Mii @ mii @awful.systems

I wrangle code, play bad music, and write things. You might find some of it

---

Read more

Posts

6

Comments

245

Joined

1 yr. ago

AI use at my company is now mandatory. We all have gotten really expensive Claude of ChatGPT licenses and (apparently) it's being monitored whether we actually use them. For while I've resisted, telling everyone I don't touch that shit with a ten-foot pole. Now I just write the occasional non-invasive prompt ("check for spelling errors in the comments" or "format the buffer so that all lines have less than 80 columns) just so someone can report to the c-suite that our department is using AI.

Meanwhile I have to do code reviews on patches that have been actually written by a chatbot and doing that takes sometimes as long as writing the whole fucking code from scratch.

But hey, style guidelines are out the window too because the slop machine can't follow them consistently anyway and apparently fixing that shit manually is too much to ask. So at least I never have to fight anyone again over indentation and braces.

If infinite monkeys with typewriters can compose Shakespeare, then infinite monkeys with slop machines can produce Einstein (but you need to pump in infinite amounts of money first into my CodeMonkeyfy startup, just in case).

Remember Devin, the AI coding bot that turned out to be a complete scam?

Well, looks like Goldman Sachs decided to let it loose on its code base. In YOLO mode too, it seems.

There’s something I’ve been wondering about: is the management (of not just Microsoft but really any company who’s pivoting to AI this hard, so also Duolingo and Klarna, &c.) really convinced that there’s going to be a foom event soon and the chatbots become capable enough to actually do the work they attribute to them? Or do Nadella and friends know this is all bullshit and just play along until they can’t anymore and leave the sinking ship?

I honestly can’t tell anymore.

I looked this up because I thought it was a nickname for something, but no, Cursor seems to have a setting that’s officially called YOLO mode. As per their docs:

With Yolo Mode, the agent can auto-run terminal commands

So this guy explicitly ticked the box that allowed the bullshit generator to execute arbitrary code on his machine. Why would you ever use that? What's someone’s rationale for enabling a setting like that? They even name it YOLO mode. It’s like the fucking red button in the movie that says, don’t push the red button, and promptfans are still like, yes, that sounds like a good idea!

That was one wild read even worse than I was expecting. Holy sexism Batman, the incel to tech pipeline is real.

"In college, you don't learn the building skills that you need for a startup," Tan says of his decision. "You're learning computer science theory and stuff like that. It's just not as helpful if you want to go into the workforce."

I remember when a large part of the university experience was about meeting people, experiencing freedom from home for the first time before being forced into the 9-5 world, and broadening your horizon in general. But maybe that's just the European perspective.

In any case, these people are so fucking startup-brained that it hurts to think about.

Now 25, Guild dropped out of high school in the 10th grade to continue building a Minecraft server he says generated hundreds of thousands of dollars in profit.

Serious question: how? Isn't Minecraft free to play and you can just host servers yourself on your computer? I tried to search up "how to make money off a Minecraft server" and was (of course) met with an endless list of results of LLM slop I could not bear to read more than one paragraph of.

Amid political upheaval and global conflict, Palantir applicants are questioning whether college still serves the democratic values it claims to champion, York says. "The success of Western civilization," she argues, "does not seem to be what our educational institutions are tuned towards right now."

Yes, because Palantir is such a beacon of defending democratic values and not a techfash shithouse at all.

I sometimes feel that I, as someone who also likes retro computing and even deliberately uses old software because it feels familiar and cozy to me, and because it's often easier to hack and tweak (in the same way that someone would prefer a vintage car they can maintenance themselves, I guess), I get thrown in with these people -- and yes, I also find it super hard to put a finger on it.

I also feel they're very prominent in the Vim community for the exact same reasons you mentioned. I like Vim, I use it daily and it's my favorite editor because it's what I am used to and I know how to tweak it, and I can't be bothered to use anything else (except Emacs, but only with evil-mode), but fuck me if Vim evangelists aren't some of the most obnoxious people online.

New Zitron dropped, and, fuck, I feel this one in my bones.

Pretty good summary of why Alex Karp is as much a horrible fucking shithead as Thiel.

https://www.thenation.com/article/culture/alex-karp-palantir-tech-republic/tnamp/

Klarna is one company that boggles my mind. Here in Germany it’s against literally every bank's TOS to hand out your login data to other people, they can (and do) terminate your account for that. And yet Klarna works by asking for your login data, including a fucking transaction token, to do their thing.

You literally type your bank login data including an MFA token into a legalized phishing site so they can log into your account and make a transaction for you. And the banks are fine with it. I don’t get it.

The German Supreme Court even deemed this whole shit as unsafe all the way back in 2016 and said that websites aren’t allowed to offer Klarna as the only payment option because it’s an “unacceptable risk” for the customer, lol.

Oh, and they of course also scan your account activity while they’re in there, because who’d give up all that sweet data, which we only know because they’ve been slapped with a GDPR violation a few years back for not telling people about it.

Yet for some reason it is super popular.

