An essay argues that calling user input to generative AI a “prompt” imports a computing assumption into linguistic territory: that questions are neutral instructions. It argues every question carries ideology, and proposes “statement-command” instead.
https://doi.org/10.25189/2675-4916.2026.v7.n3.id944
18 Comments
Nice promotion, I’m a statement-command engineer now!
> It argues every question carries ideology,
Effectively true. Just like there is no such thing as no bias and non political.
I just call it a request. And usually frame it as such.
“Can you tell me…”
“Ok, that was a great instruction, but I’d like to…”
>statement-command
Arent those called [imperatives?](https://en.wikipedia.org/wiki/Imperative_mood)
An important point this raises is that “prompts” to AI provide themselves as further training data for the AI. Prompts aren’t just searching a database from which they’re separate; they offer themselves as a new part of the way the AI thinks.
“Prompt” is one syllable. “Statement-Command”is two words and a hyphen. Even if it’s technically correct, this proposal is dead in the water.
This is kind of missing the forest for the trees isn’t it? In the conversation about LLMs, the term *AI* has a lot more baggage and is a lot more misleading than the word *prompt*.
I agree we should be careful in how we name and talk about things, and I have a strong suspicion that the people who pushed LLMs to the mainstream deliberately did the opposite.
Is there nothing else to do research about anymore? Seriously, i think studying variance in beanie babies makes more long term sense. Or growth rate of chia pets. Utterly useless and irrelevant.
We are prompting actions with direction (command) and requests (statements). Even if we don’t say “I have a question about”, the chatbots infer it as a request for information.
Even if “statement-command” is deemed accurate, I don’t understand the different.
You prompt it to do something and it does it. It doesn’t do it unprompted.
Wow, this is very pedantic. I don’t think it matters that much.
Prompt in computer terms is only like the third definition of the word. It has long been used in other contexts to describe something that provokes or inspires a response, which seems a perfect match for what it is.
It ain’t broke. If you want, teach people what the word actually means.
Split the difference and call it “command-prompt” oh wait…
1) Using Grok for your study is intentionally choosing the model known to produce the worst results and to be most subject to intentional bias of all sorts.
2) I’m not entirely sure this study wasn’t written by Grok.
The title is not really an accurate representation of the paper’s contents – the paper instead mostly argues that the behavior of xAI’s Grok model is highly conditioned by variables the user in principle knows about but in practice is liable to forget about (memory, free vs paid tier, etc). Also, the paper’s contents are probably better served in r/philosophy.
In the early days of LLMs you could get world class answers from them. The technique was to tell them to respond in a hostile manner like a debate, and even be slightly insulting. This (somehow) caused its answers to be more truthful. It was great.
This is no longer possible, as the major tech companies have neutered their LLMs into being happy, helpful servants.
Instead of “fast” and “deep think” modes. They should offer servile mode and brutal mode. In brutal mode, the LLM would give you the straight facts regardless of feelings. In some topics, you need the bot to be correct, not be sociable.
I just call it “Input” and “output” from a model.
I think that encompasses it better. Your input is also more complex; it’s the history and context as well as the most recent direct input to be replied to. Output is generated by feeding that into a complex model with some amount of randomness and what is essentially a recursive prediction until you get an end result.
Some people gain success by working hard to create something new. Others try to feign success by renaming someone else’s hard work. This is the latter. It’s sad and should be ignored.