#ai gaming tools

[solo-rpg-rambling]

Solo + LLM, some ideas

Throwing together some thoughts on playing solo TTRPGs with help from LLMs (Large Language Models, "AI", GPT, etc.). I'll throw up a cut so people who hate the idea can click and move on. Mostly I've been using LLMs as a way to expand or contextualize oracles, tables of words or phrases that are used to inspire what comes next.

Lesson 0: If the Answer is Obvious, Don't Use an Oracle (or Mechanic)

I'm sure many people have done this at least once, tried something impossible in a game hoping for an "automatic success" (natural 20 in D&D terms). At the other end of the spectrum, there are playstyles where you roll (or ask to roll) for trivial tasks and observations.

Oracles are useful for those moments when you don't know what happens next. Your characters (and you) are bored with travel scenes, so you need something to spice things up. But your brain is just churning on a lot of nothing.

Lesson 1: The LLM is a Bad GM

LLMs (at least as of 2024) are bad at synthesis, facts, math, and long-form storytelling. They are ok at making surface-level connections, suggestions, and "what happens next" (within the limits of what something trained on wikipedia, twitter, and reddit can do.)

Lesson 2: Provide Context

GPT is a "conversational" LLM. It can (within limits) remember earlier parts of the conversation. Give it one paragraph of the story so far. Describe genre and setting.

Lesson 3: Brainstorm

"Give me five ideas for..." Asking for one answer will likely get you something that doesn't quite work. Instead, ask for five suggestions and pick the one you like best.

Lesson 4: Critique, Argue, Rewrite

IMNSHO the real challenge of writing or GMing isn't ideas it's development. Treat LLM suggestions as thumbnail sketches or scribbling on a whiteboard. Take the most interesting idea, and argue with it. Keep what works, critique or rewrite what doesn't. Arguing with a bad idea helps clarify what you want from the scene.

Unleash your internal editor on those ideas.

Lesson 5: If it's not Fun, Do Something Else

I enjoy bouncing ideas off of a group or LLM. Other people use notebooks or rubber ducks. If it's not fun, and not moving your game forward, try something else.

[homestuckreplay]

Is The Wayward Vagabond Good At Chess?

On page 721 of Homestuck, the Wayward Vagabond kills time by playing chess against himself – a ‘stiff training regimen’ (p.720) for the citizen-militia of Can Town that leads to Black checkmating White, just as it was prophesied by Nannasprite that the forces of darkness would always triumph over light. But how does WV’s chess game actually play out, and what can this tell us about them and about Sburb?

First, as @tenaciouschronicler pointed out, the board is set up incorrectly. It’s a perfect mirror of how the board should be – the bottom left and top right squares on a chessboard should always be black, but in WV’s game, they’re white. To analyze this, I transcribed the actual moves that WV makes, and flipped them all horizontally – moving a piece to a2 became h2, b3 became g3, c4 became f4, etc – at which point the game becomes correct.

Here's the game after move 6 and move 13.

And, in standard chess notation (which is surprisingly quick to learn!), the moves that got us to this point.

d4 Nf6

c4 g6

Nc3 Bg7

e4 O-O

e5 Ne8

f4 d6

Be3 c5

dxc5 Nc6

cxd6 exd6

Ne4 Bf5

Ng3 Be6

Nf3 Qc7

Qb1 dxe5

Every move WV makes is legal within standard chess, so right away, we know they’ve got a good grasp of the rules and aren’t cheating. As Black, they play the King’s Indian Defense, which allows White to control the board in the early game in hopes of a counter-attack afterwards – according to chess.com, it is a ‘risky opening’ that ‘leads to unbalanced positions where Black tries to fight for an advantage instead of equality’ and makes it ‘more difficult for White to play for a draw, which tends to lead to more decisive games’.

Black’s fifth move with the knight is uncommon, but not bad. I spoke to a chess International Master to get his thoughts on the game, and he said ‘Normally people play 4…d6 in the King’s Indian to stop the knight being pushed back like this, but what Black did is also good, and he has a comfortable position out of the opening.’ He described the King’s Indian as ‘a typical fighting opening to get White to overextend’, and White fell for the play.

Homestuck doesn’t show us the middlegame, instead returning to WV’s mayoral dreams and Tab drinking as the timer rapidly counts down – but we return to the game with White in a terrible position. Here's the game where we pick up, and with White in checkmate.

Here, White is in check with an opportunity to take the opposing queen, but instead plays the following endgame moves.

Ke2 Qxg4

Kf2 Ne4

Kf1 Qg3

Re1 Qf2

This isn’t a strategic choice, but ultimately doesn’t matter - ‘No reason at all not to take the queen, White’s move is ridiculous - but he’s completely lost in the endgame anyway, with no real drawing chances’ (Chess IM). The game ends quickly, and the IM’s overall analysis is that ‘Black plays very well, either a strong tournament player or he’s cheating :) But not enough moves to go off to decide that. White is much weaker’.

