|
Did VG Close the Poll? This poll is closed. |
|||
|---|---|---|---|
| Yes |
|
1 | 100.00% |
| No |
|
0 | 0% |
| Total: | 1 votes | ||
|
Apparently they do not defend.
|
#
?
Aug 6, 2018 00:17
|
|
|
|
| # ? Jun 7, 2024 19:36 |
|
|
Error: bots not programmed to defend highground QoP actively bought back and then blinked out of the base mid to push creeps
|
|
#
?
Aug 6, 2018 00:18
|
|
|
It looks like they defend by split pushing
|
|
#
?
Aug 6, 2018 00:18
|
|
|
Bots would have won ti4 and ti7
|
|
#
?
Aug 6, 2018 00:18
|
|
|
It will be interesting to see how the bots do when they get full micro and can rat like right fuckers.
|
|
#
?
Aug 6, 2018 00:19
|
|
|
Yes Purge, that was a rage buyback from the AI.
|
|
#
?
Aug 6, 2018 00:23
|
|
|
https://twitter.com/syndereNDota/status/1026247215386378242
|
|
#
?
Aug 6, 2018 00:24
|
|
|
Very impressive showing from OpenAI. They clearly understand how to play the earlygame, presumably because most of their games are short. But it's clear that they have no idea how to play from behind, expecting to be able to go for aggressive ganks and push plays. Still, huge advancement from OpenAI, I'm excited to see how much better they can make it. When you think that 1 year ago it could only play 1 hero, in one position, with huge restrictions on items and game rules. I expect it'll be able to beat most pro teams with a really wide hero pool by this time next year.
|
|
#
?
Aug 6, 2018 00:25
|
|
|
I'm just following the thread from a baseball game but 
|
|
#
?
Aug 6, 2018 00:28
|
|
|
Im Sick Of OpenAI Being Immune From Punishment, They Just Ran Down Mid Again ON STREAM And They Didn't Get Low Prio!
|
|
#
?
Aug 6, 2018 00:32
|
|
|
it'd be cool if the TI7 champs face open AI instead of 5 random pros. alternatively Fnatic should play them
|
|
#
?
Aug 6, 2018 00:40
|
|
|
It's interesting that the bots really didn't know how to properly play Slark or Sven but they were almost there. Wonder if more games would solve that or does the AI need to learn it from humans or something. An algorithm that could learn from not only its own games but also from the millions of games already played would seem to be optimal.
|
|
#
?
Aug 6, 2018 00:41
|
|
|
goatface posted:Yes Purge, that was a rage buyback from the AI. Are you serious?
|
|
#
?
Aug 6, 2018 00:46
|
|
|
No, and I'm fairly sure he wasn't either.
|
|
#
?
Aug 6, 2018 00:48
|
|
|
goatface posted:No, and I'm fairly sure he wasn't either. Yes no kidding.
|
|
#
?
Aug 6, 2018 00:49
|
|
|
Kild posted:it'd be cool if the TI7 champs face open AI instead of 5 random pros. alternatively Fnatic should play them I don't think being unpredictable would really effect this iteration of the bots at all because their gameplan is pretty much the same every time - pick the same heroes every game, win lanes, deathball mid. You can't pull a PLD and camp the courier mid, trying to splitpush against them wouldn't make them back off, there's no NP to make unpredictable rotations on them, there's not even a ton of great anti-deathball heroes like Veno or Warlock who I think would annihilate the bots. The limited hero pool with no bans and unkillable couriers basically guarantee that the one strategy the bots know, they're able to reliably do every single time and there's very little anyone could do to disrupt it. You saw that even not being able to get the heroes the bots were used to put them at a 98% loss chance before the game even started. There's only sixteen heroes in the pool and the bot had no idea what to do on a couple of them. Like I said above, I think the limited hero pool favored the bots also - I think if you give the humans either Veno, Warlock, or Bloodseeker they could have beaten the bots. This was a cool demo, but I think the AI has SIGNIFICANTLY longer way to go before they're able to contend with actual teams in a real tournament gamemode than people are claiming.
|
|
#
?
Aug 6, 2018 00:58
|
|
|
Really it just proves how broken Gyro was. The bot learned "Draft Gyro, last hit, five-man"
|
|
#
?
Aug 6, 2018 01:16
|
|
|
Yeah their gameplan is win the lanes as relatively hard as possible then group up and death ball with all the regen they ferry out.
|
|
#
?
Aug 6, 2018 01:30
|
|
|
according to a reply it's actually 180 years per day.
|
|
#
?
