[666maslo666]

I decided to play with the newly released Claude 3.5 Sonnet AI model by feeding it a pretty long document summarizing the Freespace Universe and the FS1 campaign (attached), then prompting it to create a new campaign set in the same timeframe as FS1, but in a different fleet. Then I fed the result back a few times, each time prompting the AI to improve and refine it, make it more interesting and varied etc.. The (attached) results are... interesting indeed.  It is still pretty AI-sloppy and derivative, but I am impressed that it is at least consistent, there are no obvious mistakes or things that would not make sense. With some human touches, it could even be a basis for a (pretty mediocre, but passable) campaign. LLMs have really come a long way. So, welcome to the GTD Horizon and the 6th Fleet! 


(The logo is by ChatGPT 4o. It was not very usable for creating the campaign itself due to short usable context window though, Claude models are much better in this regard.)

[Shivan Hunter]

As with everything AI - this is definitely interesting in a "You may live to see man-made horrors beyond your comprehension" kind of way

I've definitely seen the objection before that AI doesn't have true comprehension, that it can "only" take ideas already fed to it and mash them together in a new form, plus a bit of randomness that can look like creativity if you squint real hard. And that of course raises the question of whether this is a fundamentally different thing than humans do - a difference in kind - or simply a difference in scope (i.e. we are ChatGPT but operating on a deeper level, with deeper ideas and more knowledge to draw from).

This AI is definitely doing that here - taking a bunch of FS info and schlocky sci-fi tropes, and kind of making something semi-coherent from them by accident. But I notice that it's the same kind of slop as a lot of very early, amateur FS campaigns (as well as a certain Mortal Engines crossover which I won't name). In fact, "Eiooes of Desitny" actually feels more coherent than some of those campaigns.

So I take a step back and acknowledge that AI has already surpassed some humans, if not all humans. Amateur writers, if not good writers. Algorithms could beat average humans at chess long before they could beat Kasparov.

As a wee lad, I used to write campaign outlines (they were terrible) and I had the same kind of derivative, uncreative "idea-magpie" tendency to take stuff I thought was cool and jam it all into the campaign. So I know what that process feels like from the inside.

I've got another campaign idea in the works, and it feels like my approach is different: I'm fleshing out a particular worldbuilding idea based on how I think GTVA politics might progress based on the events of FS1/FS2, and planning my campaign around that. But am I really doing something fundamentally different now, which AI couldn't replicate just by scaling up? Or am I still just an idea-magpie, borrowing on a deeper level from sociology/politics/game design rather than "I saw this and thought it was cool"?

[Mobius]

You may wish to just add the AI generated text straight into your posts, as attachments are generally short lived.