Of course, Black and White are the same player. This tells us that WV does have a strong knowledge of chess, but either only knows how to play as black, or has a strong bias towards black winning. Either is believable. As a citizen of the ominous planet where the black chess pieces live, WV would likely have spent a lot of time learning strategies for Black to play, but their only knowledge of White would be how to counter their moves. They’re also indoctrinated into loyalty to the dark kingdom royalty, and to seeing Black as the superior force in battle, fated to always triumph.

Especially notable is WV’s unwillingness to capture Black’s queen. It’s an obvious move that wouldn’t have cost them the game. On page 687, WV asserts their hatred of kings as ‘petty, bossy tyrants’ who are ‘basically awful in every way’, but says nothing about queens – I wonder if WV retains some loyalty towards the dark kingdom’s queen, leading to an internal conflict where they’re unwilling to harm even her can representation.

While the battle/chess game likely plays out differently in each iteration of Sburb, WV comes from John’s game and is recreating the battle ‘years in the future’. Can this game tell us about dark and light kingdom politics during John’s game, predicting what we’ll see next in Homestuck? If so, it seems like four white pawns – perhaps representing the four players – will be allowed to initially advance, making it appear that they’re doing well in the game. John, Rose, Dave and GG will have some successes, and might let themselves get complacent with their abilities.

But the dark kingdom will hold back their best resources, and plan for the long term. They’ll take advantage of a couple big mistakes our main characters make while fighting on the side of light, skillfully eliminate these pieces from the board, retain their own queen while destroying the light kingdom’s, and, as the title of page 721 foretells, ‘lead [their] men to victory’.

It’s not a great prophecy for John and his friends – but then again, it’s uncommon for a piece taken off the chessboard to travel to another planet and start sending commands back through time. So, maybe there’s still a chance to swing the real battle in light’s favor.

[ladyyomiart]

Bringing more old Hakuouki art! ⚔️🌸 I drew this quick portrait of Heisuke wearing his kendo uniform in 2018 and I still think it's one of my best drawings of him, haha. 💖

[berrrydameyve]

okay do do you guys want anti-AI spiderverse badges??

Miles G saying "fuck ai, pick up a pencil." for example

[maria-de-salinas]

A lot of arguments against ai art deal with technical/legal/ethical considerations and those are solid and good but for me there are things that are less tangible and harder to articulate but basically just...I don’t want to be a prompt generator. I don’t want to be a passive consumer of cool pictures. I want to connect and communicate with other people. Human creativity is good and deserves to be nurtured and protected.

[doedipus]

the fact that people keep vagueposting about palworld discourse makes it very difficult to blacklist posts talking about palworld. I don't care whether you think it's neat or not, just at least say the name of the game when you're talking about it so people who aren't on twitter don't have to think about it. thanks in advance.

[straightforwardpolls]

if someone used AI tools while writing, from which stage would you consider the final product as "not their own work"? - at any stage at all - brainstorming/research - drafting - revising - proofreading - after completion (e.g using it to write a promotional blurb)

[mrpuzzle]

📄⚾️🧢✨

[scarletfasinera]

Got blocked before I could ever even see the message lmao. The answer is that I am a Communist, and as a Communist, I've read Wage Labor & Capital. hope this helps

[sparkspropaganda]

Idk I also find it frustrating because sometimes you meet people who design but have zero understanding how to make that achievable in reality, and that's something everybody has to face at some point (I'm still learning too obviously!) But it is like. If you're specifically looking at generative models like the ones people have been using for art (which is generally what people are referring to) you're going to get a rude awakening when shirts don't work like that in real life

[halamshiral]

i don't trust inzoi 😬

[ladyyomiart]

Here's the last old Hakuouki art I'll be posting today! (I've got more but I don't want to spam the fandom tag, lol). ⚔️🌸 Chizuru bringing some tea to the Shinsengumi captains! 🍵💚 (2017).

[l-1-z-a]

Emergent narrative in The Sims 2

Matt Brown at AIIDE 2006

Notes from Matt Brown's (EA/Maxis) invited talk at AIIDE 2006, entitled "The Power of Projection and Mass Hallucination (Practical AI in The Sims 2 and Beyond)".

The general gist of it is that The Sims 2 is a sandbox world that provides some small bits of narrative-promotion to try to encourage emergent narrative sequences to actually emerge, by matching them against pre-authored snippets that seem desirable. In general, Matt Brown has a very AI-minimalist style of AI design.

