Samstrike 68
Brummbär & Störmi
Ich hab keinen Plan von Cloud Computing und dem ganzen drum herum.
Was ich aber Z.b. von Turn 10 oder Respawn verstanden habe ist, das z.B. die Ki / Bots in der Cloud berechnet werden kann und die one diese freigewordene Rechenleistung für andere dinge nutzen kann.
Passt das oder ist das auch nur geblubber ?
Selbst wenn sich keine Mehrleistung z.B. bei Forza ergibt, so sind mir die gespeicherten Spielerdaten lieber als eine "normale" Ki. Wenn es denn so stimmt wie erzählt wird.
Hier noch was dazu aus dem Forza Thread.
You said that Drivatar offers “cloud-powered opponents.” Can you tell us a bit more about that?
Drivatar is like nothing else. I've actually been struggling all week to describe it. Everyone's looking for, “is like ghosts or is it like replays?” No, that's what Rivals is. We have Rivals Mode, and Rivals Mode is great, it's gonna be in Forza Motorsport 5 and that is racing against other people's ghosts and you earn money for beating them, and that's really cool. We have multiplayer, dedicated servers, big matches; again, great racing. But Drivatar is not like AI, it's more like big data, it's more like search. In AI you've got a number of parameters that you can fill in either manually or by using learning, so we had learning in the past, or 'Drivatar 1.0', which was in the original Forza. It was learning AI, but it had a set, finite number of parameters that would fill up through learning. Or you could have a designer train it by filling in numbers on a spreadsheet like a traditional racing game.
This is different, this is big data, so we log a lot of data: where were the other cars when you made the decisions you make? Did you set up a complex pass? [Did you] bank right, dive left? How did you actually set that up? We log all of that and the system doesn't actually need to know anything about it. For example, there's a car in front of me, I'm following up behind him more or less straight away, and it hits its brakes. Well, in race craft we call that 'check braking'. The system doesn't need to know that, the system just needs to see that say of a million people, 300,000 did it and now it learns, 'hey, there's this thing that people do and here's where they tend to do it. These tracks, these cars, and all the other cars are in these positions or not, and they do it against their friends or not against their friends, and it generalises and learns these parameters.
Now every Drivatar can't necessarily do that, so you have to then train your Drivatar to do it. So, it's an ever-evolving system, it's learning race craft. Now it already knows a lot of race craft right now because we've got a lot of people on our team and we're not bad racers, we're racing all the time, and we're also pulling in people all the time to further train it. You just have to imagine millions of people racing for hours every night, and how the system just gets bigger and bigger and bigger and starts learning more and more things, so basically the game at launch is going to be a different game a month later.
Where the rubber meets the road though is the difference in racing. Back at the studio I saw three cars go cleanly three abreast through a [corner]. AI doesn't do that. Show me a game where you've seen cars go three abreast through a corner. It's super rare.
Wo wir schon beim Thema sind.
Apple-Gründer zu Cloud, Prism und iOS 7:
Wozniak vergleicht Cloud mit Kommunismus
http://www.crn.de/software/artikel-99658.html?cid=CRNnewsletter-cloudchannel-html
Seine Haltung zu IOs passt dann aber auch nicht so recht.
Was ich aber Z.b. von Turn 10 oder Respawn verstanden habe ist, das z.B. die Ki / Bots in der Cloud berechnet werden kann und die one diese freigewordene Rechenleistung für andere dinge nutzen kann.
Passt das oder ist das auch nur geblubber ?
Selbst wenn sich keine Mehrleistung z.B. bei Forza ergibt, so sind mir die gespeicherten Spielerdaten lieber als eine "normale" Ki. Wenn es denn so stimmt wie erzählt wird.
Hier noch was dazu aus dem Forza Thread.
You said that Drivatar offers “cloud-powered opponents.” Can you tell us a bit more about that?
Drivatar is like nothing else. I've actually been struggling all week to describe it. Everyone's looking for, “is like ghosts or is it like replays?” No, that's what Rivals is. We have Rivals Mode, and Rivals Mode is great, it's gonna be in Forza Motorsport 5 and that is racing against other people's ghosts and you earn money for beating them, and that's really cool. We have multiplayer, dedicated servers, big matches; again, great racing. But Drivatar is not like AI, it's more like big data, it's more like search. In AI you've got a number of parameters that you can fill in either manually or by using learning, so we had learning in the past, or 'Drivatar 1.0', which was in the original Forza. It was learning AI, but it had a set, finite number of parameters that would fill up through learning. Or you could have a designer train it by filling in numbers on a spreadsheet like a traditional racing game.
This is different, this is big data, so we log a lot of data: where were the other cars when you made the decisions you make? Did you set up a complex pass? [Did you] bank right, dive left? How did you actually set that up? We log all of that and the system doesn't actually need to know anything about it. For example, there's a car in front of me, I'm following up behind him more or less straight away, and it hits its brakes. Well, in race craft we call that 'check braking'. The system doesn't need to know that, the system just needs to see that say of a million people, 300,000 did it and now it learns, 'hey, there's this thing that people do and here's where they tend to do it. These tracks, these cars, and all the other cars are in these positions or not, and they do it against their friends or not against their friends, and it generalises and learns these parameters.
Now every Drivatar can't necessarily do that, so you have to then train your Drivatar to do it. So, it's an ever-evolving system, it's learning race craft. Now it already knows a lot of race craft right now because we've got a lot of people on our team and we're not bad racers, we're racing all the time, and we're also pulling in people all the time to further train it. You just have to imagine millions of people racing for hours every night, and how the system just gets bigger and bigger and bigger and starts learning more and more things, so basically the game at launch is going to be a different game a month later.
Where the rubber meets the road though is the difference in racing. Back at the studio I saw three cars go cleanly three abreast through a [corner]. AI doesn't do that. Show me a game where you've seen cars go three abreast through a corner. It's super rare.
Wo wir schon beim Thema sind.
Apple-Gründer zu Cloud, Prism und iOS 7:
Wozniak vergleicht Cloud mit Kommunismus
http://www.crn.de/software/artikel-99658.html?cid=CRNnewsletter-cloudchannel-html
Seine Haltung zu IOs passt dann aber auch nicht so recht.
