OpenAI's sudden discontinuation of its video model #Sora, just a few months after a spectacular deal with Disney, shows just how little these AI companies care about their so-called products. AI video apps, to quote @hito Steyerl, are «onboarding tools» — their purpose is more ideological than commercial.

Rather than offering revenue-generating services (OpenAI is burning money with every single Sora generated clip), these apps are designed to accustom us to a world in which so-called generative AI seems inevitable — and to provide investors with impressive demonstrations of these models' performance.

In other words, companies like OpenAI do not develop consumer products; they merely launch demos that can be retracted once they have fulfilled their purpose. Never mind that, in the meantime, they are flooding our media landscape with a deluge of #slop and destroying entire cultural sectors in the process.

@bildoperationen few thoughts. 1.we are past "onboarding" .the military, is on board; the boat is afloat, AI bubble not bursting bc war economy. 2. what this means for image generators is unclear yet. one cautious guess: one part (slopaganda etc) will just get more normalised. about another part (small custom made models) we just dont know yet.
@hito Thank you! Totally agree, although I'd see «onboarding» as a continuous process. Academia is still in the middle of it, and although the war machine is the big prize, the tech companies are also highly interested in education and universities, because it easily scales and provides secure revenue once the institutions are on board (and they are eager to get on board)
@bildoperationen absolutely. they will just have to do it without meme gadgets this time.
@hito That might be a main reason OpenAI discontinues Sora and focuses on ChatGPT - AI video generation ist not only extremely expensive, it's also hard to sell as anything else than a gadget, at least when you're aiming at government contracts (first, they tried to promote it as decisive step to a «world model», but I never heard about that one again)
@bildoperationen if you look at fine print it’s there: Sora only used now to generate training data (——“world model”, simulated drone flights, “physics”)
@hito That'd be interesting, do you have any sources on how this might continue and/or is already in use? I always thought it was more of a unfulfilled promise, given all the physical inconsistencies and instabilities of AI video