I tried to find that out but it doesn't seem to be mentioned in any of the articles. However, searching for the names associated with it (wife, brother-in-law, and the third business partner) brings up "Kadima Ventures" from Arizona, which has these three names listed as management. However, information about that company beyond some weird mentions and a LinkedIn profile seems to be scarce (and the AZ business registry is either down right now or not accessible from where I am).

I looked again and found a 404 article on the whole story that's better and has more info than the one I linked before, though. I'm adding that to the original post.

Road rage victim 'speaks' via AI at his killer's sentencing [Archive]

I fucking can't right now.

[Judge] Lang allowed Pelkey's loved ones to play an AI-generated version of the victim — his face and body and a lifelike voice that appeared to ask the judge for leniency.

“To Gabriel Horcasitas, the man who shot me: It is a shame we encountered each other that day in those circumstances," the artificial version of Pelkey said. "In another life, we probably could have been friends. I believe in forgiveness."

Edit: 404Media article on the story that's much better

Marc Andreessen claims his own job's the only one that can't be replaced by a small shell script.

https://gizmodo.com/marc-andreessen-says-one-job-is-mostly-safe-from-ai-venture-capitalist-2000596506

“A lot of it is psychological analysis, like, ‘Who are these people?’ ‘How do they react under pressure?’ ‘How do you keep them from falling apart?’ ‘How do you keep them from going crazy?’ ‘How do you keep from going crazy yourself?’ You know, you end up being a psychologist half the time.”

Along the same lines of LLMs ruining language stuff: I just learned the acronym MTPE (Machine Translation Post Edit) in the context of excuses to pay translators less and thanks I hate it.

Not-so-fun fact: that's a marketing term for what amounts to basically a scam to pay people less.

I used to work for a large translation company when this first came up. Admittedly, that was almost ten years ago, but I assume this shit is even more common nowadays. The usual procedure was to have one translator translate the stuff (commonly using what's called a TM or Translation Memory, basically a user dictionary so the wording stays consistent), and then another translator to do an editing pass to catch errors. For very high-impact translations, there could be more editing passes after that.

MTPE is now basically omitting the first translator and feeding it through a customized version of what amounts to Google Translate or DeepL that can access the customer's TM data, and then handing it off to a translator for the editing pass. The catch now is that freelance translators have two rates: one for translating, depending on the language pair between $0.09 and $0.5 per word, and one for editing, which is significantly less. $0.01 to $0.12 or so per word, from what I remember. The translation rate applies for complete translations, i.e. when a word is not in the customer's TM. If it is in the TM, the editing rate applies (or, if the translator has negotiated a clever rate for themselves, there might be a third rate). With MTPE, you now essentially feed the machine heaps of content to bloat up the TM as much as possible, then flag everything as pre-translated and only for editing, and boom, you can force the cheapest rates to apply to what is essentially more work because the quality of what comes out of these machines is complete horseshit compared to a human-translated piece.

For the customers, however, MTPE wasn't even that much cheaper. The biggest difference was in the profit margin for the translation company, to no one's surprise.

Back when I worked there, and those were the early days, a lot of freelance translators flat-out refused to do MTPE because of this. They said, if the customer wants this, they can find another translator, and because a lot of customers wanted to keep the translators they'd had for a long time, there was some leverage there.

I have no idea how the situation is today, but infinitely worse I assume.

Shit, I actually like Hank Green his brother John. They’re two internet personalities I actually have something like respect for, mainly because of their activism, John’s campaign to get medical care to countries who desperately need it, and his fight to raise awareness of and improve the conditions around treatment for tuberculosis. And I’ve been semi-regularly watching their stuff (mostly vlogbrothers though, but I do enjoy the occasional SciShow episode too) for over a decade now.

At least Hank isn’t afraid to admit when he’s wrong. He’s done this multiple times in the past, making a video where he says he changed his mind/got stuff wrong. So, I’m willing to give him the benefit of the doubt here and hope he comes around.

Still, fuck.

Christ on a stick, they found a way to remove the black person from the “I’m not racist, one of my friends is black” dogwhistle.

I fucking can’t.

Fair points, but I still take cleaning up someone’s own bad Node.JS code over cleaning up LLM Node.JS slop because the optimist in me hopes that the human who wrote bad code can at least learn something and become better over time. After all we all have started with writing garbage, I know that I have.

On the other hand, I guess I should find a job where I don’t have to touch web development with a ten-foot pole because it’s probably not getting better.

Our company is currently looking for a new programmer and we've interviewed a few so far. I don't want to generalize but it really seems that a non-negligible part of the younger ones at least tries to use LLMs to make up for a lack or experience, and that really shows.

I normally don't like doing programming challenges during an interview because they have little to no real-world connections, but I've been throwing small questions around lately just to see what people do, and how they approach them, and there's a subset of people who will say, "I would ask ChatGPT now" in those scenarios.

I haven't met a vibe-coder in real life yet, but I'm afraid it's only a matter of time.

I am fairly certain that their stupid AI stance at least played a role. I’m part of a largish writing community, where at least several hundred people did NaNo each year, a good part of them donated too. Last year, no one partook, instead we did our own internal thing.

If this happened across other communities too, and I wouldn’t be surprised if it did, they must have felt it financially.