31–40 of 138 posts
Earlier quoted context omitted.
Even if they did release the likely-very-big model, your personal computer is likely not fast enough to make 2-3 actions per second (x5) and update weights in real time without a beefy GPU.
You don't need to update any weights, the model is already trained.
Earlier quoted context omitted.
It's not that exciting and it's very easy to beat humans at games when your reaction times are far faster and you have access to more information at a single moment in time. It's like being impressed by a calculator.
"We’ve increased the reaction time of OpenAI Five from 80ms to 200ms. This reaction time is much closer to human level, though we haven’t seen evidence of changes in gameplay as OpenAI Five’s strength comes more from teamwork and coordination than reflexes."
Earlier quoted context omitted.
It's not that exciting and it's very easy to beat humans at games when your reaction times are far faster and you have access to more information at a single moment in time. It's like being impressed by a calculator.
What information do they have access to? I do think a simulated reaction time or limited actions per second might be more fair, if they don’t have that already..
Earlier quoted context omitted.
It's not that exciting and it's very easy to beat humans at games when your reaction times are far faster and you have access to more information at a single moment in time. It's like being impressed by a calculator.
dota 2 is not as extreme about this as, say, starcraft. in starcraft, most "strategies" are heavily scripted build orders that almost never deviate from a handful of openings (similar to opening moves in chess) and a large amount of winning comes from the ability to quickly give orders to your troops ("micro"). This is an example of a strategy that is not viable without superhuman reactions: https://www.youtube.com/w…
As somebody who plays StarCraft casually (gold/low plat in ladder), this is not true. It's even less true for pro players. The level of strategy in StarCraft is impressive, it's really hard to guess in which direction games will go when two very good players are playing against each other.
Sure, perfect execution when it comes to one strategy (say, mech-heavy Terran) will give you the largest advantage against your opponent, but failing to scout appropriately and guess what your opponent is up to means your strategy is dead. You also have to decide when to attack, how much you're willing to sacrifice to damage somebody's economy, when you want to focus one economy vs building units, ...
The video you sent with zerglings is a gimmick made for fun (it's a hard-coded AI using the siege tank's aim logic to divert zerglings from that). That would not win you a game. (because most likely a pro Terran would have destroyed your base before that)
Congrats OpenAI on game #1
Very impressed with the bots. The main strategy I see from the bots that is different from pro meta is how they spam abilities aggressively. Now the main limitation is 5 unkillable couriers which enable this. Watching, I do feel like given more practice humans would beat this version of the ai easily. It looks exploitable, and still having many rules limitations it has a ways to go. It will win in the end though, tha…
Earlier quoted context omitted.
"We’ve increased the reaction time of OpenAI Five from 80ms to 200ms. This reaction time is much closer to human level, though we haven’t seen evidence of changes in gameplay as OpenAI Five’s strength comes more from teamwork and coordination than reflexes."
Notably a lot of the clutch plays in Game #1 by OpenAI were due to very well timed skills.
Earlier quoted context omitted.
Even if they did release the likely-very-big model, your personal computer is likely not fast enough to make 2-3 actions per second (x5) and update weights in real time without a beefy GPU.
Thank you for the reply minimaxir - are you sure a 2015 macbook pro cannot run a compiled version of the neural net so it can play itself on my computer? Just wanted to double check - can someone else verify this before I spend weeks seeing if I can get it to work?