Support NeoGAF

[Aces High]

Actual problem with AI writing is that is completely safe and boring, is like playing Dragon Age: Veilguard.

It's whatever you tell it to be, minus the limits implemented by OpenAI.

If you want a story about nonbinary characters, it can do that.

If you want Stellar Blade, it can do that, too.

Example:

I have a female bard teammate. I created a virtual interface for her sex appeal and flirt factor and set it to 9.

This was pretty bad because she flirted like a male, very agressive verbally and always with focus on me.

That's not how females do it.

I told GPt to change that to an authentic female sexuality:

Women do sexual confidence checks with men.

In public, they use their body in a way that drives men crazy, but also makes them question themselves whether this was on purpose or just by accident. Is she into you? C can't tell. That's fun and tense.

Women are biological beings so they are horny just like men. But they don't want to be seen as sluts. Which means 1) they never go all-in verbally in public and 2) when in private, they do it in a way that communicates their desires in a very very subliminal way. And unlike men, they direct the focus to themselves instead to the other person when being flirty.

I explained it to GPT, he agreed, and the bard's character is super fun now. It's actually one of the most fun female characters I've seen so far in games.

But there’s nothing original about LLMs.

The very essence of a large language model is mass digestion of copied content. It’s just regurgitating it back to you.

That's the case for manmade content too, tho.

I mean just look at Tolkien's Lord of the Rings which is heavily based on Norse mythology.

Avada Kedavra is an actual curse. The Vatican said publicly that J.K. Rowling has some serious knowledge of this magic shit.

Everything is a remix.

Some human creators are just good at hiding their influences.

 

[GeorgioCostanzaX]

Side note I’ve been trying forever to find a DM in the US and set up a GAF Discord for noobs who need someone to walk us through D&D. Welcome to join if one of you doesn’t mind teaching us the ropes? I figure once we know wtf we’re doing we can maybe just use CHATGPT to DM after that.

Join the GAF plays Dungeons and Dragons Discord Server!

Check out the GAF plays Dungeons and Dragons community on Discord - hang out with 8 other members and enjoy free voice and text chat.

discord.gg

 

[StereoVsn]

I did want to add that the current general LLMs are terrible at math. And that goes double for geometry or trig proofs, not to mention higher level mathematics.

I think the new “slow ChatGPT” model is better but for the most part others kind of suck.

 

[ScHlAuChi]

The future of gaming!

ChatGPT playing Chess - "Never play chess against a pigeon"

Summary He summons pieces that weren't on the board Teleport pieces all over the place Captures its own pieces Does Wololo turning white pieces into black and still loses LMAO

www.neogaf.com

 

[Werewolf Jones]

Me and Elon with ChatGPT as the DM.

 

[JimmyRustler]

Assuming AI gets that far, won’t this basically be the pipe dream for RPGs? I mean, we you have AI reacting to your actions/decisions real time, then there is true uncertainty for consequences. I find this somewhat intriguing.

 

[duck_sauce]

I explained it to GPT, he agreed, and the bard's character is super fun now. It's actually one of the most fun female characters I've seen so far in games.

The Problem is, chatgtp always agrees with you in situations like that. That is the problem. There is no will or intention behind its words.
A imaginary world needs imagination and that needs the will and intention to create something new and not just cater to an user.

 

[Aces High]

The Problem is, chatgtp always agrees with you in situations like that. That is the problem. There is no will or intention behind its words.
A imaginary world needs imagination and that needs the will and intention to create something new and not just cater to an user.

Yes, GPT4o is designed to be very agreeable. Much more than GPT4 in my opinion.

GPTo1 Preview, however, is a lot less agreeable.

So it's most likely a matter of tuning and training. It's a product after all.

For game development you would need a dedicated gaming LLM.

 

[Aces High]

Think about this:

Agreeability is a subjective factor caused by the fact that natural language always leaves room for interpretation.

It's a potential weak point of end user LLMs like GPT4o.

A dedicated gaming LLM for professional use in game engines would probably use a synthetic programming language.

Programming languages are inherently unambiguous, with strict syntactical rules.

An LLM trained to understand and output code could precisely define game mechanics, AI behavior, or environment setups.

This could greatly enhance their precision and utility.

 

[nikos]

No thanks.

 

[yet_another_alt]

I've been treating LLMs like one of those Choose Your Own Adventure books. Except choices are not binary, it will take any input and try to spin something out of it. I've seen some very creative responses, some quite serious and some laugh out loud funny. It's really very entertaining and I can definitely see how people lose themselves talking to AI.

 

[Aces High]

So I was watching videos on this YT channel which has a super basic but also fascinating "ambient walking" concept:



And I wondered:

Why does watching this video feel more immersive than actually playing the game?

The problem is that devs perfected environmental design, but they didn't come up with gameplay systems to support it.

On the contrary:

Modern gameplay systems prevent immersion.

Instead of letting us explore freely, we’re pushed through a series of checkpoints that hand out rewards.

While these “carrots" can be motivating at first, they eventually make the game feel like a treadmill—just doing tasks to unlock the next prize.

It ends up feeling like a Skinner box, where you’re only playing to get the next reward, not because you’re enjoying the world itself.

This really damages immersion.

Even a beautifully crafted world like AC Origins, with rich Egyptian lore and detailed environments, won’t shine if players are just rushing through it to get to the next goal. Everything feels more like background scenery.

A better approach would be to give players reasons to explore that don’t rely so heavily on external rewards.

AI could achieve this in the future.

With AI, we could see a focus on emergent gameplay:

Systems that interact with each other so interesting situations happen naturally, without scripted missions. Creatures, factions, and cultures could respond to player actions and feel alive, and not just decoration.

We could finally move away from a pure “carrot-on-a-stick” model and trust players to find their own enjoyment.

 

[struggler_guts]

I've been treating LLMs like one of those Choose Your Own Adventure books. Except choices are not binary, it will take any input and try to spin something out of it. I've seen some very creative responses, some quite serious and some laugh out loud funny. It's really very entertaining and I can definitely see how people lose themselves talking to AI.

The Problem is, chatgtp always agrees with you in situations like that. That is the problem. There is no will or intention behind its words.
A imaginary world needs imagination and that needs the will and intention to create something new and not just cater to an user.

I've been playing over the last week after seeing this thread and I've basically come to the same conclusions. It's a lot of fun when I want to do a bit of 'create your own adventure' but I've found it starts forgetting details after about a day or 2 of casual playing and it tends to repeat expressions in conversation too