Embedded Files
Part 3: Prompting tips
Another way to write prompts in websim is to compose them in the style of a URL. This isn't my preferred approach, but plenty of other people like to work this way (see websim's beginners' guide and Cluck0matic's URL prompting bible).
Claude doesn't usually have the ability to access real webpages on the Internet through websim, but what it can do remarkably well is predict what kind of content it might find on any given page. You can think of it as the "suggested autocomplete" for web pages that don’t yet exist. By entering a made-up URL into websim’s address bar, such as "http://anthropic.wikimania.com/wiki/soap_bubbles," Claude can start to guess what might be on a page like it, even though you are essentially pointing it at an imaginary site.
It looks like Brian Sunter used this approach to make a lava lamp. If you look at his prompt history by clicking on the websim address bar, you can see that he entered "https://websim.ai/lsd-trip-effect" and then "https://websim.ai/lava-lamp-animated" several times in a row.
Note that you don't need to start your URL with "http://" and you can invent your own kind of protocol. For example, Katan'Hya experimented with "gopher://" and Janus did the same with "loom://." To make generative art, you could try a URL that starts with something like "processing://," which Sirkitree has done to get some cool results, like this one and this one. I've thought about doing something similar to try and nudge Claude into a p5.js "attractor basin" (basically, making it output a certain set of results more frequently than it normally would by default) since p5.js is a JavaScript library based on the Processing language, and it's mostly used for generative art.
Next page: "Scattershot" prompting
Page updated
Google Sites
Report abuse