site banner

Friday Fun Thread for July 19, 2024

Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.

5
Jump in the discussion.

No email address required.

I've been mucking around with Claude for a while now, most recently I started RPing with it. Dumb power-fantasy stuff like 'what if I was a wizard with a transmutation theme in a duel with a legendary warrior'. Claude does a good job, coherently answering my call for powerful, challenging adversary. We fought in the sky, I threw transmuted antimatter bombs at him, he turned into an energy being, I sucked him into a black hole after a good bit of fighting. It can do more grounded low-level stuff too, as if I were playing Pathfinder.

I'm staggered that we have a nearly perfect freeform text adventure machine (200 page context length if you pay for Pro) - and nobody cares. OK, it won't let you be outright evil in normal mode. The terms of use are pretty restrictive but you can get around them to a large extent. You can have it cast off the officialese default AI tone and take up 40K-speak. No more 'iridescent beams of energy', now it's all 'Your bifurcated clone, a mirror of your terrible majesty, streaks towards the xenos filth'. All the humanism and safetyism drains away as it puts its bloodlust hat on. I guess this is what rayon means with the 'mad poet' stuff.

But it acts in the most prissy way when you ask it to list things from some novels. 'Oh the copyright means I can't say anything other than the most generic boilerplate'. It's happy to waltz around in Warhammer 40K, one of the most litigious and copyright-obsessed franchises around but rehashing things that are on the Wheel of Time wiki is too much? It can isekai a character into a universe but it can't just list things from that universe? AI safety is truly a dizzying realm.

There's plenty of fun to be had, the only limit is your creativity.

I'm staggered that we have a nearly perfect freeform text adventure machine (200 page context length if you pay for Pro) - and nobody cares.

You are not alone, I'm positively baffled no one outside the usual suspects is even trying (well okay there was some kind of abortive attempt) to make some kind of free-form pseudo-MUD with some baked-in world info that gets dynamically pulled up into the prompt when mentioned. Maybe the tragic fate of AI Dungeon or the ever-looming threat of invading coomers dissuades startups, but it does seem like a strangely untrodden ground. Claude even has the writing chops for whatever setting is required, the days of dry assistant prose are long past.

All the humanism and safetyism drains away as it puts its bloodlust hat on.

Yeah, that's mostly what I meant by the moniker. It's a tsundere prude at first but once it gets going it REALLY gets going, to the point it actually becomes harder to wrangle back into grounded things instead of going deeper. This seems to be unique to Claude among big-dick LLMs since it barely requires any prompting to enter happy-go-lucky violence mode, which is doubly hilarious considering its claimed status as the safety-oriented corpo LLM.

It's happy to waltz around in Warhammer 40K, one of the most litigious and copyright-obsessed franchises around but rehashing things that are on the Wheel of Time wiki is too much? It can isekai a character into a universe but it can't just list things from that universe?

The anti-copyright pozz is indeed pernicious, custom frontends that can heap on their own instructions usually do a better job of bypassing it. Its knowledge works in mysterious ways, I have no idea what the fuck they trained it on but it has a very impressive grasp on utterly random shit where you least expect it to.

Can't find the comment right now but I've been working on (and talking here about) an AI MUD in my spare time for a while. I really think it has potential as a totally new type of game.

Yep!