Conversation

Notices

1. Ars Technica (arstechnica@mastodon.social)'s status on Thursday, 29-Aug-2024 02:06:38 JST Ars Technica

   New AI model can hallucinate a game of 1993’s Doom in real time

   "Why write rules for software by hand when AI can just think every pixel for you?"

   https://arstechnica.com/information-technology/2024/08/new-ai-model-can-hallucinate-a-game-of-1993s-doom-in-real-time/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social

   • nick splendorr ✨🏳️‍⚧️ (splendorr@mastodon.social)'s status on Thursday, 29-Aug-2024 02:15:16 JST nick splendorr ✨🏳️‍⚧️

     in reply to

     @arstechnica starting out with "diffusion can produce an entire game" and ending with "it only generates 3 seconds of gameplay after watching infinite footage of an existing game that only fools people 40% of the time" including nividia's marketing assertion that in 5 to 10 years ALL game graphics will be generated this way... is such an enormous stretch that it verges on journalistic malpractice :)

   • Jonathan Hartley (tartley@mastodon.social)'s status on Thursday, 29-Aug-2024 03:47:27 JST Jonathan Hartley

     in reply to

     @arstechnica it can do that because it's been trained on a million videos of Doom. Now try getting it to generate a game that isn't already written. Useless.

   • neonharbinger (neonharbinger@mastodon.social)'s status on Thursday, 29-Aug-2024 04:10:20 JST neonharbinger

     in reply to

     @arstechnica The more times you read this...

     "The researchers say that ten human raters sometimes failed to distinguish between short clips...of actual Doom game footage and outputs generated by GameNGen, identifying the true gameplay footage 58 percent or 60 percent of the time."

     ...the funnier it gets.

     GameNGen had one job, really, to COPY, and it couldn't even do that successfully. It failed at an open-book test.