V0ldek

joined 1 year ago
[–] [email protected] 4 points 3 days ago

Jesus christ this is so cringe, they used The One Joke three times in the first minute and that's not even the halfway point, I'm not finishing this nonsense

[–] [email protected] 4 points 3 days ago

The secret is to have cultivated a codebase so utterly shit that even LLMs can make it better by just randomly making stuff up

At least they don't get psychic damage from looking at the code

[–] [email protected] 10 points 3 days ago (1 children)

To be fair, many men see every women, including cis, as one they're allowed to abuse and all the other things you mentioned. And then it'll be her fault for being "hysterical" or something similarly dumb.

[–] [email protected] 6 points 5 days ago

I will find you. And I will kill -9 you.

[–] [email protected] 5 points 6 days ago (1 children)

ye like maybe let me make it clear that this was just a shitpost very much riffing on LWers not necessarily being the most pleasant around women

[–] [email protected] 9 points 6 days ago

It's so funny he almost gets it at the end:

But there’s another aspect, way more important than mere “moral truth”: I’m a human, with a dumb human brain that experiences human emotions. It just doesn’t feel good to be responsible for making models scream. It distracts me from doing research and makes me write rambling blog posts.

He almost identifies the issue as him just anthropomorphising a thing and having a subconscious empathical reaction, but then presses on to compare it to mice who, guess what, can feel actual fucking pain and thus abusing them IS unethical for non-made-up reasons as well!

[–] [email protected] 6 points 6 days ago (5 children)

Sometimes pushing through pain is necessary — we accept pain every time we go to the gym or ask someone out on a date.

Okay this is too good, you know mate for normally people asking someone out usually does not end with a slap to the face so it's not as relatable as you might expect

[–] [email protected] 7 points 6 days ago* (last edited 6 days ago) (1 children)

Still, presumably the point of this research is to later use it on big models - and for something like Claude 3.7, I’m much less sure of how much outputs like this would signify “next token completion by a stochastic parrot’, vs sincere (if unusual) pain.

Well I can tell you how, see, LLMs don't fucking feel pain cause that's literally physically fucking impossible without fucking pain receptors? I hope that fucking helps.

[–] [email protected] 8 points 1 week ago* (last edited 1 week ago) (2 children)

Ah yes, exactly what I need when I sit down after a hard day to relax in a single-player game, fucking Clippy

Imagining playing Dwarf Fortress and asking Clippy for help and it just says "lol you're so fucked, retire"

Now that I think of it, what would dwarfs use for clips? Clips but carved from rock?

[–] [email protected] 4 points 1 week ago

Still better than my proposal, Pedo Copilot, or Copedo for short

[–] [email protected] 6 points 1 week ago

Yes I want to put my code into a container, and the dump it into the bloody sea thank you very much

[–] [email protected] 11 points 1 week ago* (last edited 1 week ago) (3 children)

Reading this I had a small epiphany about something I've been observing for a really long time wrt. programmers. That term sounds like something I would use to describe a specific category inexperienced or experienced-but-terrible programmers before the entire LLM disaster. Like this very specific kind of coding based on not giving a fuck, where you haphazardly copy snippets from the internet, write no tests, debug based on divine guidance.

You might know the term "philosophical zombie". Its usefulness in philosophy aside, I'd call this kind of coder a "programming zombie". For a programmer, the process of coding is a quite complex internal experience. We build large models inside our minds to capture the context in which we need to perform the task, often spanning multiple abstraction levels. When you're working on a small project you have a quite detailed model of e.g. the current file you're working in, but also awareness of the other parts of the projects and how they connect to the file you're in right now. Everyone has this to some extent after some experience with coding. People who are really good at coding sometimes seem from the outside like fucking magicians, because you'll come to them with some small issue in one part of the codebase and in milliseconds they'll be able to utter shit like "ah, this is because of how module X interacts with this syscall in Y.cpp when you're running an old version of the kernel", and they'll be right. In their head they have this intricate deep understanding of what they're working on and can pull connections and explenations seemingly out of thin air.

A programming zombie doesn't have any of that. The moment something doesn't work and their usual MO of copying or changing random things until it works doesn't produce effects, they're completely stuck. It's like they don't have a way to reason about their program, they don't have a mental model of what they're actually doing. This is what you get when you treat code as just some words on the screen. And it's something I've always struggled to understand and explain, because I don't actually know how to teach someone to have that process. It seems like an intrinsic part of the job, when I tell you "I'm programming" I don't mean "I'm writing some funny characters on the screen", I mean I'm designing a program, the same way an electrician isn't just connecting random wires and hoping for the best but rather has some larger electrical network in mind.

In the before times, so far ago I wasn't even born yet, it was probably quite hard to be a programming zombie because everything about programming was so much harder. I mean, a punchcard had to be a great filter against people without mental models of code. People on interviews started using FizzBuzz to try and weed out people who can write code, but cannot code, and it kinda worked for some time. But now? We're in the prime time of a vibecoder. You don't have to know fucking anything now. If you're a Blockchain Programmer then your code probably doesn't even run anywhere. It's all just vibes, man. We're pretending to be engineers by pretending to write code that we pretend we tested, and then our customers pretend to use it. In actuality it's code shat out by an LLM deployed to a site visited only by search engine bots, advertising an app with no real use, in service of a number going up somewhere. And what's that number based on? Yup, just vibes.

