• Log InLog In
  • Register
Liquid`
Team Liquid Liquipedia
EST 02:43
CET 08:43
KST 16:43
  • Home
  • Forum
  • Calendar
  • Streams
  • Liquipedia
  • Features
  • Store
  • EPT
  • TL+
  • StarCraft 2
  • Brood War
  • Smash
  • Heroes
  • Counter-Strike
  • Overwatch
  • Liquibet
  • Fantasy StarCraft
  • TLPD
  • StarCraft 2
  • Brood War
  • Blogs
Forum Sidebar
Events/Features
News
Featured News
Team Liquid Map Contest #235[ASL22] Finals Preview: Sweet Victory!12[ASL22] Ro4 Preview: Mirror Mirror6[ASL22] Ro8 Preview: Within Reach5[ASL22] Ro8 Preview: In A Tizzy11
Community News
Weekly Cups (Nov 9-15): Clem and herO take over1Dark completes military service, returns to SC215[BSL23] RO16 Group Stage20Weekly Cups (Nov 2-8): herO back on top?2Weekly Cups (Oct 26 - Nov 1): MaxPax, herO, Solar win0
StarCraft 2
General
Team Liquid Map Contest #23 Dark completes military service, returns to SC2 Weekly Cups (Nov 9-15): Clem and herO take over StarCraft open world shooter announced at BlizzCon No more SC2 streamers on team liquid?
Tourneys
Sparkling Tuna Cup - Weekly Open Tournament RSL goes to London! 2026 Offline Finals Nov 21-22 Sea Duckling Open (Global, Bronze-Diamond) Dark returns! GSL CK #7 4v4 Team Battle (Nov 19) soO & FanTaSy return! SC2 Kevin Scrim 4v4 (Nov 16)
Strategy
[H] ZvP Mid-Late Game: Stalkers Collossi HT
Custom Maps
[G] How to playtest arcade maps Starcraft 2 Advance [D] the control group Simple Questions/Answers
External Content
Mutation # 552 Secret Storm The PondCast: SC2 News & Results Mutation # 551 Robotic Revival Mutation # 550 Graveyard Shift
Brood War
General
List of top players active on ShieldBattery BW General Discussion NaDa's Body Shield Battery Server New Patch BGH Auto Balance -> http://bghmmr.eu/
Tourneys
SC1 Kevin Scrim hosted by SOOP (Nov 18-19) [IPSL] Winter 26-27 Round of 16! [Megathread] Daily Proleagues 2v2v2v2 Tournament Stage 2: Players
Strategy
Pluto's TvP 5Marine 2Medic Push Simple Questions, Simple Answers Infested Terran strategies in 2026 Nuke strategy worked in K-League 2026-11-09
Other Games
General Games
Nintendo Switch Thread Stormgate/Frost Giant Megathread Free Space RTS Game, optimised f. Phone/Tablet etc List of Game Threads The PlayStation 5
Dota 2
League of Legends
[TL LoL EUW IHs] Teemo shall perish
Heroes of the Storm
Heroes of the Storm 2.0
Hearthstone
Deck construction bug
TL Mafia
TL Mafia Community Thread
Community
General
US Politics Mega-thread Trading/Investing Thread Artificial Intelligence Thread Russo-Ukrainian War Thread Things Aren’t Peaceful in Palestine
Fan Clubs
Fan Club Resource Thread Reynor Fan Club The Flash Fan Club
Media & Entertainment
Movie Discussion! Anime Discussion Thread [Manga] One Piece
Sports
Football (Soccer) Thread TeamLiquid Health and Fitness Initiative For 2023
World Cup 2022
Tech Support
Computer Build, Upgrade & Buying Resource Thread
TL Community
The Automated Ban List Recent Gifted Posts
Blogs
The time I organised my…
AssyrianKing
Why I built Repalyzer (1.1.…
BlueStar
Innovate or Game Over?
TrAiDoS
ShieldBattery on Mac
Kraekkling
Can Bots Beat Pros?? Starcr…
namkraft
StarCraft improvement
iopq
Socialism Anyone?
GreenHorizons
Customize Sidebar...

