My friend from Italy introduced me to panzerotti which is essentially a pizza folded like a calzone in thin dough, but pan-fried instead of baked and it’s awesome. It’s toasty like grilled cheese and made me go oh, this is what Hot Pockets is trying to emulate…
Exhibit 14a: Hot Pocket vs. Panzerotti
My main interactions with Camembert have been through IF, so now I will always remember how to spell it because it’s CAT + MEMBER.
The fork is because I don’t have the co-ordination for knife-spearing. Also handy for any vegetable sides (peas, beans, miniature tomatoes or saeurkraut) that get served.
Chopsticks are OK if I have a lot of time and want to make everyone else at the table laugh.
I hope this is not too on-topic for the off-topic forum, but as I mentioned in my Pangamebook thread this weekend, I used ChatGPT to generate test gamebooks (i.e. choice-based interactive-fiction stories). Think I ended up with seven generated stories, exposing three or four bugs I had not noticed before in my script. The stories were almost flawless, from a technical pov. Correct syntax (as I had instructed) and almost no broken cross-references between text-sections. Mostly coherent stories as well, although most were short (15-20 sections). One was a longer 100+ sections story that I created by ask it repeatedly to keep generating more content. That resulted in some broken links and repeated sections, but nothing that could not be solved in a few minutes by manual edits. This was ChatGPT 3.5. 4.0 is probably better.
It is not anything I would recommend to write real branching stories to be consumed by humans (although I have honestly seen worse gamebooks, written by humans). But for generating test-data like this to use when working on a tool for interactive-fiction it is not bad at all.
My first idea was to generate random test-stories using python and just spitting out boring random words, but using ChatGPT was easier and better (and more fun). Plus the occasional syntax error, or just the use of syntax I would not have thought of trying, was useful for randomly finding bugs.
I find that it’s fun to see how ChatGPT generates “creative” storylines, but its real and useful strength, for me, is in analyzing human creations for consistency and errors. In other words, I’d get more useful results using it as a shaping tool.
@pelle That’s a cool idea for testing. The sorta things I occasionally use ChatGPT for are asking it to rewrite a passage or dialog in a given sounding voice (valley girl, newfie accent, etc.) or correct for grammar errors. The pizza example is a kind of language translation where understanding needs to be used. I don’t know how good ChatGPT would be at localizing a game to a different language. I would guess short snippets would be difficult to handle, but have you ever tried this?
@cpollett I have not really. For one of my books I asked it to generate it in Swedish, just to get a test with some non-English characters. Just now I asked for a python-script to generate a bigger random book. With a few tweaks I got a script that generates a 2000 section markdown gamebook with nonsense words in. It is not fully connected, but there are randomly 0-3 links going out from each section, so it looks like it is mostly a complete graph, and there is at least one link to each section. I never saw a gamebook longer than some 1500 sections in the wild (unless we count the entire Fabled Lands series as one big book). Looks like 2000 is not a problem, but Graphviz struggles to generate a PNG graph of the book (have waited for several minutes while it is running at 100% CPU).
Wouldn’t a highly connected graph featuring that many nodes will result in so many criss-crossing paths? I’d be surprised if the resulting graph won’t feature some very dark areas.
Like some really complicated string art.