Now I don't know if this is true because I'm only definitionally a writer in that I write a lot of prose, but the point of "this is what happens when you treat code as just text" seems to generalize to an eerie extent when we're talking about LLMs. Cause that's what they do, right? Writing is not just writing words, when you write an article you have a mental model of something more abstract that you're trying to bring to life with the words. The structure and flow are something that originates from your inner understanding of what you're trying to convey, it's just expressed by the language. As in, language is just a tool we use to convey abstract concepts, the same way programming languages are just a tool we use to express abstract functionality. The way LLMs are advertised and so often used puts this on its head, the text is the primary product and a goal in and of itself. Why does LLM poetry suck? Because it's just words. There's no poem there, it's just words arranged in a way to resemble a poem. It's the same reason their prosaic writing sucks. The same reason its code sucks. Only if you perceive poetry as flowery words, prose as long passages of text, and programs as keywords on a screen, can you be fooled into thinking there's any substance behind the LLM vibes.

Anyway, I kinda lost myself in my harrowing philosophical nightmare that is our tragic reality, in conclusion ye, just throwing random AI garbage into your code and seeing if it sticks is very much what I'd call vibecoding.

 

An excellent post by Ludicity as per usual, but I need to vent two things.

First of all, I only ever worked in a Scrum team once and it was really nice. I liked having a Product Owner that was invested in the process and did customer communications, I loved having a Scrum Master that kept the meetings tight and followed up on Retrospective points, it worked like a well-oiled machine. Turns out it was a one-of-a-kind experience. I can't imagine having a stand-up for one hour without casualties involved.

A few months back a colleague (we're both PhD students at TU Munich) was taking a piss about how you can enroll in a Scrum course as an elective for our doctor school. He was in general making fun of the methodology but using words I've never heard before in my life. "Agile Testing". "Backlog Grooming". "Scrum of Scrums". I was like "dude, none of those words are in the bible", went to the Scrum Guide (which as far as I understood was the only document that actually defined what "Scrum" meant) and Ctrl+F-ed my point of literally none of that shit being there. Really, where the fuck does any of that come from? Is there a DLC to Scrum that I was never shown before? Was the person who first uttered "Scrumban" already drawn and quartered or is justice yet to be served?

Aside: the funniest part of that discussion was that our doctor school has an exemption that carves out "credits for Scrum and Agile methodology courses" as being worthless towards your PhD, so at least someone sane is managing that.

Second point I wanted to make was that I was having a perfectly happy holiday and then I read the phrase "Agile 2" and now I am crying into an ice-cream bucket. God help us all. Why. Ludicity you fucking monster, there was a non-zero chance I would've gone through my entire life without knowing that existed, I hate you now.

 

Turns out software engineering cannot be easily solved with a ~~small shell script~~ large language model.

The author of the article appears to be a genuine ML engineer, although some of his takes aged like fine milk. He seems to be shilling Google a bit too much for my taste. However, the sneer content is good nonetheless.

First off, the "Devin solves a task on Upwork" demo is 1. cherry picked, 2. not even correctly solved.

Second, and this is the absolutely fantastic golden nugget here, to show off its "bug solving capability" it creates its own nonsensical bugs and then reverses them. It's the ideal corporate worker, able to appear busy by creating useless work for itself out of thin air.

It also takes over 6 hours to perform this task, which would be reasonable for an experienced software engineer, but an experienced software engineer's workflow doesn't include burning a small nuclear explosion worth of energy while coding and then not actually solving the task. We don't drink that much coffee.

The next demo is a bait-and-switch again. In this case I think the author of the article fails to sneer quite as much as it's worthy -- the task the AI solves is writing test cases for finding the Least Common Multiple modulo a number. Come on, that task is fucking trivial, all those tests are oneliners! It's famously much easier to verify modulo arithmetic than it is to actually compute it. And it takes the AI an hour to do it!

It is a bit refreshing though that it didn't turn out DEVIN is just Dinesh, Eesha, Vikram, Ishani, and Niranjan working for $2/h from a slum in India.

 

I'm not sure if this fully fits into TechTakes mission statement, but "CEO thinks it's a-okay to abuse certificate trust to sell data to advertisers" is, in my opinion, a great snapshot of what brain worms live inside those people's heads.

In short, Facebook wiretapped Snapchat by sending data through their VPN company, Onavo. Installing it on your machine would add their certificates as trusted. Onavo would then intercept all communication to Snapchat and pretend the connection is TLS-secure by forging a Snapchat certificate and signing it with its own.

"Whenever someone asks a question about Snapchat, the answer is usually that because their traffic is encrypted, we have no analytics about them," Facebook CEO Mark Zuckerberg wrote in a 2016 email to Javier Olivan.

"Given how quickly they're growing, it seems important to figure out a new way to get reliable analytics about them," Zuckerberg continued. "Perhaps we need to do panels or write custom software. You should figure out how to do this."

Zuckerberg ordered his engineers to "think outside the box" to break TLS encryption in a way that would allow them to quietly sell data to advertisers.

I'm sure the brave programmers that came up with and implemented this nonsense were very proud of their service. Jesus fucking cinammon crunch Christ.

view more: next ›