Website Feedback

Closed Threads



Active: 10124 users

an AI researcher's take on the ladder bot - Page 3

Forum Index > BW General
Post a Reply
Prev 1 2 3 All
StRyKeR
Profile Blog Joined January 2006
United States1756 Posts
Last Edited: 2026-09-26 21:42:19
September 26 2026 21:41 GMT
#41
On September 23 2026 12:20 evanthebouncy! wrote:
one thing I noticed after watching more of Pluto's playing is that as soon as it attacks, the opponent loses in 30 seconds.

I think the reasons are as follows:
1. pluto is trained in self-play, so it imagined its human opponent to be itself, i.e. with perfect micro
2. pluto maintains some kind of "army value advantage" (let's call it AVA) estimation, to decide if it can beat or not beat the opponent army
3. the online pluto uses maphack, so it has a perfect information to calculate AVA
4. so when pluto commits to an attack, it is imagining that it can beat the perfectly microed opponent army
5. but human's micro is far worse than pluto, so human will lose a chunk of relative army value
6. pluto's AVA goes wayyy up, and goes into a frenzy and human dies


On the flip side, it's way too conservative as Zerg as a result. It thinks its Terran opponent can click hack and precisely target hurt Mutas, so it never harasses with Mutas. It always either runs away immediately or completely commits to a fight. It never tries to trade.
Ars longa, vita brevis, principia aeturna.
PurpleWave
Profile Joined January 2021
24 Posts
Last Edited: 2026-09-26 22:36:21
September 26 2026 22:20 GMT
#42
On September 27 2026 04:00 ywxr wrote:
I wonder if the creator is moving away from pure W/L-signal RL + strategy bandit to try to learn efficiently from these human games (which would be a uselessly small sample size otherwise), or from relatively small amounts of high-quality counterfactual-pair training data on things like drops and dark swarm. Given that it's already got a probably pretty well-calibrated win probability/value head, I think fine-tuning it to learn these things could probably be done very cheaply.


The replays are I think not likely useful for training. You don't want to imitate the humans because the humans are worse players. And because the current version (Artosis games, but presumably not the Korean pro games) is a newly trained model, the replays are very off-policy.
ywxr
Profile Joined September 2026
4 Posts
September 26 2026 23:52 GMT
#43
On September 27 2026 07:20 PurpleWave wrote:
Show nested quote +
On September 27 2026 04:00 ywxr wrote:
I wonder if the creator is moving away from pure W/L-signal RL + strategy bandit to try to learn efficiently from these human games (which would be a uselessly small sample size otherwise), or from relatively small amounts of high-quality counterfactual-pair training data on things like drops and dark swarm. Given that it's already got a probably pretty well-calibrated win probability/value head, I think fine-tuning it to learn these things could probably be done very cheaply.


The replays are I think not likely useful for training. You don't want to imitate the humans because the humans are worse players. And because the current version (Artosis games, but presumably not the Korean pro games) is a newly trained model, the replays are very off-policy.


I wasn't suggesting imitating humans. If the value head is broadly well-calibrated in a generalizable way, specific missing skills like drops/dark swarm could be fine-tuned very efficiently using synthetic data of counterfactual pairs. I agree the actual replays aren't necessarily that useful, but in this case it might not matter too much that they're off-policy. And with generalizable enough calibration, I think there could even be a way for it to be a few-shot online learner, which would be really cool.
vndestiny
Profile Blog Joined September 2011
Singapore3475 Posts
Last Edited: 2026-09-27 01:26:49
September 27 2026 01:25 GMT
#44
I’m wondering whether the training model is steer-able deep into mid game.

Self-play training games are able to be steered/ weighed with certain openings.

