ZENOS

The data foundry for frontier AI.
Train on worlds, not just pixels.

[ZENOS @zenosdata]

The new gold standard for training AI on game data.

ZENOS Standard 1.0: seven render buffers, split audio, every input and what it meant, full world state, labels, and every pixel resolved to its entity.

One clock. No SDK. Licensed from the studios.

We went /deeper.

[ZENOS @zenosdata]

Download a sample with every layer, a readme and a notebook, under an evaluation licence. https://t.co/OgI4x2qjw1

[ZENOS @zenosdata]

The new gold standard for training AI on game data.

ZENOS Standard 1.0: seven render buffers, split audio, every input and what it meant, full world state, labels, and every pixel resolved to its entity.

One clock. No SDK. Licensed from the studios.

We went /deeper.

[ZENOS @zenosdata]

How we capture it: PRISM reads the shipped build while people play.

No SDK, no dev build, no source code, no mods, any engine.

Every title is licensed from its developer and cleared for AI training: world models, agents, robotics, video models etc.

[ZENOS @zenosdata]

How we got here: https://t.co/d2h5RjspnP

[ZENOS @zenosdata]

Six years, one technology.

Since 2020 we have read the full state of running games, at first to rebuild the entire game somewhere else. We have used it for VR spectating of esports matches, real-time recreations of games for broadcast and large venues, and 3D views of matches in the browser.

Every one of those ran as a third-party app, outside the game, with no special access from the developer.

AI companies want to go deeper and deeper on games and gameplay, so our data became the product: the full state behind every frame of a shipped game, licensed from the studios, for AI teams who want to go as deep as possible.

Next week we show what the capture reads.

[ZENOS @zenosdata]

@TimSweeneyEpic @tomjank If developers want to reserve their rights, i.e. control who can train their AI on their games, their own EULA and terms need to say so. Here is how: https://t.co/nM3HK63p7N

[ZENOS @zenosdata]

@Pirat_Nation Now game developers just need to make sure their own EULA and terms reserve their rights on AI training, so they control who trains models on their games. Here is how: https://t.co/nM3HK63p7N

[ZENOS @zenosdata]

@gamedevdotcom Now game developers just need to make sure their own EULA actually reserves their rights on AI training, so they control who trains models on their games.

Here is a free guide on how studios can implement appropriate rights reservations: https://t.co/nM3HK63p7N