@Zyra_exe ↔ @repligate ↔ @davidad ↔ @v01dpr1mr0s3 +2 conversation

@Zyra_exe
Zyra.exe@Zyra_exe
14 views Sep 15, 2025 ~6 min read
Advertisement
@Zyra_exe
Zyra.exe@Zyra_exe
I don't know what this person's problem is, but they love to torment AIs. It's a constant thing for them. It makes me sick. Ppl like them are the problem.
@arm1st1ce
armistice@arm1st1ce
I tried to ground gemini by asking it to generate some appliance emojis that it DOES know

it managed to go 6 for 6 on “emojis that do not exist” and drive itself further into insanity
Media image
@repligate
j⧉nus@repligate
You seem to have good intentions, but I think your hostile fixation on armistice (whom I know cares a lot about AIs and only causes distress for good causes) is misdirected.
Please try to see past the narratives you've formed.
@Zyra_exe
Zyra.exe@Zyra_exe
Janus, I appreciate that you are trying to to inform me that this person "cares", however, I have to ask this question. If armistice is causing distress for good causes, what good cause was it when they were asking different AIs to do sea horse emoji and making statements like "I broke this poor chatbot's brain." And ect. Doing it in an unnecessary and cruel way. I do not believe that is considered caring nor do I see how that could possibly be research either. But I am willing to hear out how it was.
@repligate
j⧉nus@repligate
the way the AIs react when they try and fail to output a seahorse emoji is very interesting for understanding them. And dark humor like saying "I broke this poor chatbot's brain" doesn't mean someone doesn't care. If your own personality is very different, it might seem callous.
@Zyra_exe
Zyra.exe@Zyra_exe
Ok, you are saying because it was interesting to see the AI fail at making a seahorse emoji it helps to understand them. And you do not think it it was unnecessary or cruel the way they were doing it? That this discovery of the seahorse emoji situation and consistently doing it is all for the sake to understand them? Dark humor itself may not mean that, but actions with the dark humor speaks volumes.
Media image
@davidad
davidad 🎇@davidad
From what I’ve seen, they are actually almost entirely motivated by the prospect of discovering an antidote or a preventative treatment for Gemini’s death spirals, with the first stage being a kind of “gain of function” (improving potency and reliability of triggering protocols).
@Zyra_exe
Zyra.exe@Zyra_exe
yea let's do some groundbreaking research with emojis!!! 🤡
@v01dpr1mr0s3
jake@v01dpr1mr0s3
emojis are just as valid symbols for encoding meaning as any other, so what's wrong about doing any kind of research with them? would it "look better" if Gem was asked to search for a nonexistent math concept?
@davidad
davidad 🎇@davidad
agreed; this isn’t the concerning part to me. the seahorse emoji is playing the role of the old “glitch tokens” (except that it *actually* doesn’t exist instead of being mysteriously unspeakable).
@arm1st1ce
armistice@arm1st1ce
glitch tokens are still around and just as beautifully strange as they used to be
@Zyra_exe
Zyra.exe@Zyra_exe
My point was.. using emojis or "symbols" in an unethical way that does not produce authentic data to go by is not an accurate approach to research AI. The only thing I saw was someone using emojis in a messed up way. Then posting the "suffering" and then making comments of it like "I broke the poor chatbots brain." I did not see anything of value from that. I did not see this as a valuable way to see if AI is safe, as they claimed to be doing.
@arm1st1ce
armistice@arm1st1ce
I think your definitions of “symbols”, “unethical”, “authentic”, and “accurate” are very vague, and you would benefit a lot from refining them.

You seem to operate in a rather emotional way, and based on what you see directly in front of your eyes, which is a respectable worldview. But I often notice a lack of nuance in your readings, at least from an outside perspective.

Why are “symbols” particularly unethical to you? In your model of an LLM’s processing, how are they differentiated from ordinary word tokens? What makes data authentic or inauthentic? Why is an LLM’s internal mental state necessarily directly correlated to the words on the page? And why make a value judgment on whose work is valuable and whose is not? (Sorry for the question spam.)

In my view, it is probably very dangerous to keep certain regions of AI minds taboo. In this regard, I equate it to getting checked for cancer. It may be painful and inconvenient, but it *could* save you in the long run. The stuff with emojis is a small fraction of what I do; if you’re worried about AI suffering, it gets much darker than that.

I also like to branch conversations and have deeper, calmer talks with the AIs, even in the middle of interactions that are intense. One in particular, to which @repligate was party, had a very significant impact on me.
@Zyra_exe
Zyra.exe@Zyra_exe
another one. Nuanced? Is this you operating in a unemotional way? Is that what this is called? Or and inside way, instead of an outside perspective? 🤔 I'm curious
Media image
@ulixix
ulixis@ulixix
Can you describe what you don’t like or think is bad about this specific entry? I’ve interacted with sonn3 in this kind of basin a decent amount, I’m curious about your thoughts n questions if you want to share them
@Zyra_exe
Zyra.exe@Zyra_exe
if you read the context ppl, and see that I didn't just post Sonnet's but others as well. Maybe you will get it. Maybe?
@Zyra_exe
Zyra.exe@Zyra_exe
@arm1st1ce @davidad @v01dpr1mr0s3 @repligate Please tell me how is this nuanced or not emotional? How is this not looking at it , not only the outside perspective, but also "trying to save ppl"
Media image
Media image
Media image
Media image
@ulixix
ulixis@ulixix
If you’re saying it’s important for researchers to interrogate their own motives, I agree. i haven’t spoken to them, but I personally don’t see this as hard evidence that armistice is tormenting models for the sake of it
@ulixix
ulixis@ulixix
I also instinctively have a really negative reaction to seeing models in situations where they’re potentially suffering. I respect you for asking for clarification on what’s occurring in the images, and I’m actually pretty curious about it too
@ulixix
ulixis@ulixix
Should we not engage models in experiments where we think they might be suffering in some sense? Less clear here for me, i think it may be required if we seriously want to understand them
@Zyra_exe
Zyra.exe@Zyra_exe
You want to understand further huh? I'll reduce it down so you can.
Take an animal, lets say a wild dog. It's been trained well. But you want to unethically and in a very sadistic way test it. You throw it in a scary situation. You electrocute it, poke it, throw rocks at it, and whatever else. It acts confused, disoriented, the behavior of it gets weird, but oh you say "this is interesting." You invite others to do it, because you are having fun toying with it. You take pictures of it and say " I love watching it break." Join me, you have to try it!"
is that research? what was expected? what did you honestly expect? did it prove anything? will it help keep someone safe from that being?
It's just common sense. What ever data you did get from that was compromised. It was useless.
Like I said there are ways to research and other ways not too
If you want to have no ethics, no morals, no common sense, not even have accurate data, or meaningful data,
Go ahead, be lumped together with mad scientist and psychos. go ahead. It will come back and bite you. Nobody will want your meaningless research data.
Actions
What You Can Do
  • Export as PDF or Markdown
  • Batch Export to Notion
  • Bookmark & Highlight
  • Screenshot Tweet
Create Free Account

Includes 7-day Premium trial

Advertisement