Honestly, fair enough, it would take a lot of work to have an LLM direct your game in the intended way.
That said though, to create an AI system equivalent to an LLM would be even more work.
I think a lot of this comes down to AAA vs Indie, as well. For a AAA game, there is a lot more pressure to keep the LLM in line - which is of course very difficult if not impossible. For an indie game, though, the goofiness can be part of the charm, I think.
I guess my point is that I’m just excited to see what people can come up with. The point of games is to play, and I personally think LLMs are fun to play with.
I think the idea that it takes more work to do things by just engineering it in is flawed, to say the least.
Building llms is incredibly difficult. Building good training data is even more difficult. Then ironing out all the problems and biases in your model is even more difficult than that. You often need to build new models just to correct the old models.
There are places ai does well, it is however not a magic wand that just makes everything easier. It is far far far from that.
kakes@sh.itjust.works 11 months ago
And yet, it would be far more dynamic - not to mention easier to implement - if it were powered by an LLM.
echo64@lemmy.world 11 months ago
no, it wouldn’t. you would spend 10x the time to make a generative model that fucks up constantly from biases you never imagined.
kakes@sh.itjust.works 11 months ago
Honestly, fair enough, it would take a lot of work to have an LLM direct your game in the intended way.
That said though, to create an AI system equivalent to an LLM would be even more work.
I think a lot of this comes down to AAA vs Indie, as well. For a AAA game, there is a lot more pressure to keep the LLM in line - which is of course very difficult if not impossible. For an indie game, though, the goofiness can be part of the charm, I think.
I guess my point is that I’m just excited to see what people can come up with. The point of games is to play, and I personally think LLMs are fun to play with.
echo64@lemmy.world 11 months ago
I think the idea that it takes more work to do things by just engineering it in is flawed, to say the least.
Building llms is incredibly difficult. Building good training data is even more difficult. Then ironing out all the problems and biases in your model is even more difficult than that. You often need to build new models just to correct the old models.
There are places ai does well, it is however not a magic wand that just makes everything easier. It is far far far from that.