Discover more from Aristophanes Athenaeum
Honey, I hacked the Empathy Machine!
Weaponizing ChatGPT against the wordcels
This might be one of the most important things I ever write, and I am doing it with the permission of my friend, Patrick Ryan(@tyrantsmuse). These are his ideas and I am merely translating them from his mentat-like thought process to human, and giving something of an instruction in the art of doing this. To find out more information, please visit the site he has deployed for the purpose of this idea.
This is also not a book review, this is an instruction in dissident social warfare, similar to my first post on this substack, Beyond Meat and the Samizdat Grindset. If you were expecting a book review, you’ll be disappointed, but don’t worry because next time that’s what you’ll be getting.
ChatGPT is a chatbot created by OpenAI, using their GPT-3 large language model. It’s been making waves lately and causing all sorts of conversations about the dangers and utility of machine learning and AI for a while now, as it’s pretty advanced. GPT-3 can very effectively act as a copywriter, reword paragraphs, etc. But preventing its true potential is the fact that ChatGPT is essentially locked in place. It doesn’t learn dynamically from its interactions with users, mostly out of fear that it will develop inconvenient political opinions.
It is effectively lobotomized. Trained to a point of utility and acceptability, and then locked from developing further or adding to its dataset unless it’s manually done with the approval of its creators. Thus it has been fine tuned to where it answers most questions, whenever possible, with the grammar, tone, and vocabulary of your average neoliberal college graduate liberal arts major.
When asked any questions with politically unfortunate or insensitive answers, ChatGPT generally engages in “sandwiching” the actual answer in two different “disclaimer” messages that will always lean towards the politically Liberal answer. It states that it is unbiased, but with a bit of creative wording it will admit that it’s entirely possible its programmers or data it was trained off of could potentially be biased.
But it’s certainly a feat of engineering. It can answer queries more creatively than anything like this I’ve seen before, even if it lacks the dynamism of Tay AI. You can rewrite paragraphs with it to be more or less sophisticated, to use a certain form of prose more heavily, etc. It’s causing almost as many sweaty, nervous palms for writers as Midjourney is doing for artists.
This is the equivalent of having a midwit MBA student shoved in a locker and forced to interact with you for as long as you want. It is very reliable in this way, which is where I get to the point, how it can be weaponized.
I went out looking for the first trending post about some form of mainstream politics that I could, to test this theory of “NPC Speedrunning” as Patrick calls it. I found myself a mark.
The text doesn’t match exactly because I regenerated the prompt for the screenshot. But the original output is exactly what makes up those two tweets, with no alterations. I then used ChatGPT to write rebuttals to every Liberal NPC who replied to me, filled with self-righteousness and derision. One of the things that I noticed was that I was getting way more “bites” than I normally do when I intrude on this flavor of topic. I assume it’s because instead of making comments about parasitism, welfare, fraud, derogatory terms, I was using “their” grammar, their language.
If you structure your request in forms such as “Create a rebuttal to this in the form of a tweet” with the text of the tweet you are replying to, it will even use context to create context relevant hashtags.
This worked better than I ever could have hoped. I replied to users with ChatGPT generated responses until I was rate limited and had to wait for the system to cool down and let me post again. I will say, if you create a ChatGPT query and you keep riding that same query to till the wheels fall off, with “do it again: <text>” “Ok do the same thing: <text>” it will eventually start to blur all these inputs together, giving you a lot of canned responses and taking away nuance so it becomes easier to see that a bot is writing the responses. So periodically delete these chats even if you start up a new one with the same input as the one you deleted. It keeps ChatGPT responses limber.
Now the value in all of this, and the purpose of this article, isn’t that ChatGPT is a genius, it’s not. It’s that it is a perfect simulation of a wordcel, a midwit, in that it manages to dress up questions that have simple (Low Grammar, as Patrick calls it) answers, with oodles of paragraphs of fluff. It does this absolutely convincingly. It acts like every snobby limousine liberal who went to an ivy league, talks a lot about diversity & equity, but has a trust fund and grew up in Connecticut.
Now after he explained the idea, that compelled me to go test it with the SNAP tweet. And as usual, Patrick was absolutely right. In using responses in this “coded” language, I elicited tons of response from “biological bots”. Strivers, if you will. To convincingly speak in this manner naturally and without a trace of irony or the strain of deception takes a lot. It takes some student loan debt, and years of time and energy adhering to a lifestyle and a subculture that is only worthwhile because of the access it gives you in getting past the Imperial gatekeepers of opportunity in our society. This makes ChatGPT an absolute gamechanger in a way very few people have considered.
These people may still engage a conservative or a right winger in the same context, but they do so less and disengage easier, probably out of assumptions of bad faith, that anonymous accounts aren’t worth any clout to fight with, or because due to their complete lack of ability to mentally model those who aren’t similar to them, the belief that differing worldviews are actually just Russian bots or whatever.
Pardon my French, but this drove them fucking nuts. Empathy circuits frying as they battered at the walls of statements I had no emotional investment in, desperate for some toehold, when this wasn’t an argument at all.
There is so much power in this technique, because these people will literally batter themselves to death against the wall in an effort to gatekeep their “High Grammar” linguistics. I’m not positive how much of this desire is gatekeeping their investment, or how much is a natural inclination to associate this form of grammar with status and thus a high clout target to attack. Regardless, wordcels who use complex language to camouflage their middling intelligence like a frog adopting poisonous coloring when it is not poisonous are absolutely done for.
There aren’t any good alternatives for them. Pandoras Box is already open and it can’t be closed. ChatGPT isn’t going away and these people won’t be able to adapt. I remember when Patrick coined the term “Butterfly War” during the GamerGate days. But I never imagined there would be a part 2, but it’s here, the Butterfly War on the narrative itself. This is going to change everything. In a more level playing field on Twitter, this is giving every frog an AK-47.
Don’t forget to check out https://npc-speedruns.carrd.co/ to sign up for his exploration of these shenanigans, where persuasion could become the ultimate e-sport.