However for current major weakness of Pluto: dealing with drop (reaver/shuttle/dropship), inability to use drop, building placement (nexus/cc/hatch & wall), understanding darkswarm ability.

These might take millions & billions more games and still not discover them in reinforce selfplay by chance (sounds like evolution ha).
IntoTheWow
Profile Blog Joined May 2004
is awesome32282 Posts
September 27 2026 02:50 GMT
#45
Just dropping by to say that:

- it's great to see your name here again!
- Good context on how the bot works. What other training strategies could be used. I guess the other extreme would be for it to study a set of curated (high ELO) replays. Is there a way to reach the best of both worlds? Give it a baseline based on current human know ledge and let it push the boundaries of infinite APM + attention?
Moderator<:3-/-<
PurpleWave
Profile Joined January 2021
24 Posts
September 27 2026 03:01 GMT
#46
On September 27 2026 11:50 IntoTheWow wrote:
Is there a way to reach the best of both worlds? Give it a baseline based on current human know ledge and let it push the boundaries of infinite APM + attention?


This is what AlphaStar did (minus infinite APM, which like Pluto they somewhat limited):

- Train a model to imitate humans, over 1 million Battle.net replays. This model was ~Gold league as-is
- Copy the model many times over, and train them against one another, with different reward signals for each (reward winning + preference different units),

When they trained with just reinforcement learning they didn't manage to get behavior more sophisticated than worker rushes. The imitation step was required. So Pluto requiring no imitation step is impressive.
evanthebouncy!
Profile Blog Joined June 2006
United States12816 Posts
September 27 2026 03:16 GMT
#47
On September 27 2026 05:50 419 wrote:
The bit about Pluto calculating likes its playing against itself makes me wonder if Pluto having contempt would make it even more effective against human players.


yeah I think to beat humans Pluto needs a kind of "skill check" mechanism, which is something that can be adjusted at inference time.

for instance, it can measure the difference of "predicted reward" based on its neural network, and "observed reward" based on how the game actually panned out. For example, Pluto might have calculated that an engagement it will lose 8 marines worth of "score", however after the actual engagement it only lost 2 marines worth of "score". Then it can realize that "oh wow this opponent micros terribly" and now that can be used as additional signal to steer its behaviour, for instance, being more aggressive.
Life is run, it is dance, it is fast, passionate and BAM!, you dance and sing and booze while you can for now is the time and time is mine. Smile and laugh when still can for now is the time and soon you die!
evanthebouncy!
Profile Blog Joined June 2006
United States12816 Posts
September 27 2026 03:20 GMT
#48
On September 27 2026 06:41 StRyKeR wrote:
Show nested quote +
On September 23 2026 12:20 evanthebouncy! wrote:
one thing I noticed after watching more of Pluto's playing is that as soon as it attacks, the opponent loses in 30 seconds.

I think the reasons are as follows:
1. pluto is trained in self-play, so it imagined its human opponent to be itself, i.e. with perfect micro
2. pluto maintains some kind of "army value advantage" (let's call it AVA) estimation, to decide if it can beat or not beat the opponent army
3. the online pluto uses maphack, so it has a perfect information to calculate AVA
4. so when pluto commits to an attack, it is imagining that it can beat the perfectly microed opponent army
5. but human's micro is far worse than pluto, so human will lose a chunk of relative army value
6. pluto's AVA goes wayyy up, and goes into a frenzy and human dies


On the flip side, it's way too conservative as Zerg as a result. It thinks its Terran opponent can click hack and precisely target hurt Mutas, so it never harasses with Mutas. It always either runs away immediately or completely commits to a fight. It never tries to trade.


exactly! and that's how humans wind up beating Pluto by just making a huge army and never actually fight it :D
pluto will just hallucinate this impossibly strong opponent and in its immediate reward function to "reduce unit score loss" never take a fight

I've not seen Pluto take any hail mary attempts, because it is so caught up in optimizing the short-term gain. so Pluto will be very weak in actual positional plays, where you sacrifice the "now" for the "future". For this same reason it never figures out why upgrades are useful either.

