A thoughtful piece that has me thinking about translational problems for technology in the game industry. Like how Microsoft and Google and Facebook all seem like they "just don't get it" when it comes to games and keep pulling defeat from the jaws of victory.

There was that cloud gaming fad a few years back when it seemed for... some reason... every "big tech" company had to be doing something. [1] Google's Stadia had some good ideas on the controller front yet did nothing to really change the industry: GCP stocks a large number of huge machines with multiple GPUs which would be capable of running multiple player games where a single machine runs a shared simulation for all players and streams the graphics to them. Risky? Yes. But had such an effort succeeded it would have had a major impact on the industry, and it was something Google could have done better than anyone else.

I like the theme too that "instruction following will never teach an LLM to talk like an NPC" but I do think the innovation out of this space are going to come out of things like "character.AI" where the relationship to the character is the endpoint whereas you might add really revolutionary NPCs to a game and find users don't really engage with them.

[1] ... itself a structural problem in the industry

i remember i was working on some behavior/moderation related issues at a game company a while back, and always thought it'd be cool to have AI NPCs playing a role in incentivizing good, non-toxic behavior. For example, let's say you're playing an rpg, and based on your interactions with the NPCs and other players, the AI NPC would reward you for a quest/mission you complete with an item having a rarity that corresponds to your behavioral score.

I expect this to come, but wonder when..