Personally and as a gamer since the 80s (and nowadays making games myself), I think the last great breakthrough improvement was procedurally generated game spaces, and that stuff dates back to Minecraft in 2009.
The improvements in visuals are well into the diminishing returns part of the curve, the richness and size of custom crafted game spaces has hit a cost ceiling (hence budgets in the $100 million mark for AAA games, even with partial authomatisation of things like model generation and painting via stuff like Houdini and Substance), and the only direction of growth I can see is the gameplay itself, where the improvements from the naturally emergent gameplay of multi-player were a one-off and have been more or less stuck at the same point for a decade.
For a while I had some hope that AI (specifically LLMs) would yield a massive jump in the richness of the game world in story terms (imagine an RPG were all NPCs have genuine complex stories with realistic interactions, all generated on the fly and even influenced by you) but plain LLMs have large hardware requirements merelly to interact with one person (powerful GPUs, at least 12GB of VRAM) on top of the requirements to run the game itself, so that kind of game improvement seems unlikely before the end of the decade.
There’s been a lot of resources invested in improving local LLMs very recently, and Meta of all companies has been investing a ton of researcher time into local LLMs since the start of the AI boom with Llama and the like
Given the timelines for game development I’m still hoping for more emergent storytelling and gameplay at some point via AI, even if that’s just to generate interiors for buildings that can’t be entered and dialogue for silent NPCs
Personally and as a gamer since the 80s (and nowadays making games myself), I think the last great breakthrough improvement was procedurally generated game spaces, and that stuff dates back to Minecraft in 2009.
The improvements in visuals are well into the diminishing returns part of the curve, the richness and size of custom crafted game spaces has hit a cost ceiling (hence budgets in the $100 million mark for AAA games, even with partial authomatisation of things like model generation and painting via stuff like Houdini and Substance), and the only direction of growth I can see is the gameplay itself, where the improvements from the naturally emergent gameplay of multi-player were a one-off and have been more or less stuck at the same point for a decade.
For a while I had some hope that AI (specifically LLMs) would yield a massive jump in the richness of the game world in story terms (imagine an RPG were all NPCs have genuine complex stories with realistic interactions, all generated on the fly and even influenced by you) but plain LLMs have large hardware requirements merelly to interact with one person (powerful GPUs, at least 12GB of VRAM) on top of the requirements to run the game itself, so that kind of game improvement seems unlikely before the end of the decade.
There’s been a lot of resources invested in improving local LLMs very recently, and Meta of all companies has been investing a ton of researcher time into local LLMs since the start of the AI boom with Llama and the like
Given the timelines for game development I’m still hoping for more emergent storytelling and gameplay at some point via AI, even if that’s just to generate interiors for buildings that can’t be entered and dialogue for silent NPCs