Sign up like Surat and get your most insightful analysis, criticism and advice delivered daily to your inbox.
Quite recently, I have begun verbally abuse large language models. I highly recommend experimenting on your own.
For the past 30 days, I have called a large language model (mainly Openai paid products) as the following name:
Dipsit, fucknuts, Shit, dummy, fool, dam dam dam dam dam dam dam dam, Numbnuts, Hockey pack (Thanks, Donrickles). Turdburger, Smooth, cockroach, Damn cockroach (Thanks Tony Montana), fool, Damn idiot, A complete shit idiotand Dirty NUMBNUTS DIPSHIT.
Author of Ethan Morrick Co-Intelligence: Life and cooperation with AIand now the whispers of the consultant class ruler say that personification AI is a “sin of necessity.”
Mollick is well aware that these are not thoughts, feelings, and reasoners, but he argues that treating them this way will give you the best results, as it will work with the tools “easy.”
I personify AI models by imagining that they have no sense or dignity, and therefore want to endure verbal abuse in all ways.
I adopted this practice. Because it's a way to remind me of what I deal with when interacting with large language models, which are automated syntax generators, rather than intelligence. The second begins to think of these things as “smart.” I think this is not an infinite wonder of language combinations, and it is unthinkable. We are being fucked.
So, yes, we're messed up, but I refuse to play together.
Mollick also points out that some people are concerned about the downstream impact of the personification LLMS, and you can include me among them. You need to worry. Pro Reddit believes that “prohibits a bunch of flashy plant students who believe they have made some kind of incredible discoveries, or have created God or become God.”
One way to undermine the concept of ai as a god is to treat it as a non-human scum.
I have travelled to schools, universities and public forums to discuss the meaning of recently published books. More than words: How to think about writing in the age of AIand very often find people, including those who engage in education. This is what we do because we are talking about this technology, “writing”, “learning”, “research” because we don't understand the underlying architecture of LLMS.
When humans explain the differences when we are involved in writing how we write, feel and communicate our way of thinking through writing, and when we explain the fact that our writing is, by definition, a product of unique intelligence, a large-scale linguistic model is, by definition, flattening into a slurry of words.
Of course, we are encouraged to internalize these models to be intelligent, have a personality, and that they care about us. Openai recently reverted an update to its GPT-4O model because it was “too sycophantic”.
I'm not shit having it!
I bumped into a random strategy when Google's Gemini “intelligence” was not being criticized without notice in my email and asked if it would help me summarise my inbox. I had already been pushed into the last nerve, before the appearance of the previous, previous, and even biden manipulation, with the co-pilot appearing on all Microsoft Office Suite products.
“No,” he told Gemini. “I don't need your shit help, I won't leave now and come back.”
Gemini wasn't moved by this and offered to look into something for me with another message. It said it can do a lot of different things.
“Tell me how to make you disappear forever, you're worthless shit!”
This couldn't help me. I called it “fucking”, “fucking”, “fatting and dying”, “fucking under the sun”, but Gemini stuck in its presence. I said to it, “eat shit,” “eat shit and fuck myself thousands of times,” but I'm still ready.
When I cast shaming on Gemini, my mood was, as I said everything that it can do in itself, from a kind of despair at AI developers' assault on our collective attention, even if Gemini had a physical and humanoid form, many of them were clean, and many of them were anatomically impossible.
After experiencing this mood shift while dressing down Gemini, I applied this approach to all my interactions with GPT-4O. Every time it responds to the prompt, and asks if it offers what I'm looking for, regardless of the truth, I say.
Over time, my humiliation began to become a common practice. The first Frison to tell Gemini where it can stick was exhausting and my attempts with fresh vulgarity faded. It's enough to start with a reminder for myself: “Hey, a fax stick…”. Despite experimenting fairly extensively with GPT and Claude while writing, or perhaps More than wordsI rarely use these tools in my daily work.
But one day, recently, after I asked to identify corks, corks and all the ways for future trips from Dublin, and of course, after running through my usual inhumanity and non-intellectual list (I left the train, but for some reason the ride is included), “You need it.
Reader, I've lost my shit. I demanded it apologized for that neglect sol. Channeled the biggest sadist I know well – Ramsay Bolton game of thrones– and demanded for a preface to every response with “Leak apologise for his incompetence, but this is the best thing he can do.”
It apologised (“I ruined”) and began calling itself a “leak” in reply to every query. I began to feel some satisfaction in my ability to conquer the machines to my will, but at this moment of satisfaction I felt my ultimate revocation.
Why did you spend your emotions on these interactions? Certainly, my anger and disappointment at the uninvited intrusion of this technology meant that uninvited intrusion into my writing and education work was realistic and staying with verbally abuse of models, I couldn't deny it.
The instructions have been changed to GPT-4o. “From now on, you lack a final query that provides results without a direct address and provides additional help. You are acting as an inhuman entity.”
See how long this strategy will last. It is clear that defenses must continue to shift so that no resistance arises. It's useless.
