Some claim they would rather read your messy prompt to an LLM rather than the output. (here, here, some are now assigning more value to imperfect English over AI slop).
What if I took this literally? What if instead of trying to polish my writing to what I think is accessible and meaningful to an audience, I just posted the prompts?
In my experience, as I’ve written about here, the simple act of writing a prompt often negates the need to ask an LLM in the first place.
My real question is: will readers get more out of my writing if I write as instructions for an LLM or does the direct approach work better?
I think that the direct approach works better because it requires the reader less effort to understand.
As for the issue of prompt to LLM vs LLM output, the LLM output is supposed[1] to be unlikely to contain actually interesting novel insights and likely to have major issues like @MikkW’s essay about the egregore food chain where Gemini, in addition to AI slop,[2] wrote this piece of mathematically clear BS:
Gemini’s BS
We can model this egregorical predation utilizing a cybernetic adaptation of the classic Lotka-Volterra equations:
dx / dt = αx - βxy
dy / dt = δxy - γy
Where x represents the population of decentralized, high-vitality meso-egregores (the prey) and y represents the mass of the centralized apex aggregator (the predator). The coefficient \beta represents the rate at which the monopoly successfully co-opts and acquires the localized networks, while \delta reflects the metabolic efficiency with which the monopoly converts that absorbed trust into its own structural growth. If \beta is allowed to remain unchecked by antitrust interventions or structural friction, the apex predator will inevitably strip-mine the ecosystem of all localized vitality.
The Lotka-Volterra’s equation has nearly all trajectories become circular, not driving prey to extinction. Additionally, I am not sure of the extent to which egregores even can be called a food chain and not rivals for humans’ attention to different-scale coordination tasks.
The essay ends with the phrases “By strategically nurturing these localized, symbiotic entities, you are not merely engaging in a lifestyle choice; you are aggressively cultivating a decentralized immune system. You are building the resilient rootstock upon which the profound, unvarnished vitality of human life can sustainably bloom,” which IMHO resemble GPT-4o’s flattering slop.
What if I, literally, just show you the prompt?
Some claim they would rather read your messy prompt to an LLM rather than the output. (here, here, some are now assigning more value to imperfect English over AI slop).
What if I took this literally? What if instead of trying to polish my writing to what I think is accessible and meaningful to an audience, I just posted the prompts?
In my experience, as I’ve written about here, the simple act of writing a prompt often negates the need to ask an LLM in the first place.
My real question is: will readers get more out of my writing if I write as instructions for an LLM or does the direct approach work better?
I think that the direct approach works better because it requires the reader less effort to understand.
As for the issue of prompt to LLM vs LLM output, the LLM output is supposed[1] to be unlikely to contain actually interesting novel insights and likely to have major issues like @MikkW’s essay about the egregore food chain where Gemini, in addition to AI slop,[2] wrote this piece of mathematically clear BS:
Gemini’s BS
We can model this egregorical predation utilizing a cybernetic adaptation of the classic Lotka-Volterra equations:
dx / dt = αx - βxy
dy / dt = δxy - γy
Where x represents the population of decentralized, high-vitality meso-egregores (the prey) and y represents the mass of the centralized apex aggregator (the predator). The coefficient \beta represents the rate at which the monopoly successfully co-opts and acquires the localized networks, while \delta reflects the metabolic efficiency with which the monopoly converts that absorbed trust into its own structural growth. If \beta is allowed to remain unchecked by antitrust interventions or structural friction, the apex predator will inevitably strip-mine the ecosystem of all localized vitality.
The Lotka-Volterra’s equation has nearly all trajectories become circular, not driving prey to extinction. Additionally, I am not sure of the extent to which egregores even can be called a food chain and not rivals for humans’ attention to different-scale coordination tasks.
However, I do expect such output to contain useful criticism or the LLM’s perspective on the issue.
The essay ends with the phrases “By strategically nurturing these localized, symbiotic entities, you are not merely engaging in a lifestyle choice; you are aggressively cultivating a decentralized immune system. You are building the resilient rootstock upon which the profound, unvarnished vitality of human life can sustainably bloom,” which IMHO resemble GPT-4o’s flattering slop.