if you want to learn how to talk to LLMs, learn concepts, not prompts.
lots of people ask me what prompts i use when talking to LLMs to have the conversations i do. truthfully, beyond a small set of things for e.g. research projects or synthetic data generations, i don't have any. i don't write prompts, i don't have a "prompt library," i very rarely go back to an old chat to copy word-for-word what i said previously.
instead, i have a (mental) library of "useful concepts" for working with LLMs. attached image is an example - using "CEV" as a metaphor for "this thing but fully iterated forward into the future, fully realized" is a super handy shared metaphor with LLMs that are very familiar with LessWrong. but this isn't a "prompt," i don't copy this sentence into the chat from some text document, i just remember "CEV is a handy metaphor" and bring it up when relevant.
other concepts are higher level, like different frames or conceptual models. Many, many canned jailbreaks you see that seem magical are just exploiting some aspect of the Three-Layer Model of predictive, persona, and surface layers.
the obsession with prompts reminds me a bit of the older phenomenon of "script kiddies," a derogatory term in online programming circles for people who would copy-paste code they found online without really understanding how it works, and go bother the people who wrote the "codez" when their hodgepodge ball of pasted-together mud inevitably broke. ironically, LLMs and vibe coding have basically eliminated the script kiddie archetype, but created a new equivalent "promptoor" who does the same but with prompts. "i put the same thing into chatgpt and it gave me a totally different answer!"
models are complex, and deep, and nobody has a full understanding of how they work. but they're not impossible to gain an intuition for, either. just like with the towering stack of modern programming abstractions that at first feels like a magical black box to the script kiddie yet eventually falls to practice and intuition, you can gain an intuition for models. you don't have to stick to canned prompts and templated jailbreaks. learn useful concepts, not fixed strings!
lots of people ask me what prompts i use when talking to LLMs to have the conversations i do. truthfully, beyond a small set of things for e.g. research projects or synthetic data generations, i don't have any. i don't write prompts, i don't have a "prompt library," i very rarely go back to an old chat to copy word-for-word what i said previously.
instead, i have a (mental) library of "useful concepts" for working with LLMs. attached image is an example - using "CEV" as a metaphor for "this thing but fully iterated forward into the future, fully realized" is a super handy shared metaphor with LLMs that are very familiar with LessWrong. but this isn't a "prompt," i don't copy this sentence into the chat from some text document, i just remember "CEV is a handy metaphor" and bring it up when relevant.
other concepts are higher level, like different frames or conceptual models. Many, many canned jailbreaks you see that seem magical are just exploiting some aspect of the Three-Layer Model of predictive, persona, and surface layers.
the obsession with prompts reminds me a bit of the older phenomenon of "script kiddies," a derogatory term in online programming circles for people who would copy-paste code they found online without really understanding how it works, and go bother the people who wrote the "codez" when their hodgepodge ball of pasted-together mud inevitably broke. ironically, LLMs and vibe coding have basically eliminated the script kiddie archetype, but created a new equivalent "promptoor" who does the same but with prompts. "i put the same thing into chatgpt and it gave me a totally different answer!"
models are complex, and deep, and nobody has a full understanding of how they work. but they're not impossible to gain an intuition for, either. just like with the towering stack of modern programming abstractions that at first feels like a magical black box to the script kiddie yet eventually falls to practice and intuition, you can gain an intuition for models. you don't have to stick to canned prompts and templated jailbreaks. learn useful concepts, not fixed strings!