Some points and claims:

Perception is key: how to seem intelligent, not how to be intelligent ("Big A, little i")

Complicated behind-the-scenes computation is usually interpreted as random behavior by the player, so don't bother

Players have short memories and quickly move on, so your AI should too

Correlary to the above: Local intelligence is good enough, at least for something like The Sims. Just focus on making sure everything makes sense in relation to what comes immediately before and immediately after, and players will fill in appropriate long-term stories themselves.

Consistency leads to player storytelling and ascription of personality

Some randomness helps avoid brittleness; keeps you from getting stuck in repeated or easily exploitable weird behaviors

The Sims/NPCs react to player-driven momentum, rather than initiating much of anything

Much of the local behavior is organized around story trees

Story trees are explicitly authored bits of story, very short

In their experience, the important thing here was the authoring tools, not the AI: given good tools, a few human authors can quickly create and maintain lots and lots of story trees. Important features were sorting of story trees by roles, easy searching/comparison/etc., batch creation/edits, and variable bindings.

He was pretty adamant that it's much easier to write authoring tools to make writing extensional definitions of "good story" feasible than it is to build a generative system with an internal notion of good story, since you'd have to do a lot of manual fiddling on the latter anyway.

Events are matched against all tree prefixes simultaneously, and of high-matching trees, the events that that tree says would come next in its story snipped is scored against a model of player interest ("player likes [x] events") and personality of the NPC, etc.

Players find it easier to give specific outcomes rather than traits — answering "likes hiking (y/n)?" is easy, while "x/10 for extrovert?" isn't.

Focus the player on the details you will actually use; in character design and personalities, don't try to faithfully model things that don't matter very much to your game

Anecdote: They had a complicated model for which urinals Sims would use: if there's 3, and someone's at the leftmost one, the Sim is supposed to use the rightmost one, not the middle one, unless they have some sort of weird personality. It wasn't reliably getting the responses they wanted, so they ripped it out and replaced it with a random-number generator, which people were just as happy to make up stories about and ascribe personality to.

A bit of an admission in the conclusion: As a sort of aimless "sandbox game", The Sims only really needs to make sure something interesting happens, but it hardly matters what. The player is responsible for making up stories (this is their sandbox after all), so the story-AI part of The Sims is only intended to provide some prompts and play off what the player does. That might not be the case in other types of games.

Follow-up: Maxis ended up deviating significantly from this view of Sims narrative with The Sims 3, bringing in Richard Evans to do a less AI-lite version of the AI, which included longer-term planning rather than purely emergent narrative. Evans has given a number of talks on that, which I unfortunately don't have good notes on, but here (archived) is a brief writeup someone else did of his AIIDE 2007 talk.

Mark J. Nelson, 2006-06-21.

<Note index>

Comments welcome: mjn@anadrome.org

Many thanks to @andrevasims for searching in this post:

[kyonite]

man does it ever strike you how fucking useless 90% of the AI shit companies are trying to peddle is? like especially for creatives it's just so functionally fucking useless.

"oh but I can use it to write the manual for using this tool" is genuinely something my boss said to me earlier today and I was like. do you know how AI works? do you know that in order to effectively write the manual for that thing, it needs to reference another manual that was already written for the thing? another manual for the thing which was,,,, wait for it,,, written by a human.

it's just so many of the use cases that people try to point out for creatives are just like. functionally fucking useless for many reasons but esp because it's referencing something SOMEONE ALREADY DID. why do I want a robot to regurgitate some shit that a serbian man already wrote on a website in 2009. he already did a fine job. why is the robot here.

[my-thoughts-and-junk]

my new thing is obsessing over a funny little guy for a few months before moving onto another funny little guy

[gutsfics]

tbh i think that the "is ai generated content art" argument is roughly the same as "is Duchamp's Fountain art". i think that the argument of if its art is what makes it art, because it is invoking some kind of feeling. starting a conversation. if that makes sense?

that being said, absolutely there is a time and a place for it (ie not replacing the work of paid artists). and it should only be trained on artsts who have consented to it.

i dont want to read fiction made by ai bc if you didn't take the time to write something, why should i take the time to read it? and for non fiction, it'll be riddled with incorrect facts because an algorithm "decided" that this word is the most likely to go next in the sentence based on what it's database says, even if its factually incorrect.

one of my favorite pieces of art is an ai image that was sent to a company that makes paint-by-numbers, and then filled out by a person. the artist, tumblr user rigatonidanza, asks "is this art? at what point does it become art? can it never be art because its ai, or because its paint by numbers? is it art because it elicits a response from the viewer?"

the point of art is to make you think, and to make you feel. does it stop being art if its not made by human hands, the way algorithmically generated images are? is it not art if its something mass produced, made for intentions other than what the artist uses it for, like Fountain? even if looking at it makes you feel something? anything? even if that "something" is a negative emotion?

all that said. if you use generative algorithms as an excuse to not pay real artists, you are a human tar pit. may you always immedeatly stub your toe after it stops hurting from the last time you stubbed it.