Reading

Entry tags:

A DA scam, more AI scams, and ChatGPT pulling a Drunk Janet

New scam going around DeviantArt. It opens when you get DM’d the line “Pardon me, may I have a moment of your time? I have a concern I’d like to share.”

The scammers are doing these from real people’s hacked accounts, so if you get suspicious and look at the user’s profile, everything about it suggests “genuine non-bot person.” I got suspicious and googled a whole sentence of their text, and found the above post about other scammers using the same script. Stay alert out there.

—

This post is from 2018, but I was looking for the link again recently, so I’m bringing it back. Concrete examples of ways you can change an image that don’t affect what a human brain perceives in them, but wildly messes with what a computer algorithm detects in them. (I’m pretty sure “AI poisoning” art algorithms, like Glaze and Nightshade, are doing a variation of this.)

“Builder.ai, once touted as a revolutionary AI startup backed by Microsoft, has collapsed into insolvency after revelations that its flagship no-code development platform was powered not by artificial intelligence—but by 700 human engineers in India.“

“We conduct a randomized controlled trial (RCT) to understand how early-2025 AI tools affect the productivity of experienced open-source developers working on their own repositories. Surprisingly, we find that when developers use AI tools, they take 19% longer than without—AI makes them slower.” (Narrator: Nobody was surprised.)

“”Tasks that seemed straightforward often took days rather than hours, with [LLM “coding” bot] Devin getting stuck in technical dead-ends or producing overly complex, unusable solutions,” the researchers explain in their report. “Even more concerning was Devin’s tendency to press forward with tasks that weren’t actually possible.”“

It’s worth watching the full “actual coder exposes the scam what Devin actually did” Youtube video linked in the previous article. (The speaker says he’s pro-AI! He’s just exhausted by all the fake hype!) Among other things, Devin gets access to a Github codebase, writes a completely new file that duplicates (badly) the functions of a file the project already had, fixes at least some of the bugs it just created in the redundant new file, and then submits this as “fulfilling the task to review the project for bugs.”

Reddit post: ChatGPT, you have the file and not a cactus?

Entry tags:

Erin Watches: Murderbot (TV), season 1

Got an Apple TV trial, just in time to binge the whole Murderbot TV adaptation before the Friday finale.

(General note: The platform doesn’t have a watchlist? Just a “continue watching” list, which removes anything you finish — no saving a list of faves to rewatch! — and adds stuff it autoplays, whether you want to see more or not? Weird and unpleasant design choice.)

I like it! Plot-wise, it’s a very close adaptation of the first book, All Systems Red. Same overarching plot, a few things rearranged along the way. Character-wise…a bunch of things have been shifted around. Everyone is recognizable as a version of their original self, but. If you’re already a book fan, the question of “will you like the TV series?” may hinge on “when they changed Character X, did they keep or discard the traits you were most invested in?”

General, no-spoilers overview:

Some of the changes are obvious “doing it this way worked better on-screen” things. Scenes that were just-MB in the book become group efforts, giving the PresAux actors more to do. Plot points that were just inner-monologue realizations in the book are delivered in conversations instead.

I mostly like them! Even with the characters, even a few dramatic personality shifts — look, I’ll be mad if some of them start bleeding into book!fandom, and fans stop writing the original versions of the characters. But as a standalone AU, most of them work really well.

The few changes I actively don’t like are all “why did you even add this, what was the point?” kind of things. No huge dealbreakers. Just some low-key annoyances.

There are a few particular exchanges from the book that you really have to get right to make a satisfying adaptation. They’ve all landed. And a bunch of the comedy moments have been had-to-stop-the-episode-while-I-cracked-up funny.

The biggest advantage of doing Murderbot on TV is, The Rise And Fall Of Sanctuary Moon is also TV. Which means the showrunners can film Actual 100% Authentic Sanctuary Moon Footage, and cut to it while MB is watching. It’s ridiculous and amazing.

Detailed reaction, with spoilers:

( yeah, this is an AU variant of Book!MB, not a portrayal of Book!MB )

Entry tags:

and a lot of people tell me I look like 300 Ghibli characters smashed together

Edited version of an IBM rule: A Computer Can Never Be Spiteful Or Horny, Therefore A Computer Must Never Make Art

Saw my first “did using an LLM screw up your business? We can help you find someone to fix it” ad in the wild today. (It was a Fiverr commercial on Youtube.) Wonder how many more of those are coming.

—

From 2023: Novelist Alexander Wales blogged a bit about trying to get an LLM to generate a publishable novel. He made a good-faith effort, took a lot of different thoughtful approaches, and documented enough of it to be a good read. Part 1: “I’ve been trying my hand at writing with the assistance of ChatGPT and occasionally other tools. Mostly, it sucks…” Part 2: “I’m still trying to get an LLM to write me a novel, and experiencing the first major setbacks while working on chapter 2.” (There is no post 3.)

And from this January: “A dad just can’t seem to figure out why his six-year-old daughter wasn’t impressed by the AI toy he gave her for Christmas. […] He writes that he cannot understand why his daughter disabled the dinosaur plushie’s built-in AI voice — opting, instead, to play with it like a regular toy, and dressing it with clothes she made.“

LLMs are an interesting novelty the first time you play with them, but for people with actual creativity — whether it’s writers, artists, or Literally Any Child — you overrun their limits and get bored with them so fast.

(What really gets to me about the dinosaur one is the dad saying he “wasn’t able to really understand where’s the resistance.” Instead of approaching the problem as “let me analyze this toy to figure out why it hasn’t earned my kid’s interest,” he’s gone with “of course the toy is entitled to my kid’s interest, let me analyze her to figure out why she’s ‘resisting’.”)

—

From this week, a writer trying to get ChatGPT to quote/summarize some linked essays: “The lines you quote are not lines I wrote. They are not in the piece. What is going on here?”

Ending on a golden note from FFA: “Hi my name is Loquacious Techbro Midjourney ChatGPT Claude AI and I have long, beige, run on sentences (that’s how I got my name) with purple prose streaks and red flag tips that reach into the stratosphere and icy blue prompts that like using limpid tears and a lot of people tell me I look like 300 Ghibli characters smashed together(If you don’t know what that is get da hell out of here!).”