Life is run, it is dance, it is fast, passionate and BAM!, you dance and sing and booze while you can for now is the time and time is mine. Smile and laugh when still can for now is the time and soon you die!
StRyKeR
Profile Blog Joined January 2006
United States1756 Posts
September 27 2026 04:44 GMT
#49
On September 27 2026 12:20 evanthebouncy! wrote:
Show nested quote +
On September 27 2026 06:41 StRyKeR wrote:
On September 23 2026 12:20 evanthebouncy! wrote:
one thing I noticed after watching more of Pluto's playing is that as soon as it attacks, the opponent loses in 30 seconds.

I think the reasons are as follows:
1. pluto is trained in self-play, so it imagined its human opponent to be itself, i.e. with perfect micro
2. pluto maintains some kind of "army value advantage" (let's call it AVA) estimation, to decide if it can beat or not beat the opponent army
3. the online pluto uses maphack, so it has a perfect information to calculate AVA
4. so when pluto commits to an attack, it is imagining that it can beat the perfectly microed opponent army
5. but human's micro is far worse than pluto, so human will lose a chunk of relative army value
6. pluto's AVA goes wayyy up, and goes into a frenzy and human dies


On the flip side, it's way too conservative as Zerg as a result. It thinks its Terran opponent can click hack and precisely target hurt Mutas, so it never harasses with Mutas. It always either runs away immediately or completely commits to a fight. It never tries to trade.


exactly! and that's how humans wind up beating Pluto by just making a huge army and never actually fight it :D
pluto will just hallucinate this impossibly strong opponent and in its immediate reward function to "reduce unit score loss" never take a fight

I've not seen Pluto take any hail mary attempts, because it is so caught up in optimizing the short-term gain. so Pluto will be very weak in actual positional plays, where you sacrifice the "now" for the "future". For this same reason it never figures out why upgrades are useful either.



Indeed. Despite how strong it is right now, it's beatable, especially if you abuse its blind spots (like not knowing what Dark Swarm is).

There's a lot of ways it can get even stronger. If people actually continue to work on this it's not hard to imagine a version that beats Pluto almost every time.

The creator mentioned publishing a paper on this research. I wonder if he'll be less interested in pursuing further research once he achieves that.
Ars longa, vita brevis, principia aeturna.
Jealous
Profile Blog Joined December 2011
10424 Posts
Last Edited: 2026-09-27 04:57:08
September 27 2026 04:54 GMT
#50
On September 27 2026 13:44 StRyKeR wrote:
Show nested quote +
On September 27 2026 12:20 evanthebouncy! wrote:
On September 27 2026 06:41 StRyKeR wrote:
On September 23 2026 12:20 evanthebouncy! wrote:
one thing I noticed after watching more of Pluto's playing is that as soon as it attacks, the opponent loses in 30 seconds.

I think the reasons are as follows:
1. pluto is trained in self-play, so it imagined its human opponent to be itself, i.e. with perfect micro
2. pluto maintains some kind of "army value advantage" (let's call it AVA) estimation, to decide if it can beat or not beat the opponent army
3. the online pluto uses maphack, so it has a perfect information to calculate AVA
4. so when pluto commits to an attack, it is imagining that it can beat the perfectly microed opponent army
5. but human's micro is far worse than pluto, so human will lose a chunk of relative army value
6. pluto's AVA goes wayyy up, and goes into a frenzy and human dies


On the flip side, it's way too conservative as Zerg as a result. It thinks its Terran opponent can click hack and precisely target hurt Mutas, so it never harasses with Mutas. It always either runs away immediately or completely commits to a fight. It never tries to trade.


exactly! and that's how humans wind up beating Pluto by just making a huge army and never actually fight it :D
pluto will just hallucinate this impossibly strong opponent and in its immediate reward function to "reduce unit score loss" never take a fight

I've not seen Pluto take any hail mary attempts, because it is so caught up in optimizing the short-term gain. so Pluto will be very weak in actual positional plays, where you sacrifice the "now" for the "future". For this same reason it never figures out why upgrades are useful either.



