OpenAI just closed down Sora, one of those AI slop video generation tools. AI video generation is relative resource intensive, and OpenAI lost huge amounts of money on it, while receiving very little revenue. Now they could have kept Sora, and just raised the price to at least break even. But that price would have been eye-watering, and would have destroyed the myth that AI is a cheap way to replace humans. So the first threat to the AI slop business model is that cheap or free image/video generation is on its way out.
But the bigger problem is probably on the revenue generation side. If you think of for example the monetization of free products like the Google search engine, it works because your input of a question into Google has a value. If you are searching for the best headphones, chances are that you would be willing to spend money on some headphones; Google is able to either sell that information to the right people, or even more bluntly steer you towards the headphones somebody paid them to advertise. The monetization of social media like YouTube works by assuming engagement: If I click on a video with a review for a board game, I am probably interested in that board game, and at least engaged in the board game hobby. Again, that can be used by YouTube to serve me advertising on whatever interests me, increasing the chance that I click on the advertisement and buy stuff.
OpenAI is not getting any valuable or sellable information by somebody using Sora to generate an AI slop video. And on YouTube that video might get a lot of clicks, but the engagement is very shallow. What advertising do you serve somebody clicking on Shrimp Jesus? Sea food? Chances are, they are more grossed out by the images than interested in any product.
Ultimately the whole AI slop business chain consists of a number of steps, none of which is actually creating any real value. The people using the AI image/video tools aren't buying anything and aren't willing to pay much for their low effort content creation business. The people clicking on those images and videos aren't engaged, and create just empty clicks of no value to advertisers. The real cost to generate the AI slop by far exceeds the economic value created by it. The people paying that cost are the investors in the AI companies and the advertising companies, and both of these will one day realize that there is no pot of gold at the end of the rainbow. Shrimp Jesus has feet of clay, economically speaking.
Id say the exact same thing applies to the text based LLMs. My wife uses chatGPT like a google replacement. If ChatGPT started charging for its use she would just go back to Googling. So how do they ever make any money off her? They wont. I guess they could introduce ads but the minute its more convenient to just Google people will go back to doing that.
ReplyDeleteThat goes for the vast majority of their users. Sure they might get some folks who will pay for access just like those who sign up for their premium features now but is it enough to sustain the cost and the enormous amount of money being invested in the industry? I doubt it.
The LLMs can be pretty effective for search.
ReplyDeleteThere is no enshittification. Cory Doctorow is a hack without formal education, dropping high school for an alternative high school and then quitting four universities.
ReplyDeleteHe described a small fraction of a process that doesn't exist in a vacuum without mentioning the larger context.
Then why is it still popular? Because causation and correlation.
It's easily observable that platforms change to the detriment of the user and the platform being "greedy" feeds perfectly into the outrage of the user.
"They want more money!" is understandable even by the dumbest person and won't prompt questions as to how or why. On top it positions the user as the rightful audience that is being disadvantaged.
Platform degradation does happen, but it happens due to:
- platform growth (databases, servers, processes, etc. need to be scaled - a cheap server costs nothing, running a data center thousands)
- competition from other platforms (anyone remember Myspace? Facebook bought Instagram as a Snapchat competitor; studios launched their own streaming services after Netflix)
- normal market forces (advertisers have dozens of other competing options) and users never paid for the service to begin with
- markets correcting the evaluation of the product by suppliers (MGM et al. likely sold movies at a penny of a dollar to Netflix before realising the value of streaming)
- normal social effects (Facebook is less cool when your parents comment on your posts)
Here is a real life example:
You and your three friends go to your local pub and have a great time. Drinks are cheap, service is fast as you are one of the few patrons, your favourite niche sport is on, the local brewery has a poster in the corner, you can play Darts, there is enough room.
A year later, the pub is overflowing, they had to build two extensions, service is unable to accommodate everyone on time, your sport is no longer on but generic football, rival supporters are regularly shouting at each other, a big named brewery has a banner on the wall, Darts had to be removed for safety reasons, prices have gone up.
Who is to blame? "Obviously" the owner!
Full due to increasing popularity? Nope. Changed channel due to different audience? Nope. Better supply from a larger brewery? Nope. Safety regulations banning darts? Nope. The economy taking a dive and raising prices all around? Nope.
Anyway, with that out of the way:
Yes, venture capital backed companies don't have to care about profitability so they don't. Once established, they can take care of that - and a lot will never get to that stage and fail before.
The difference is that digital products can be in early access since you are not building a physical product.
Yes, AI companies are bleeding money, are unsustainable, the output is utter crap and I hope they die in a fire - but is their goal to be profitable right now? In the long run? Or do they first need to switch consumer perception similar to the touchscreen?