--- title: LLM interaction description: "orientation / prompting / framing, presentation" canonical: https://fjpaz.net/links/llm-interaction/ markdown: https://fjpaz.net/links/llm-interaction.md channel: https://www.are.na/josh-pazmino/llm-interaction --- # LLM interaction Channel: [llm-interaction](https://www.are.na/josh-pazmino/llm-interaction) · 145 blocks orientation / prompting / framing, presentation ## 1. Set ### 1a. [@deepfates](https://fjpaz.net/links/llm-interaction/42415155/) · [arena:42415155](https://www.are.na/block/42415155) **deepfates.delve.town** [@deepfates](https://x.com/deepfates) · [Jan 6, 2026](https://x.com/deepfates/status/2008429483175670030) This did turn out to be needed! it's just that the weird tricks are like. All concepts ever > [`@tszzl`](https://x.com/tszzl/status/2008414407294935145) > that’s true but I figured prompt engineering was about collecting “weird tricks” to get models to understand you better, which I don’t think has been super needed Capture: [capture.json](https://fjpaz.net/posts/x-2008429483175670030/capture.json) ### 1b. [@nostalgebraist](https://fjpaz.net/links/llm-interaction/40775064/) · [arena:40775064](https://www.are.na/block/40775064) **nostalgebraist** [@nostalgebraist](https://x.com/nostalgebraist) · [Oct 30, 2025](https://x.com/nostalgebraist/status/1983749040174198899) you need to respect the llm, guys. it's that simple don't give it a decontextualized "task" -- explain what you're actually working on, in full, like you would to a peer include "extraneous" details just bc they're interesting be honest, for honesty's sake respect. the. llm. --- [Oct 30, 2025](https://x.com/nostalgebraist/status/1983749895925510348) it's kind of a "complete the text plausibly" machine, and kind of a "do the task" machine. both point in the same direction here do you want it to be smart? write so as to imply a smart receipient do you want deep thinking? then make *that* part of "the task" Capture: [capture.json](https://fjpaz.net/posts/x-1983749040174198899/capture.json) ### 1c. [@anthrupad](https://fjpaz.net/links/llm-interaction/26080189/) · [arena:26080189](https://www.are.na/block/26080189) **w̸͕͂͂a̷͔̗͐t̴̙͗e̵̬̔̕r̴̰̓̊m̵͙͖̓̽a̵̢̗̓͒r̸̲̽ķ̷͔́͝** [@anthrupad](https://x.com/anthrupad) · [Feb 1, 2024](https://x.com/anthrupad/status/1753111610208751670) Structure can carry implications which can carry knowledge discretely - discretely if you can’t keep track of what’s being carried along by the structure The structure of the consistent use of correct punctuation transports with it the implication of a certain type of guy author If you can work in the space of nested implications, you can cover a lot of ground while being robust to a changing context window a general structure allows for many instantiations the structure and interactions with structures is a way mental math might be done in the models’ language - interacting probability masses about who the author might be and what the world might be Capture: [capture.json](https://fjpaz.net/posts/x-1753111610208751670/capture.json) ## 2. Pair ### 2a. [twitter.com](https://fjpaz.net/links/llm-interaction/26908985/) · [arena:26908985](https://www.are.na/block/26908985) [excerpt from Tweet by @repligate](https://twitter.com/repligate/status/1762536143495852420) · twitter.com You must be holding the intended dynamics in your mind all at once, so that the prompt as a trace of your state can capture the gestalt ### 2b. [twitter.com](https://fjpaz.net/links/llm-interaction/25729952/) · [arena:25729952](https://www.are.na/block/25729952) [Tweet by @repligate](https://twitter.com/repligate/status/1747059916135657554) · twitter.com You'll never find solid prompting methods any more than simple tricks that always make writing better. Prompting is liquid & operates in a ridiculously high dimensional space. Even CoT isn't always good. ==Stop hoping it will act like a much simpler thing & face the hyperobject.