is sims even feasible?

[bold monolith]

Kiara mentioned during the collab that sims would be a cool game for Neuro to play, and a lot of people seem to really agree.

But asides from the part where she actually PLAYS the game, like deciding what the characters do etc. would Neuro even be able to do stuff like house building or money management?

And how would machine learning even come into play with a game like the sims?

Doing The Sims some day does sound cool on paper. But does anyone know if it would actually be doable in a way that’s interesting?

Just curious. In case anyone had any thoughts

[craggy flame]

Its possible, but its a lot more work than a game like Amogus.

I suppose the easiest way would be to give her a script of the basic controls and a premade family to play with and feed her absurd amounts of training data, but given the complexity of the game i dont think Vedals 2,5gb of remaining disk space would be enough for it....

Otherwise you could train her one screen at a time, like she looks at the main menu, thinks, "Oh, this is the main menu. I select new family." Next screen, thinking, "Oh, this is the screen to make a Sim. I shall make Vedal." selects random options from the list making an absolute abomination

[spring frost]

I think for the Sims it should be treated more like using an LLM in an AGI context. Neuro does have some "common sense" just in her natural language processing, so if you could get a list of things that are happening in the game like "jimmy needs to go to the bathroom" and then insert a query like "what should jimmy do next" Neuro can literally just answer that question with her usual chat system. Just have vedal or someone control the game and do as Neuro instructs. Since the Sims is basically just a real life simulator that might kind of work! It would at least be entertaining.

And aside from hacking into the game to get a text representation of what's going on to feed to Neuro, there wouldn't be a need to do any other programming.

[craggy flame]

True a quick and easy way to make it happen is to have Neuro tell Vedal what to do in the game, but thats not really Neuro playing the game thats more the daughter sitting on the parents lap while they're gaming

[spring frost]

For a game like the sims I think that might be most entertaining. Most other games it'd be best if Neuro could play on her own.

[craggy flame]

Maybe if Vedal could program a "Game mode" into her where she focusses on what they're doing rather than random chatter.

[teal basin]

Neuro needs to be able to create a cognitive map for complex games, as in being able to generalise knowledge...

[rich kraken]

OpenAI trained neural networks to play Dota 2 from scratch using only machine learning. Also, it learned how to play by playing with only itself.
It then defeated the top players, first 1v1, then team vs. team.
After this, they polished the data by playing against real players online; it helped with fun things like item dropping (so it wasn't an "enemy item" anymore) and other human tricks.

SIMS are not the hard game; opposite of this, actually. For the sake of the game, it also needs only the mouse, two buttons, and WASD+ time control (pause, 1x, 2x, 3x) for optimized gameplay.

Theoretically, any problems with the UI and game parameters tracking are nothing if you just force AI to play by pure visual perception, real player abilities, and machine learning. Like people, they do not seek out parameters in the code and how to access them; they just look at the image and control the mouse cursor (who is visually oriented too).

I don't know if it is possible and optimal, but just saying, 10+ years ago, I already used MMO cheats (bots) that were breaking any security just by abusing visual data gathering and player-like control.
It was purely scripted—not even an AI.
The reason for any coding isn't it to make a human interfere in the working process as little as possible? Isn't it a machine's job, in general, to make everything for us? Arent the neural networks created to avoid writing a gigantic wall of code? To create solutions in situations where you will be required to do enormous thinking and spend a lot of time and are not even sure how it needs to work at the end.

To code an almost fully scripted bot, line by line, by hand, seems generally wrong and far from machine learning and neural networks. To me, at least.

|| SIMS can be a great step to create and strengthen a harem, also an endless content creating hole. Mb she will even create you there? Like, top donaters or something. ||

[tropic echo]

just my 2 cents
theoretically I absolutely agree with the "don't write scripts" point
but this whole idea is just to complex for our man vedal to do alone, more over it is not worth in terms of "time put in"/"potential viewership" so yeah, unless there are some incredible breakthroughs in visual agent networks I would not be waiting

[rich kraken]

Well, yes, seems unlikely. But I can always hope, at least
Mb there is some professional somewhere here.

[craggy flame]

You mean our main man Vedal and his underpaid assistant Alex :P

[gleaming kettle]

What I'm more interested in is seeing Neuro create Sims actually instead of playing it. It would be interesting to see the characters she comes up with

[craggy flame]

I wonder if she could attempt to recreate people off of reference images?

[left ginkgo]

Neuro should be able to come up with goals, and act in a way to carry out those goals. Its kinda like minecraft in the open world aspect.