Indeed. Despite how strong it is right now, it's beatable, especially if you abuse its blind spots (like not knowing what Dark Swarm is).

There's a lot of ways it can get even stronger. If people actually continue to work on this it's not hard to imagine a version that beats Pluto almost every time.

The creator mentioned publishing a paper on this research. I wonder if he'll be less interested in pursuing further research once he achieves that.

If I recall correctly, tscmoo has been working on BW AI bots for a decade, if not longer. Even though he is on top of the bot ladder (and briefly the human ladder!), somehow I don't think his interest will fade just because he is on top or he published a paper The BW AI competition landscape could get very interesting after such a systemic shock (not that it wasn't already!).

E: Definitely more than a decade: (Wiki)Tscmoo
"The right to vote is only the oar of the slaveship, I wanna be free." -- бум бум сучка!
evanthebouncy!
Profile Blog Joined June 2006
United States12816 Posts
September 27 2026 07:08 GMT
#51
On September 27 2026 04:00 ywxr wrote:
I wonder if the creator is moving away from pure W/L-signal RL + strategy bandit to try to learn efficiently from these human games (which would be a uselessly small sample size otherwise), or from relatively small amounts of high-quality counterfactual-pair training data on things like drops and dark swarm. Given that it's already got a probably pretty well-calibrated win probability/value head, I think fine-tuning it to learn these things could probably be done very cheaply.


I think . . . the easy way to train it to learn about drop/dark-swarm is to find ways to synthetically generate a lot of data of games containing them, in the order of 10s of millions, in order to train pluto.
There is already some exploration intrinsic reward for making more niche units (that's why you see it upgrading ghosts or making a single corsair), except this incentive is not strong enough for it to climb the tech-tree all the way to dark swarm

A rather crude approach is to take some human games and train a kind of unit composition reward function, and use that reward function in its own game. For instance if defilers would give a high reward, the bot would learn to make them first, and then through sheer luck figure out how consume / dark-swarm works through millions of trials and error

on fine-tuning the winprob/value head, I'm not so sure. Clearly it is not calibrated against stuff like dark swarm and drop. I think those things do require significant amount of additional training data to converge
Life is run, it is dance, it is fast, passionate and BAM!, you dance and sing and booze while you can for now is the time and time is mine. Smile and laugh when still can for now is the time and soon you die!
evanthebouncy!
Profile Blog Joined June 2006
United States12816 Posts
September 27 2026 07:42 GMT
#52
On September 27 2026 13:44 StRyKeR wrote:
Show nested quote +
On September 27 2026 12:20 evanthebouncy! wrote:
On September 27 2026 06:41 StRyKeR wrote:
On September 23 2026 12:20 evanthebouncy! wrote:
one thing I noticed after watching more of Pluto's playing is that as soon as it attacks, the opponent loses in 30 seconds.

I think the reasons are as follows:
1. pluto is trained in self-play, so it imagined its human opponent to be itself, i.e. with perfect micro
2. pluto maintains some kind of "army value advantage" (let's call it AVA) estimation, to decide if it can beat or not beat the opponent army
3. the online pluto uses maphack, so it has a perfect information to calculate AVA
4. so when pluto commits to an attack, it is imagining that it can beat the perfectly microed opponent army
5. but human's micro is far worse than pluto, so human will lose a chunk of relative army value
6. pluto's AVA goes wayyy up, and goes into a frenzy and human dies


On the flip side, it's way too conservative as Zerg as a result. It thinks its Terran opponent can click hack and precisely target hurt Mutas, so it never harasses with Mutas. It always either runs away immediately or completely commits to a fight. It never tries to trade.


exactly! and that's how humans wind up beating Pluto by just making a huge army and never actually fight it :D
pluto will just hallucinate this impossibly strong opponent and in its immediate reward function to "reduce unit score loss" never take a fight

I've not seen Pluto take any hail mary attempts, because it is so caught up in optimizing the short-term gain. so Pluto will be very weak in actual positional plays, where you sacrifice the "now" for the "future". For this same reason it never figures out why upgrades are useful either.



