Sora 2

Post by Oarch » Wed, Oct 01, 2025, 12:40 PM UTC

davidmurdoch wrote: post #589

I just asked GPT 5 to generate an image of as person. I then asked it to charge the color of their shirt. It refused because "I can’t generate that specific image because it violates our content policies." I then asked it to just regenerate the first image again using the same prompt. It replied "I know this has been frustrating. You’ve been really clear about what you want, and it feels like I’m blocking you for no…

Post by andsoitis » Wed, Oct 01, 2025, 12:42 PM UTC

the_duke wrote: post #273

I haven't seen comments regarding a big factor here: It seems like OpenAI is trying to turn Sora into a social network - TikTok but AI. The webapp is heavily geared towards consumption, with a feed as the entry point, liking and commenting for posts, and user profiles having a prominent role. The creation aspect seems about as important as on Instagram, TikTok etc - easily available, but not the primary focus. Genera…

Post by mscbuck » Wed, Oct 01, 2025, 12:46 PM UTC

bonoboTP wrote: post #578

afavour wrote: post #555

Earlier quoted context omitted.

This was my reaction when I saw Meta’s “Vibes” app. Who wants to browse a stream of exclusively AI generated videos? Obviously Meta wants that because it’s a lot cheaper than actually paying real people to make content… but it’s slop.

This is not the final target. It's video generation now, but that's just a stepping stone. The real thing is that learning a generator is also learning a prior over videos, and hence over how the world works. The real application of this will be word models, vision-language action models, spatial AI and robotics. Basically a kind of learned simulator in which to plan and imagine possible futures, possible actions and…

Post by thegeomaster » Wed, Oct 01, 2025, 12:56 PM UTC

dwohnitmok wrote: post #688

some-guy wrote: post #452

Earlier quoted context omitted.

My opinion is that unless there is some insane breakthrough in power efficiency with video generation, or if energy costs go down to zero, there is no way such a thing actually becomes profitable at the scale of scrolling TikTok. It is far more power efficient (and cheaper) to have people post their own content.

I think you're overestimating how much power LLMs consume. Let's say one video pegs a top of the line Blackwell chip at 100% utilization for 10 minutes. I think a Blackwell chip (plus cooling and other data center overhead) is somewhere around 3000 watts when running 100%. So that's about 0.5 kilowatt-hours. I suspect this is a severe overestimate because there's probably a significant amount of batching that happens…

Post by failingforward » Wed, Oct 01, 2025, 1:03 PM UTC

raincole wrote: post #560

wrigby wrote: post #553

Earlier quoted context omitted.

There are two completely distinct differences that jump out to me initially that I think may help justify your feelings: 1: Reading a long book demands focus on a longer timespan than scrolling TikTok, and with focusing on a single thing for a long time, we get a sense of accomplishment. I don’t know how to justify this as valuable, but for some reason I feel that it is. 2: The Song of Ice and Fire (and GoT) were con…

I think the point 2 will rub many people the wrong way (me included) though. That would make reading Fourth Wing or Twilight a more connecting experience than most classics. (Nothing inherently wrong with that, but... you know...)