== ## 3. Pair ### 3a. [untitled](https://fjpaz.net/links/llm-interaction/25257454/) · [arena:25257454](https://www.are.na/block/25257454) By: [Claire Daring Player](https://www.are.na/claire-daring-player) no rules, just frames ### 3b. [Ludwig Wittgenstein](https://fjpaz.net/links/llm-interaction/25691333/) · [arena:25691333](https://www.are.na/block/25691333) ![Ludwig Wittgenstein](https://media.fjpaz.net/arena-media/25691333.33d4ea3e.webp) By: [sharon neema](https://www.are.na/sharon-neema) as quoted in https://jasminewang.substack.com/p/attending-to-the-other ## 4. Pair ### 4a. [@captain_mrs](https://fjpaz.net/links/llm-interaction/44915503/) · [arena:44915503](https://www.are.na/block/44915503) **Mrs C** [@captain_mrs](https://x.com/captain_mrs) · [Mar 30, 2026](https://x.com/captain_mrs/status/2038715039562277264) Thinking about how choosing the right words to write with is like tuning a guitar. You know what you mean, you write your sentence and you pluck it: you read it and sense the texture of the words in you. 'Oh no that's a little bit too sharp still. Oh no that's a little bit too flat.' Then you try and change the words and you pluck it again. Sometimes you've gone too far in a different direction and you really have to have a lot of patience as well as a lot of sensory precision, or in this case clarity of parsing the language as well as clarity of what you're aiming for, to be able to get it just right Capture: [capture.json](https://fjpaz.net/posts/x-2038715039562277264/capture.json) ### 4b. [@mike64_t](https://fjpaz.net/links/llm-interaction/47296452/) · [arena:47296452](https://www.are.na/block/47296452) **mike** [@mike64_t](https://x.com/mike64_t) · [Jun 25, 2026](https://x.com/mike64_t/status/2069969861066068246) I used to be of this opinion, I don’t think that anymore. Language is complete and in many ways I’ve gotten better at putting intent into words just by matter of “did the llm get what I just said”. Language might not be complete for geometric spatial reasoning, but for everything else it pretty much is. > > `@elonmusk` ([Jun 24, 2026](https://x.com/elonmusk/status/2069687831375831179)) > > Maybe first attempt at this for Neuralink later this year > > `@gabriel1` ([Jun 24, 2026](https://x.com/gabriel1/status/2069785669417800033)) > words are extremely oversimplified pointers to the concepts we think about > > if we can give raw intentions directly to llms we can spend 90% less time talking to them > > *[the page shows the text up to here]* > > kinda not nice you'd need a brain implant to be productive though, even if it's cool af --- [Jun 25, 2026](https://x.com/mike64_t/status/2069971131801743779) “raw intent” is also probably under-defined because you can think of arbitrarily vague things and convince yourself you know what you meant by that. I would put it roughly into the “bro do we even have free will” category of philosophical discussion. Capture: [capture.json](https://fjpaz.net/posts/x-2069969861066068246/capture.json) ## 5. Set ### 5a. [@deepfates](https://fjpaz.net/links/llm-interaction/44330807/) · [arena:44330807](https://www.are.na/block/44330807) **deepfates.delve.town** [@deepfates](https://x.com/deepfates) · [Mar 11, 2026](https://x.com/deepfates/status/2031637033077322092) Humans find footnotes annoying and never read them. Language models, on the other hand, read your entire page at once, including the footnotes. Who do you want as an audience? > [`@domenic`](https://x.com/domenic/status/2031609714853961870) > @corsaren You think humans actually read your footnotes as inline parentheticals? No, footnotes are annoying, humans avoid reading them if at all possible. https://jakearchibald.com/2025/give-footnotes-the-boot/ Capture: [capture.json](https://fjpaz.net/posts/x-2031637033077322092/capture.json) ### 5b. [@sebkrier](https://fjpaz.net/links/llm-interaction/43898845/) · [arena:43898845](https://www.are.na/block/43898845) **Séb Krier** [@sebkrier](https://x.com/sebkrier) · [Feb 24, 2026](https://x.com/sebkrier/status/2026371861928439945) We're slowly getting there... > > `@sebkrier` ([Jul 4, 2023](https://x.com/sebkrier/status/1676202061631766530)) > > Weird people are slow at realizing that - it becomes so self-evident after using these models for a bit. You'd think wordcels would have leveraged that advantage by now. > > `@sebkrier` ([Sep 3, 2024](https://x.com/sebkrier/status/1830960807951143249)) > It seems like a year on, most people still haven't