Indeed. Despite how strong it is right now, it's beatable, especially if you abuse its blind spots (like not knowing what Dark Swarm is).

There's a lot of ways it can get even stronger. If people actually continue to work on this it's not hard to imagine a version that beats Pluto almost every time.

The creator mentioned publishing a paper on this research. I wonder if he'll be less interested in pursuing further research once he achieves that.


the author is very passionate about bw I think he'll write a paper but he'll also just keep on pushing this bot further I'm sure.

I think the most immediate thing to make pluto better (vs human) is to add a skill check mechanism so he figures out how strong (or weak really) human is at micro, and adjust its value functions accordingly
Life is run, it is dance, it is fast, passionate and BAM!, you dance and sing and booze while you can for now is the time and time is mine. Smile and laugh when still can for now is the time and soon you die!
StRyKeR
Profile Blog Joined January 2006
United States1756 Posts
September 27 2026 12:33 GMT
#53
The cost to run training must be pretty high I imagine. He may be limited by compute for diminishing returns unless he's being funded by Meta for its AI division.
Ars longa, vita brevis, principia aeturna.
evanthebouncy!
Profile Blog Joined June 2006
United States12816 Posts
September 27 2026 13:59 GMT
#54
On September 27 2026 21:33 StRyKeR wrote:
The cost to run training must be pretty high I imagine. He may be limited by compute for diminishing returns unless he's being funded by Meta for its AI division.


we can only guess haha :D

and yeah these trainings would be expensive for sure. The game space for starcraft is much larger than Go, and it'll take a good amount of compute to "flush out" that space. Pluto has done such an amazing job of flushing out the space, with ostensibly less compute than what deepmind had for playing Go. Even then pluto is yet to comprehend the orange cloud thing

how to train RL agents efficiently is very much an open problem, that's why it's still paper worthy
Life is run, it is dance, it is fast, passionate and BAM!, you dance and sing and booze while you can for now is the time and time is mine. Smile and laugh when still can for now is the time and soon you die!
ywxr
Profile Joined September 2026
4 Posts
September 27 2026 16:06 GMT
#55
On September 27 2026 16:08 evanthebouncy! wrote:
Show nested quote +
On September 27 2026 04:00 ywxr wrote:
I wonder if the creator is moving away from pure W/L-signal RL + strategy bandit to try to learn efficiently from these human games (which would be a uselessly small sample size otherwise), or from relatively small amounts of high-quality counterfactual-pair training data on things like drops and dark swarm. Given that it's already got a probably pretty well-calibrated win probability/value head, I think fine-tuning it to learn these things could probably be done very cheaply.


I think . . . the easy way to train it to learn about drop/dark-swarm is to find ways to synthetically generate a lot of data of games containing them, in the order of 10s of millions, in order to train pluto.
There is already some exploration intrinsic reward for making more niche units (that's why you see it upgrading ghosts or making a single corsair), except this incentive is not strong enough for it to climb the tech-tree all the way to dark swarm

A rather crude approach is to take some human games and train a kind of unit composition reward function, and use that reward function in its own game. For instance if defilers would give a high reward, the bot would learn to make them first, and then through sheer luck figure out how consume / dark-swarm works through millions of trials and error

on fine-tuning the winprob/value head, I'm not so sure. Clearly it is not calibrated against stuff like dark swarm and drop. I think those things do require significant amount of additional training data to converge


I agree about generating large amounts of synthetic data. I don't think the value head needs to be recalibrated directly; what I mean is that since it's very likely already quite generalizable to phenomena like "suddenly lose a fight very badly" or "suddenly lose a ton of economy," even if the network function doesn't understand well the underlying cause, with the right SFT architecture it should be possible to learn far more efficiently than millions of games.

