual writing, it looks like the author is in bad need of Seroquel. LLMs are typically coherent across sentences and even paragraphs, even if the factual content is sometimes lacking entirely.

I wasn't directly equating it to a Markov chain; just pointing out that LLMs are built more like Markov chains than people seem to think they are. Both of them are essentially just a function that takes some text and outputs the most likely next piece of text. LLMs are obscenely good at it. But still, like the classic algorithm for generating incredibly drunk versions of random text corpora, LLMs are built around predicting the next token. They can just do it much more convincingly and coherently. It can then sometimes be useful to exploit the coherency of the output to perform a lot of hard NLP tasks; they aren't completely dumb.

You know how the "MA" stands for "Markov chain" in LZMA? Well, Fabrice Bellard made a text compressor from LLMs and it's incredibly powerful. It's part of the reason why I think of LLMs as extremely overpowered Markov chains.

 

[args]

when you convert null to a string you get "null" and not an empty value?

That's exactly what is does. In JSON the special symbol for null is simply null.
When you encode null and "null" you'll notice they're encoded as "null" and "\"null\"" respectively.

 

[y a t s]

It's almost like most AI research were primarily done to empower search algorithms, ads targeting, and translations, and that's what they are best suited for.

Funnily enough, the original intent behind the modern attention-based transformer mechanisms you see most AIs using today was to improve the efficiency—primarily, but not entirely, for training—of natural language processing (NLP) tasks, such as machine translation. It turns out it was useful for a lot more than just that.

Edit: attached PDF.

 

[Creative Username]

Funnily enough, the original intent behind the modern attention-based transformer mechanisms you see most AIs using today was to improve the efficiency—primarily, but not entirely, for training—of natural language processing (NLP) tasks, such as machine translation. It turns out it was useful for a lot more than just that.

Edit: attached PDF.

yeah afaik they kept trying to staple attention to lstms and other highly backprop-unfriendly architectures in various elaborate ways for a while until they realized that you can use it sort of like one of those linear cnns but not exactly
note: the above is probably retarded and wrong but it's how i understand the build-up to the transformer explosion