Almost everyone in the comments misunderstands what is happening here, right? To the best of my knowledge, this is an AI video generated in real time, in the same vein as an LLM.
Please correct me if I’m wrong, but its essentially predicting the most likely next frame of the screen based on “learning” from watching videos of the game. So if the videos have people push the jump button, the most likely next frame is that the character jumps. If the most likely next frames from the dig action being taken is the block removed and the item appearing in the inventory bar then that is what would occur. But there’s nothing here actually “happening”, this can never be a game - that is, everything that makes a game a game is the ability to succeed. There is no health, experience, etc. It is total frothy slop, and just a surface level moving photo of the game.
It looks that way to me, less “AI generated gameplay” and more “AI generated minecraft let’s play”
For actual AI gameplay, this video was interesting to meI found a YouTube link in your comment. Here are links to the same video on alternative frontends that protect your privacy:
If the most likely next frames from the dig action being taken is the block removed and the item appearing in the inventory bar then that is what would occur
You’ll notice it’s not actually got as far as the item appearing in the inventory bar - it doesn’t manage to pick up any blocks in the video, and when placing blocks the item counter just jumps around randomly.
I suspect, like with the Doom video BelieveRevolt mentioned, it can only make about 3 seconds of footage at a time too, which is the reason it keeps jumping to new scenes.
You’ve got it right. There was already a video of someone “playing” Doom and it was mess with objects and enemies not reacting to weapon fire, randomly disappearing, health obviously not being tracked etc.
This was fucking Doom, a 30 year old game that will run on anything, and they were trying to brag about the energy guzzling treat printer making a horribly mangled version of it.
AFAIK the premise is basically “given the last frame and the current input, what is the next frame?” so there’s no object permanence or structure behind it it’s just making a video that for any given span of two consecutive frames looks like a video of minecraft.
A Reddit link was detected in your post. Here are links to the same location on alternative frontends that protect your privacy.
