It looks like an interesting idea worth exploring in the modding scene, but practical uses may be like 10-20 years away. I could imagine the new ways to make video game dialogue would be based of not giving characters ANY pre-created answers to any question, but just what they're able to know about the world and quests, and then to be allowed to give an examples of what humans would be able to come up with such knowledge.
Currently it looks like the VR games, most gaming population simply doesn't have the hardware to run these sorts of things, as many games have stayed in more traditional way.
I think we have to wait at least 1 or 2 more console generations to add these computer chips which would be designed around generating AI given content.
I mean when first 3D games came out, like Doom, Duke Nukem 3D and Quake, we had no idea how long it would take to perfect that kind of 3D world rendering... in fact even to this day we're still not fluidly able to maintain 3D worlds without all sorts of quirks.
To see real working and practical AI generated dialogues (in pre-generated quests) we'd likely have to go to year 2040-2050, and then to have AI be able to actually generate quests that are worthy of exploring add another 10-20 years.
At that point in time, instead of worrying about the content of the game or graphics, the biggest competition will likely be in how interesting the rules are in a said game, the actual core of what makes good gameplay and interesting mechanics. You take those out and I think we're going to see a bunch of clones of similar dialogues, all voice AI's sounding about the same with the 1 000 different AI sound models, with most conversations being just like the demo shows, just a sliding bars to determine which character has more humor or less toxicity or is more helpful.
Certainly worth exploring this and maybe even adding it to something less forced role, like having able to have chats with randomly generated NPC's and allowing them to tell their life stories, I think that's what we'll see them first used.
So I wouldn't rush to expect this to emerge as the new standard in future games just yet, but we'll likely see some implementations of it, while still maintaining story focused narratives.
For this to be any practical, I think games would have to be much-much larger to be able to develop entire zones with people with certain attitude, events that people be able to refer to. It would take an entirely different approach to quests and how they're given. So that's why I don't see this technology taking off any time soon.
Look good, just like VR, and how many times that has been tried, along with 3D movies, and they always seem to fail eventually, before resurfacing 5 to 10 years later.
I'd say just wait till year 2040, expect to pay like 200€ for this sort of game that would have to require at least 50 million copies sold... and this would just be the entry cost, with likely monthly subscription for even a single player game that would require a constant stream of 100 employees working their butts off, and building up new narratives, fixing old ones, generating all sorts of reasons to take on these adventures etc. Those kind of games would be so large that there would be stores within stores. You'd be able to make your own life in a video game, building up content for others to consume etc. It would take similar scope that went from building Doom with 5 people, into creating Doom with 200 people, developing real world AI generated world with infinite activity would take up to 8 000 people.
Remember that this is all just a prototype in a LAB environment, with the whole video being edited likely multiple times to take out hiccups, and to actually publish this could've taken 200 re-shoots of the AI code, to finally get a simple NPC to work as noodle salesman.
This current version of conversation AI will likely end up like Nintendo's VR console, like a cool gimmick demo, which causes nausea and comes with only a red color display and you have to strap your face to this console.
Maybe in next Cyberpunk, we could see this in use in random NPC dialogues, maybe even added to Cyberpunk 2077 NPC's later, but nothing more. And these wouldn't be more interesting than "Have you heard the news? What happened in Crystal Palace... that's horrible." where other NPC responds to "Crystal Palace? Some rich a****** got killed?", first guy saying "Yeah, they said it was some hitmen who worked for MiliTech.", 2nd guy saying "Oh Militech? That's where my father used to work. They're such great guys.", conversation gets back and forward and both characters look like Skyrim NPC's and there's no big expressions on what they say, there's probably not going to be some kind of conflict with NPC's etc. It's just generally not going to be interesting plotwise. It would be only interesting because it's AI giving the dialogue, not because it's going to offer any insight that already wouldn't be able to be given by pre-recorded scene.
And this one is big reason for why we wouldn't see this taking place for long time:
Why use generative AI that offers more work, when you could create a special interaction with just 1/10th or 1/100th of the work? People who play video games aren't going to replay the same 100h games more than 1 or 2 time, most people don't even complete their games, so doing this generative AI thing would be just a gigantic money sink, especially open world games. Maybe some indie games with far less budget would be able to create this clutter that pretends to be content to offer to people to interact with, but trying to scale this to anything meaningful, especially with modern hardware is just fools errand.
TLDR: Like I said, it's likely to arrive, but it's going to be far less impressive than what's advertised, and it'll take generations to mature. The entry cost is going to be very high, and gaming already is very exclusive hobby which many aren't able to afford today.