It was using cairs too! I know it feels really clunky when you’re watching it but holy shit was it ripping through Larva and Soma. I get humans can’t micro like Pluto can but the early goon/cair/slow zeal builds seemed to nullify Zergs having free rein over the early game.

What’s the harm in experimenting?

They did. Many years ago. It just didn't work on a human level.

To defend 973 with only a few cannons + goons zealots, not having a wall is actually better, but it means you can't defend the earlier ling flood. AI can defend that because its probe pull is... well, inhuman. Not to mention that Zerg can go 2H Muta, (which is what Larva seems to be doing every game now because it's the only way to abuse Pluto's weakness), and once again, defending that with goons is inhuman, because the AI literally targets the weakest muta in a stack one by one.

fLyiNgDroNe

Belgium4129 Posts

9 hours ago

#482

On September 28 2026 20:36 LUCKY_NOOB wrote:

Show nested quote +

Not a single new idea.
You can read posts from random D ranks going back as far as the game release saying things like: why dont u use ghosts?

The execution is new but also superhuman...

Its not like pros dont know ghosts exist.

Of course if it inspires humans to be better in some way that would be great.

Its not a TAS in Super Mario Bros.

Thats a very lazy take on this. New idea =/= using underused unit.

LUCKY_NOOB

Bulgaria1593 Posts

9 hours ago

#483

On September 28 2026 21:25 fLyiNgDroNe wrote:

Show nested quote +

Thats a very lazy take on this. New idea =/= using underused unit.

Sure its not the same. But ideally it has to be limited to the human level apm equivalent [whatever that may be] to be relevant to human play. Now it's abusing it's machine advantage to make up for shortcomings like awful building placement.

More research is needed.

TMNT

3372 Posts

9 hours ago

#484

Larva finally beat it (6-5, after losing the Bo9 4-5).

But yeah I think namkraft's assessment in the previous page is correct. After a while this human vs AI battle may be worn out because the game becomes either:
- straight up, honorable macro game where the AI just smashes human with its incredible control
- human abusing its lack of understanding in certain aspects (lack of defending mineral line, drops, cheese, lack of expansion, doesn't rebuild buildings...), it still takes a lot of effort from the human though. I mean Larva still has to work his ass off despite abusing every trick in the book to punish the AI.

Once/if the AI learns about drops and defending backstabs, it's probably impossible to beat it.

iFU.pauline

France1768 Posts

8 hours ago

#485

On September 28 2026 21:45 TMNT wrote:
Larva finally beat it (6-5, after losing the Bo9 4-5).

But yeah I think namkraft's assessment in the previous page is correct. After a while this human vs AI battle may be worn out because the game becomes either:
- straight up, honorable macro game where the AI just smashes human with its incredible control
- human abusing its lack of understanding in certain aspects (lack of defending mineral line, drops, cheese, lack of expansion, doesn't rebuild buildings...), it still takes a lot of effort from the human though. I mean Larva still has to work his ass off despite abusing every trick in the book to punish the AI.

Once/if the AI learns about drops and defending backstabs, it's probably impossible to beat it.

Yeah the AI doesn't "understand" to begin with. It is interesting to see however what it takes to an AI to rivalise and win vs top human in terms of energy and mechanical advantages and it's quite huge.
It took 3 to 4 days for a human to adapt and kill the AI. The energy that needs to be spent now for a better AI (handling abuse which is far from trivial I believe) might be exponentially bigger. It shows how cost-effective we are and there is no comparison to make. The fact that there is so much data for a model to train on gives a false impression. It is too much energy for little in fact.