Or more speculatively, even learn few-shot online by looking for sharp discontinuities in the value head which will be like high "perplexity" and then tracing back credit.
quaristice
Profile Joined February 2021
177 Posts
September 27 2026 16:10 GMT
#56
wonder if it would be faster to train something like this by save-stating moments in time and repeatedly repeating from that state trying different alternatives
evanthebouncy!
Profile Blog Joined June 2006
United States12816 Posts
September 28 2026 02:16 GMT
#57
On September 28 2026 01:06 ywxr wrote:
Show nested quote +
On September 27 2026 16:08 evanthebouncy! wrote:
On September 27 2026 04:00 ywxr wrote:
I wonder if the creator is moving away from pure W/L-signal RL + strategy bandit to try to learn efficiently from these human games (which would be a uselessly small sample size otherwise), or from relatively small amounts of high-quality counterfactual-pair training data on things like drops and dark swarm. Given that it's already got a probably pretty well-calibrated win probability/value head, I think fine-tuning it to learn these things could probably be done very cheaply.


I think . . . the easy way to train it to learn about drop/dark-swarm is to find ways to synthetically generate a lot of data of games containing them, in the order of 10s of millions, in order to train pluto.
There is already some exploration intrinsic reward for making more niche units (that's why you see it upgrading ghosts or making a single corsair), except this incentive is not strong enough for it to climb the tech-tree all the way to dark swarm

A rather crude approach is to take some human games and train a kind of unit composition reward function, and use that reward function in its own game. For instance if defilers would give a high reward, the bot would learn to make them first, and then through sheer luck figure out how consume / dark-swarm works through millions of trials and error

on fine-tuning the winprob/value head, I'm not so sure. Clearly it is not calibrated against stuff like dark swarm and drop. I think those things do require significant amount of additional training data to converge


I agree about generating large amounts of synthetic data. I don't think the value head needs to be recalibrated directly; what I mean is that since it's very likely already quite generalizable to phenomena like "suddenly lose a fight very badly" or "suddenly lose a ton of economy," even if the network function doesn't understand well the underlying cause, with the right SFT architecture it should be possible to learn far more efficiently than millions of games.

Or more speculatively, even learn few-shot online by looking for sharp discontinuities in the value head which will be like high "perplexity" and then tracing back credit.


I think you're right. Then I suppose what's missing is simply having the bot sees a ton of games with dark swarm. These things are black magic, only the creator can tell how many games does it take xdddd
Life is run, it is dance, it is fast, passionate and BAM!, you dance and sing and booze while you can for now is the time and time is mine. Smile and laugh when still can for now is the time and soon you die!
evanthebouncy!
Profile Blog Joined June 2006
United States12816 Posts
September 28 2026 02:19 GMT
#58
On September 28 2026 01:10 quaristice wrote:
wonder if it would be faster to train something like this by save-stating moments in time and repeatedly repeating from that state trying different alternatives


Maybe, so treating it like alpha Go and using monte carlo tree search. I think for micro counterfactual reasoning might not be important but for build order or reacting to builds it would be very important.

I'm unsure how it might be incorporated into the training though as the effects of build orders don't manifest for minutes down the line.
Life is run, it is dance, it is fast, passionate and BAM!, you dance and sing and booze while you can for now is the time and time is mine. Smile and laugh when still can for now is the time and soon you die!
evanthebouncy!
Profile Blog Joined June 2006
United States12816 Posts
October 03 2026 00:57 GMT
#59
we're doing a quick cast with namcraft on pluto (live now) if you want to tune in
https://www.youtube.com/live/J8nlGdR2kfU

