This Stalker Micro is absolutely Mind blowing.

I know its a bot but still

funny comment:

"AI Is learning how to control a group of killer robots. This is fine."

RIP Humans

even though they initially said they'd limit the AI's input and apm, they certainly did not
the Ai can see the whole map and her APM does not seem like it'd be limited
the research is pretty retarded at this point because the most primitive AI could beat the best humans with inhumane control and macro, strategy does not even matter at that point.
they should completely limit the input to what humans can do
I don't think the AI even uses a mouse cursor to select units at this point

wintergaming >

Skynet is coming yes.


Imagen a swarmhost, would make thouse stalkers run from midmap to base, over and over, on every locust wave

They did limit it, they showed graphs, but the problem is that even with limited APM he needed far less clicks because each click was an efficient one.

I think it was very interesting to watch, but obviously yeah the AI's ability to control blink stalkers was unfair to say the least.
And viewing the entire map was also something that was in his advantage, I think nevertheless it's a cool development.

I feel like a lot of people are a little bit too butthurt to actually look at what an achievement it is that they developed it this far.

I mean I was really hyped for it, but this was really a disappointment
we haven't really seen the AI out think anyone, it just simply brute forced his was with perfect micro and macro with little strategic thinking.
not impressed yet.

define "out think"

the AI has a complex solution space to fullfill its agenda. it is way too big for bruteforce only. it developed its pathways through this space using many tools and doesn't really matter where it used brute force.

having a limited action pool only adds an additional layer, as this new resource needs to be best used/timed

if you expected anything else, you were bound to be disappointed

eg. create a diversion but i doubt the AI even has or will have a concept of it because it does not understand the limitations of the other player because it is not limited for the same input as in having a limited screen and having to use a mouse and a keyboard
what it has accomplished so far is not impressive especially without the restrictions I've mentioned

This post was edited by Cpt_Ghost on Feb 11 2019 02:03pm

I only watched like half of the video you linked on the hungarian sub, but as far as that went this thing only really played against itself sofar

it would need some serious playtime against humans, to learn about playing against them. thats not very easy to do, as our clockspeed doesnt scale very well.

you're right, I am sure there is a garage on every corner with kids cranking up AIs like this

the only reason for alphastar being hard to beat is that he does not make mistakes and its micro and macro is top notch
yea, it's not hard to come across AI's capable of that, the only difference at the moment is that alphastar is self learning on top of that, which in itself is a feat for sure. But the reason i'm not impressed is that its just not impressive to beat a human under these conditions because it isn't even close to a fair fight, so it is false to advertise alphastar with "beating pros in starcraft"