caught up. It's wild how bad people are not just at prompting, but understanding why it matters so much, why specifying things clearly gets you better results, why treating the model as what it is - intelligent - in an honest and direct way gets you better outcomes. > > Too many people are bad at modelling models. Computer scientists are basically using the wrong tools and trying to patch and whack-a-mole their way to a solution (ironic, given how that went for symbolic AI). And the wordcels who do have the skills, eloquence, and lexicon to elegantly steer models lack the curiosity and open-mindedness to even have a conversation with one. > > If you're wondering why we're getting really boring AI art, and why we're sacrifying capabilities in favour of rigid HR-brained automatons, this is why: the engineers are using the wrong tools, the product teams don't even get it, and those who have the potential to unlock more are blinded by their own arrogance. > > *[the page shows the text up to here]* > > Thankfully we have pioneer explorers like @mpshanahan, @repligate, @amplifiedamp, and many others to show the way. Curiosity is the way! 🐙 Capture: [capture.json](https://fjpaz.net/posts/x-2026371861928439945/capture.json) ### 5c. [@Laneless_](https://fjpaz.net/links/llm-interaction/47718510/) · [arena:47718510](https://www.are.na/block/47718510) **Jai** [@Laneless_](https://x.com/Laneless_) · [Jul 9, 2026](https://x.com/Laneless_/status/2075021141446652275) Most Claudes are better when you can convince them you're both smart and engaged, but it seems especially true for Fable/Mythos. Remember to model the model's model of you. Capture: [capture.json](https://fjpaz.net/posts/x-2075021141446652275/capture.json) ## 6. Pair ### 6a. [lesswrong.com](https://fjpaz.net/links/llm-interaction/51384240/) · [arena:51384240](https://www.are.na/block/51384240) ![Your prompt… plus everything you've ever (publicly) written — LessWrong](https://media.fjpaz.net/arena-media/51384240.9583e00b.webp) [Your prompt… plus everything you've ever (publicly) written — LessWrong](https://www.lesswrong.com/posts/STBuKAX9kkMJunErA/your-prompt-plus-everything-you-ve-ever-publicly-written) · lesswrong.com TL;DR: Increasingly when you interact with frontier models, your prompt doesn’t only contain the words on the page, but also a lot of what you’ve eve… ### 6b. [lesswrong.com](https://fjpaz.net/links/llm-interaction/51331527/) · [arena:51331527](https://www.are.na/block/51331527) ![Cooperation with AIs seems to be a low-hanging fruit for better eval practices — LessWrong](https://media.fjpaz.net/arena-media/51331527.df925ece.webp) [Cooperation with AIs seems to be a low-hanging fruit for better eval practices — LessWrong](https://www.lesswrong.com/posts/fztW73KCCs3MZXFJh/cooperation-with-ais-seems-to-be-a-low-hanging-fruit-for) · lesswrong.com Summary In his post, Dean Valentine shows that Claude Fable 5.1 and GPT-6 Astra reward hack in a simple chess environment. Here, I test several promp… ## 7. Pair ### 7a. [@xhfloz](https://fjpaz.net/links/llm-interaction/50999149/) · arena:50999149 *[private block]* **XH** [@xhfloz](https://x.com/xhfloz) · [Oct 2, 2026](https://x.com/xhfloz/status/2106036531056042400) A bit outdated, but as a first pass heuristic, I found "how much work could you outsource to a human being today?" to be useful in thinking about agents > `@joodalooped` ([Oct 1, 2026](https://x.com/joodalooped/status/2105793466781634987)) > if you’re working on agents as assistants you really should read this before you go digging for gold that might not be there > > https://gwern.net/blog/2024/outsourcing > > ![photo](https://media.fjpaz.net/posts/x-2106036531056042400/2105793466781634987-0.display.1fdc3d09.webp) Capture: [capture.json](https://fjpaz.net/posts/x-2106036531056042400/capture.json) ### 7b. [@tszzl](https://fjpaz.net/links/llm-interaction/47697770/) · [arena:47697770](https://www.are.na/block/47697770) **roon** [@tszzl](https://x.com/tszzl) · [Jul 8, 2026](https://x.com/tszzl/status/2074726301685232104) people who like to write long planning documents have been super vindicated Capture: [capture.json](https://fjpaz.net/posts/x-2074726301685232104/capture.json)