Aug 6, 2018 01:34
|
|
|
They probably have a simulation running with no restrictions already, but they probably don't feel that it's ready. (And I'm sure they talked about it in the Q&A but I didn't watch it). The bots exploited the mechanics perfectly but it didn't always feel like DotA. You aren't going to win normal DotA getting a Null first with zero regen because you can't rely on a personal courier. Btw if anyone wants a good tutorial/intro example to machine learning I always liked this video https://www.youtube.com/watch?v=qv6UVOQ0F44
|
|
#
?
Aug 6, 2018 01:41
|
|
|
schmitty9800 posted:You aren't going to win normal DotA getting a Null first with zero regen because you can't rely on a personal courier. Yeah I was hoping this was going to be another thing like the 1v1 OpenAI situation where it was something people could learn from. The double mango strategy that people adopted happened because of how powerful the AI showed it was, but there wasn't a ton here that people could take into their games. Shipping regen with a personal courier is already how you win Turbo mode.
|
|
#
?
Aug 6, 2018 01:48
|
|
|
Seems like the bots effective hero pool isn't actually 18 heroes.
|
|
#
?
Aug 6, 2018 01:51
|
|
|
Yeah there's a reason they didn't up the pool to 22 so you could do a standard Captain's Draft with 12 bans. I think if you gave the humans 3 bans at the beginning of the draft and they banned Gyro/CM/Lion the bots would be hosed
|
|
#
?
Aug 6, 2018 02:00
|
|
|
TOOT BOOT posted:Seems like the bots effective hero pool isn't actually 18 heroes. Seems like the allowable heroes list is based on the following inputs: -The ~10 heroes that AI can win with -Whatever heroes said AI hero pool counters -Any heroes that don't counter AI hero pool While I am impressed that OpenAI can win with the given parameters, it seems really telling that illusion and one killable courier per team is enough to significantly nerd the current build.
|
|
#
?
Aug 6, 2018 02:49
|
|
|
It's really stupid you can use 1 report to report for all 3 categories and actually have it count for all 3 categories.
|
|
#
?
Aug 6, 2018 07:37
|
|
|
quote:<ambivalent> also I think moonmeander was trying to put on a bit of a show but just died to the bots a bunch
|
|
#
?
Aug 6, 2018 08:28
|
|
|
[8:26pm] <craigk> "yeah [the ai bots] destroyed even capatalist, merlina, fogged, moonmeander" [8:26pm] <craigk> well i mean not the first time machinery has tripped up moonmeander
|
|
#
?
Aug 6, 2018 11:03
|
|
|
I think the games would be more interesting/more useful for learning for the bots if instead of personal invuln couriers they just had no couriers. As it is, they've just hyper-refined the deathball with regen strat and it's not very representative of 'real' matches.
|
|
#
?
Aug 6, 2018 12:16
|
|
|
I wonder what happens if they put two of mid-bot and three 5-bots in a game.
|
|
#
?
Aug 6, 2018 20:34
|
|
|
Did that Xyclopz situation get resolved? BTS has him casting right now and I wonder if Valve cares.
|
|
#
?
Aug 7, 2018 06:54
|
|
|
he lost his artifact access and probably won't get access to all those other games valve is developing. xyclopz rules because he just plays whatever music he wants instead of the royalty free stuff BTS usually have on intermission
|
|
#
?
Aug 7, 2018 13:45
|
|
|
he’s bad at his job tho.
|
|
#
?
Aug 7, 2018 14:39
|
|
|
yeah this thread exists in a weird parallel universe where xyclopz isnt horrendous
|
|
#
?
Aug 7, 2018 14:41
|
|
|
huh
|
|
#
?
Aug 7, 2018 14:55
|
|
|
https://twitter.com/wykrhm/status/1026846803386224643 
|
|
#
?
Aug 7, 2018 16:37
|
|
|
Rubick or void please
|
|
#
?
Aug 7, 2018 16:48
|
|
|
Rubick's arcana is going to be such a disappointment and I'm going to laugh
|
|
#
?
Aug 7, 2018 16:56
|
|
|
its going to be hilarious when Invoker beats Rubick
|
|
#
?
Aug 7, 2018 17:00
|
|
|
At least sniper didn't win
|
|
#
?
Aug 7, 2018 17:07
|
|
|
|
| # ? Jun 7, 2024 19:36 |
|
|
rubick's definitely gonna win, look at that loving vote total
|
|
#
?
Aug 7, 2018 17:31
|
|