it'll be pretty casual and I think it'll be recorded for later, when that happens I'll update the top level post
Life is run, it is dance, it is fast, passionate and BAM!, you dance and sing and booze while you can for now is the time and time is mine. Smile and laugh when still can for now is the time and soon you die!
Prev 1 2 3 All
Please log in or register to reply.
Live Events Refresh
Next event in 2h 17m
[ Submit Event ]
Live Streams
Refresh
StarCraft 2
ProTech152
SortOf 139
StarCraft: Brood War
Calm 2093
Bisu 402
Hyuk 271
ToSsGirL 212
Soulkey 140
Snow 34
ZergMaN 23
Noble 19
Larva 9
sOs 9
[ Show more ]
Force[name] 7
League of Legends
JimRising 825
Super Smash Bros
Mew2King71
Heroes of the Storm
Trikslyr45
Other Games
summit1g12328
WinterStarcraft616
ceh9303
C9.Mang0245
Happy232
ViBE60
Organizations
Other Games
gamesdonequick588
[ Show 10 non-featured ]
StarCraft 2
• AfreecaTV YouTube
• intothetv
• Kozan
• IndyKCrew
• Migwel
StarCraft: Brood War
• BSLYoutube
• STPLYoutube
• ZZZeroYoutube
Dota 2
• lizZardDota2218
League of Legends
• Rush1431
Upcoming Events
The PondCast
2h 17m
University League
2h 17m
LiuLi Cup
3h 17m
Rogue vs TBD
Maru vs TBD
OSC
8h 17m
Replay Cast
15h 17m
GSL
1d 3h
Maru vs Dark
Patches Events
1d 6h
OSC
1d 8h
Replay Cast
1d 15h
KCM Race Survival
2 days
[ Show More ]
Ultimate Battle
2 days
ZerO vs Bisu
RSL Revival
3 days
BSL
3 days
eOnzErG vs StRyKeR
Cross vs JDConan
RSL Revival
4 days
IPSL
4 days
Ret vs ggnore13
nOmaD vs Aether
BSL
4 days
Bonyth vs Doodle
Kyrie vs Jimin
Replay Cast
4 days
WardiTV Weekly
5 days
Kung Fu Cup
6 days
Solar vs Classic
ByuN vs SHIN
PiGosaur Cup
6 days
Liquipedia Results

Completed

Proleague 2026-11-16
Blizzard Classic Cup 2026
HCC Season 3

Ongoing

Acropolis #5
KCM Race Survival 2026 Season 4
International ShieldBattery League
BSL Season 23
IPSL Winter 2026-27
Calamity Cup Div. A Season 8
Underdog Cup 2026 Div 2-3
Stake Ranked Episode 6
BLAST Rivals Fall 2026
IEM Beijing 2026
PGL Masters Bucharest 2026
Stake Ranked Episode 5
1win Private Club #2
Thunderpick World Champ. '26
ESL Pro League Season 24
Stake Ranked Episode 4
1win Private Club #1
Logitech G Play Connect 2026
SL StarSeries Fall 2026
FISSURE Playground #3
BLAST Open Fall 2026

Upcoming

Sector 591
YSL S4
SC4ALL II: Brood War
NBSL 2026 Xiamen LAN
BSL 23: Non-Korean Championship
HSC XXX
Stellar Fest 2: Lunar Cup
StarCraft2 Weekly Open Tournament 2026 Grand Finals
SC4ALL II: StarCraft II
Kung Fu Cup 2026 Grand Finals
RSL Offline Finals
IEM Kraków 2027
PGL Bucharest 2027
BLAST Bounty Winter 2027
PGL Major Singapore 2026
TLPD

1. ByuN
2. TY
3. Dark
4. Solar
5. Stats
6. Nerchio
7. sOs
8. soO
9. INnoVation
10. Elazer
1. Rain
2. Flash
3. EffOrt
4. Last
5. Bisu
6. Soulkey
7. Mini
8. Sharp
Sidebar Settings...

Advertising | Privacy Policy | Terms Of Use | Contact Us

Original banner artwork: Jim Warren
The contents of this webpage are copyright © 2026 TLnet. All Rights Reserved.