Live data from Hacker News

SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

deepmind.google

81–90 of 115 posts

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#81
post #24

Earlier quoted context omitted.

To be fair, it's 65% on all tasks (with a 75% human baseline) and 15% on unseen environments. They don't provide a human baseline for that, but I'd imagine it's much more than 15%.

It really feels like we are determined to simulate every possible task in every possible environment instead of building true intelligence.

To be fair, humans are also pretty terrible at doing completely new things and until we’ve tried it a few times.

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#82
post #71

Tangential question: is this going to eventually ruin e-sports? What happens when a team of humans are playing against a team of AI, which play in the same conditions, with network lag etc. from a client computer perspective... And consistently beat the human counterparts for being faster at response time, never make mistakes and not ever getting tired? Eventually it could kill all MMOs, fill them up with AI players,…

If/When AI gets to the point where it can consistenly beat human players in video games, it will be worldwide news when it first happens, and then the novelty will wear off. See: Chess still going strong.

For the video game market in general, AI will be used by players largely where there are financial incentives (MMO market, CS:GO Skins, etc.), but usage of an AI will most often be socially viewed just like usage of any other performance improvement/script/hack: As a crutch.

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#83

OK, AI playing video games is cool. But you know what's really really cool? It looks like SIMA 2 is controlling the mouse and reading the screen at something approaching 30+fps. WANT. Computer use agents are so slow right now, this is really something. I wonder what the architecture is for this.

Python dxcam + windows hook API listening for HID messages

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#84
This gives me strong vibes of "The Lifecycle of the Software Objects" [1] by Ted Chiang. Next step is to put this digient AI into a Figure 03 robot [2].

[1] https://en.wikipedia.org/wiki/The_Lifecycle_of_Software_Obje...

[2] https://www.figure.ai/news/introducing-figure-03

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#85
post #22

I hope we can get some (ideally local) version of this we can use as a "gaming minion". There's a lot of games where I probably would have played more if I could delegate the grind. If they're not that competent, it adds to the fun a little even.

Skeeball -> Virtual Skeeball -> Virtual virtual Skeeball

"Wow! I could've sworn I was really playing virtual Skeeball!"

--

In practice I think you would lose interest very quickly in any game should you do so. Games are carefully designed to balance the drudgery with action, and to control the complexity progression. Using AI would break both.

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#86

OK, AI playing video games is cool. But you know what's really really cool? It looks like SIMA 2 is controlling the mouse and reading the screen at something approaching 30+fps. WANT. Computer use agents are so slow right now, this is really something. I wonder what the architecture is for this.

Its even cooler if humans find something to be excited about in this world, since AI is replacing everything we do.

You do realise the things that AI are replacing are very new things humans have only been doing for the last ~50 years?

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#87

I want a smarter game. Like a survival game that - as usual - starts with you collecting sticks and stone to build a stone axe. But at the appropriate tech level, transitions into automation. You discovered a new building material and want to build a castle from it? Equip your NPC's with diamond pickaxes and tell them how much better/safer life would be if they built a new castle from unobtanium . And off they go. To…

You might be interested in https://github.com/mindcraft-bots/mindcraft

It's an open-source framework for building intelligent, task-driven bots in Minecraft Java Edition powered by large language models (LLMs) such as GPT-4, Gemini, and many others. Designed for research and creative automation, these bots can connect to Minecraft worlds, perceive and act on their environment, undertake custom tasks like resource gathering or building, and even collaborate in multi-agent scenarios.

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#88
post #71

Tangential question: is this going to eventually ruin e-sports? What happens when a team of humans are playing against a team of AI, which play in the same conditions, with network lag etc. from a client computer perspective... And consistently beat the human counterparts for being faster at response time, never make mistakes and not ever getting tired? Eventually it could kill all MMOs, fill them up with AI players,…

there was a guy who did relatively simple ML with RL on a druid in world of warcraft and put it into 2v2 arenas. soon enough it was completely unbeatable regardless of what human it got paired with. a 1 man army lol XD. they guy did pick the right class for it, but still. AI players done right with RL become insanely difficult to beat.

games like CS its less useful because it will be blatantly aimbotting. as it gets better it will be more and more obvious. you might be able to train it to mimick human mistakes, but i think ultimately it will be easily spotted by other players. for games without the hand-eye-coordination like turn-based games or games with 'global cooldowns' , mmorpgs etc., it will be much harder to identify.

i think normal subtle cheats like ESP when 'done right' are much more killing esports than this would.

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#89
post #56

Earlier quoted context omitted.

My friend that is AI. However, it can get a lot better: be more aware of screen content, follow multiple instructions at once, keep context in mind throughout the conversation and from past interactions

AI commonly means LLM. Where are you determining this is using a LLM for proccessing?

You must be new around here, kid. AI has been there since the birth of the programmable computer.

Re: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

#90
post #86

Earlier quoted context omitted.

Its even cooler if humans find something to be excited about in this world, since AI is replacing everything we do.

You do realise the things that AI are replacing are very new things humans have only been doing for the last ~50 years?

You do realize those very new things have mostly replaced what humans used to do ~50 years ago?
Post reply on HN