EPTOpenAI's Dota 2 bots vs. 5 top professionals in TI - Page 5
| Forum Index > Dota 2 General |
|
rabidch
United States20289 Posts
| ||
|
Murlox
France1699 Posts
Sure there's "bot" in both, but I think the comparison ends pretty much here... | ||
|
codeafin
7 Posts
On July 16 2018 13:16 rabidch wrote: would be cool if openai did their own try at bot ti It would be, and the most interesting part wouldn't be their first run at it but their 500th run after they've done their internal learning See what creative builds and tactics they come up with | ||
|
spudde123
4814 Posts
They announced that they will allow picking from a pool of 18 heroes and some restrictions like wards and roshan are gone. At least this makes it much more interesting to watch, and the humans will be able to draft a sensible lineup instead of playing a mirror match. They'll be playing a team with Cap, Blitz, Fogged, Merlini +1 in the end of this month. Interesting to see how it looks. It's one thing to lift the restrictions and it should be done to make the game more like actual dota, but it'll be interesting to see how the bots actually use wards for example. I'd be quite surprised if the bots truly excel at all aspects of the game. Also not sure how wards for example are implemented. Previously the research team hardcoded item builds for the bots, so I assume they also hard code when they buy wards or not rather than the bots yet being able to learn it themselves. | ||
|
Sn0_Man
Tebellong44238 Posts
They have almost 3 weeks to train this version of the bot though, which is a long time with the amount of computing power they throw at it. | ||
|
polgas
Canada1770 Posts
August 5 Benchmark Test Restrictions: - Random Draft using a pool of 18 heroes (Crystal Maiden, Death Prophet, Earthshaker, Gyrocopter, Lich, Lion, Necrophos, Queen of Pain, Razor, Riki, Shadow Fiend, Slark, Sniper, Sven, Tidehunter, Viper, or Witch Doctor) - No Divine Rapier, Bottle - No summons/illusions - 5 invulnerable couriers, no exploiting them by scouting or tanking - No Scan - - - | ||
|
WolfintheSheep
Canada14127 Posts
On July 19 2018 02:35 Sn0_Man wrote: It'll be interesting to see what the draft looks like. They have almost 3 weeks to train this version of the bot though, which is a long time with the amount of computing power they throw at it. Not sure how "Random Draft" works with only a pool of 18 heroes. Either way, will be fun to see which roles are prioritized by AI. Glad to see warding, Roshan and invisibility have been integrated now. Minus the invincible 5 couriers, it'll look close to a real match scenario. Viper and Shadow Fiend are really the only heroes there that might semi-regularly use Manta Style. I think they might have actually overcompensated with the reaction time. A lot of pros probably have a sub 200ms reaction, even accounting for input lag. | ||
|
Wineandbread
United States2065 Posts
I'm very interested though with a lot of those restriction lifted. I hope they get a bo3 in against the caster crew too. Wonder who their last guy is gonna be. Lacoste maybe? Or a pro NA player in SF? I assume the pro players also get their own set of 5 invulnerable couriers? | ||
|
polgas
Canada1770 Posts
| ||
|
sertas
Sweden890 Posts
| ||
|
Sn0_Man
Tebellong44238 Posts
On July 20 2018 18:49 sertas wrote: Why dont they have liquid or any other pro team play against the bot 5v5? Considering that this event is just 10 days before TI starts, I imagine liquid will be a little preoccupied around then (as should any team at TI). I know that blitz asked some other TI players to join him and got turned down. | ||
|
sertas
Sweden890 Posts
| ||
|
Sn0_Man
Tebellong44238 Posts
| ||
|
paladin-
10 Posts
Seeing some next lvl warding by the bots is hopefully going to be cool though. | ||
|
gaymon
Germany1023 Posts
| ||
|
polgas
Canada1770 Posts
| ||
|
spudde123
4814 Posts
It's quite hard to assess what they still have to be able to improve because we have really no idea how the bots won and lost. Simply saying that they won in some matches isn't so interesting to me. It would be nice if they would release these sorts of matches for viewing. The audience event would also be a better benchmark if the humans could see the bots in action before they play them, even if it's a previous version. | ||
|
Pontual
Brazil3038 Posts
| ||
|
heinz744
3 Posts
| ||
|
polgas
Canada1770 Posts
| ||
| ||
