gpk blog - Just Send Me the Promptgpk blog
My ramblings about emacs, the internet and everything
Powered by Hugo | Themed with Poison<br>© George Kettleborough<br>CC BY-SA 4.0
If you are thinking about using an LLM to communicate with me, you will be glad to hear<br>I’ve just developed an exciting new technique which will save us both time. Rarely are<br>such wins possible, so this is surely cause for a celebration.<br>The technique goes as follows:<br>Type up your LLM prompt,<br>(Here’s where it gets really exciting), instead of sending your prompt to an LLM,<br>just send the prompt directly to me!<br>“What?” I hear you crying, “My prompt doesn’t contain all the information you need!”<br>But it does. You can’t create information from thin air and I too have access to LLMs.<br>If I want to bloat your information a hundred fold by incorporating generic drivel I can<br>do it myself. I could even do it using the LLM of my choice and tweak the prompt to make<br>it more or less verbose.<br>However, we are barely scratching the surface of what’s possible here. Because, you see,<br>I won’t even feed your prompt to an LLM at all!<br>OK, I know this sounds insane but hear me out: I have the very same language processing<br>capabilities built into my brain. An organic language model, you might say. It’s even<br>quite large, if I may be forgiven for boasting. I can understand a fair bit but,<br>admittedly, I can’t understand everything. Luckily I also have circuits I’ve developed<br>that tell me whether I really understand something or not.<br>So what if I don’t understand? What if your prompt is too terse? Too devoid of context?<br>I could always fall back to feeding it to the LLM after all, but ultimately the LLM<br>won’t have any more information than what’s the in prompt, it will just be guessing the<br>details. So what then?<br>This all depends on the medium of communication.<br>If it’s an email you’ve sent me, I will simply reply to you (sending you my prompt<br>directly, of course) asking for clarification or more details. You can then amend your<br>prompt as you would normally and send it back.<br>If it’s a published work such as a book or marketing literature then I’m more limited in<br>my recourse and I will likely just leave a bad review or not buy your product etc.<br>Either way, together we will learn to communicate clearly and effectively<br>together. We’ll learn to be clear and concise in our prompts and be fully aware of<br>implicit context at all times. I believe it’s called prompt engineering, but I prefer<br>the older terminology: writing.
« Previous<br>Automating Metabase Upgrades on Kubernetes with...Next »<br>Think Twice about Boolean Fields in your Data...