Post by @jonny@neuromatch.social

jonny (nonvenomous)

@jonny@neuromatch.social 15 hr. ago

RE: https://neuromatch.social/@jonny/117219507611629659

while surveying scam mobile games I arrived at a two step test for whether something qualified as a game:

does player input meaningfully affect the outcome of the game
if yes, does improved player input result in improved game outcomes

jonny (nonvenomous)

@jonny@neuromatch.social 2 days ago

The kart racing game demo is actually a perfect demo of LLM-generated games: as soon as I opened it my phone got like 20 degrees hotter, and it requires no player input aside from holding down "go" to win. It can imitate the form of a game in the least efficient way possible while having none of what makes a game a game.

edit: here is the kart racing game: https://neuromatch.social/@jonny/117228876689358136

16 Replies

2 Boosts

Ryan Castellucci :nonbinary_flag:

@ryanc@infosec.exchange 6 hr. ago

@jonny ergo candyland is not a game

jonny (nonvenomous)

@jonny@neuromatch.social 37 min. ago

@ryanc
I wouldn't play a scam mobile game version of candyland, that's for sure, but many scam mobile games effectively function like candy land. The object of candyland is outside the universe of the game - an interactive no-screen walk through a candy-themed area with your kids, doing headcanons and acting out the candy bosses and whatnot. The meta has a high degree of un-evaluable skill and imagination like "how good of a chocolate monster voice you can do" and "having prepared candy to resolve child conflicts arising from RNG imbalance"

But ya if you are gonna play candyland you might as well just get out the colored pencils instead

Pasadena CAS

@pcas@urusai.social 6 hr. ago

@jonny Greg Costikyan proposed a definition of games back in 1994; your points would seem to fall under his 4th point "decision-making". https://pressbooks.usnh.edu/creatinggames/chapter/costikyans-definition/

Other parts of his definition exclude titles often considered games (e.g. The Sims), so I don't mean to suggest it's perfect - just interesting to consider as an example of what other folks have written on the topic

View

jonny (nonvenomous)

@jonny@neuromatch.social 47 min. ago

@pcas
Oh yes definitely, there is a rich space of theories about games and what makes game and even what makes a game fun - this test is specifically for when I was evaluating scam mobile games and I was trying to determine if a game was even attempted. I'd first start by doing trivial or random input and see how far I got.

🎀𝓀𝒶𝓃𝑒𝑒𝓁🎀

@kaneel@mas.to 13 hr. ago

@jonny something that I've realised while working on a defunct, doomed, project of a "browser for html5 games" is that a lot of the games developed by crypto and ai people are not games.

They think "programming is the barrier for making games" and don't think about gameplay and what it is they want people to "play with", TLDR: these people don't understand what makes a game a game.

Androcat

@androcat@toot.cat 13 hr. ago

jonny (nonvenomous)

@jonny@neuromatch.social 13 hr. ago

@androcat the problem is that e.g. universal paperclips is actually a game and in my opinion fun, because the outcome of the game really does depend on player input. it inspires a space of play and metaplay that even includes interesting automation questions. people almost can't help but make fun games even when they are trying to make the minimum game idea. non-game games are a different kind of thing altogether

Androcat

@androcat@toot.cat 13 hr. ago

jonny (nonvenomous)

@jonny@neuromatch.social 51 min. ago

@androcat
Totally, yeah, speedrunning, breaking a game by beating it harder than it was designed to be beaten takes a great deal of skill and is a lot of fun, and there are games that are more and less fun to try and speedrun in part because of the high skill cap that means there is still optimization that can be done to make it even faster, but there would be no joy in speedrunning a game like this because there would be no meta, surely there are exploitable bugs but that's like throwing candy on the ground and picking it up and pretending you found it - finding bugged routes in LLM code feels different than finding some bug by reverse engineering highly optimized human code and misusing it. This game simply doesn't enter into the category of "game"

jonny (nonvenomous)

@jonny@neuromatch.social 15 hr. ago

A whole cateory of mobile games actually don't condition the win on player input at all. Like most of the gate-based games like "advancing army approaching, move side to side to shoot different guys" don't require you to do anything except hold down your thumb to win more often than you lose. If it does matter, the cap on how much improved player input improves the outcome is remarkably shallow - if you are attempting the game, you have reached the skill cap.

jonny (nonvenomous)

@jonny@neuromatch.social 15 hr. ago

It appears like a game because the game is actually a scam, and the scam is the driver of play.

I don't have as clear of a scam test, but its something like:

is limiting the amount of play the primary challenge of the game?
are there geometrically scaling quantities, like a crafting or item ranking mechanism, that produces a Shepard's tone treadmill: a feeling of constantly accelerating progression
are there multiple counters that function like currencies, and can they be exchanged for one another?
if multiple, there is at least one currency whose means of gaining it is regulated by a) real world time, and b) play time
are any of the above remedied by paying real money?
is there a leaderboard/social element? If so, is the multiplayer gameplay implausible? Are the social mechanisms indistinguishable from if all the other players were fictional and there was simply no means of connecting multiple players ?

jonny (nonvenomous)

@jonny@neuromatch.social 15 hr. ago

Most scam games are scams but not games. There are some scam games that are both scams and games. There are no scam games that are games but not scams, because then they are just games. LLM games are the otherwise unoccupied member of the quartet: scam games that are neither games nor scams.

EvoScale

@EvoScale@c.im 9 hr. ago

@jonny The logistics get even more mind boggling, when you factor in Flock cam scams.

jonny (nonvenomous)

@jonny@neuromatch.social 14 hr. ago

for posterity, here is what a scam game that is neither a game nor a scam, i mean AGI, looks like. (3x speed)

nobody, not even openai, looks at the fucking model output, not even when it is their fucking "we did AGI" press release.

thanks a lot, wittgenstein, there is no way to define a game so there's no way to benchmark for being a game.

3+ more replies

@jplebreton@mastodon.social 9 hr. ago

@jonny it's interesting (and dismaying, obviously) how there's now this whole massive-and-growing class of output where if you criticize anything about it the hype guys will immediately admit it's trash; the work itself exists to suggest proximity to non-trash, eg "yeah but just imagine how much better it'd be with [a better prompt, next year's model, more compute]". a quality void to insist on future quality, to endlessly deflect and defer judgment, until the IPO / shares vest etc.

SnoopJ

@SnoopJ@hachyderm.io 9 hr. ago

@jplebreton @jonny "just imagine a good version of [shitty thing], and please do not reflect on why you are capable of imagining those good qualities and where they may ultimately originate. please invest."

Discussion

Discussion