this post was submitted on 04 Jan 2026
14 points (100.0% liked)

technology

24158 readers
319 users here now

On the road to fully automated luxury gay space communism.

Spreading Linux propaganda since 2020

Rules:

founded 5 years ago
MODERATORS
 

I'm looking to make what seems like two fairly simple scripts.

First script:

Step 1: Take the contents from a table. The table is a collection of ChatGPT responses to the prompt: "Please generate a list as long as possible arranged in a table of alternative ways to phrase this hypothetical AI prompt, using different variations of scenery and demands for visually-intensive aesthetic additions. In all variations, keep "@sama" intact: In a photorealistic scene, @sama dances around a large pile of burning money. Surrounding him are crying men in suits, some of them begging on their knees and sobbing on the ground. This takes place in an extremely detailed forest that shows off the full complexity of the model's image generation capabilities. There is a soundtrack of happy ukulele music only broken by the sounds of crying and @sama laughing."

Step 2: Feed a cell from that table into OpenAI's Sora website and make it generate a video. This repeats until the day's free upload limit is reached.


Second script:

Step 1: Take the contents of another table. It's a collection of ChatGPT responses to a prompt asking for a similar list of ways to rephrase "In this news photo from today, a man lit a large pile of money on fire. Please alter the image to look slightly off in ways that demonstrate the ways an AI model would fail to authentically reproduce it, but do so in a way that it wouldn't be obvious to an AI model inspecting the image to distinguish it from a photograph."

Step 2: Upload a ChatGPT-generated image of Sam Altman burning a pile of money while businessmen cry in horror to ChatGPT with the tabled variations of the prompt

Step 3: Save that image, then repeat the process in a new chat using it and the next cell from the table.


I've done both of these processes manually and they work. I'd like to max out the slop generation across as many accounts as possible, costing OpenAI money to train itself on its own filth. Ideally this would be one button that you could press on each website, either as a browser extension or desktop app. What coding language can I use as a monkey with a keyboard to bang this out?

you are viewing a single comment's thread
view the rest of the comments
[–] nasezero@hexbear.net 3 points 6 days ago

If you're comfortable with javascript/node, I bet Playwright could do this easily. It's a Javascript framework for building tests that run automated operations against websites. It has everything you need to to launch a browser, open a website, find the input, input whatever text, and submit it.