I’ve been thinking a lot about the different audiences and use cases for LLMs, and the various creative endeavours that it’s seeking to change.
From the POV of the billionaires at the top of these companies, actual creative professionals are a rounding error. A loud one, at times, but not one to overly concern themselves with once the media has wandered off and all you’re trying to do is make your first trillion.
Currently, very few of the AI products are making a profit. Most are haemorrhaging cash at an impressive rate. Obviously, this isn’t a business plan that can go on indefinitely. So it won’t. While there’s lots of talk about a bubble, my expectation is that most will simply adapt and survive. I’m already hearing of cancellations of some giant data centres which were going to hoover up colossal amounts more money, as LLMs evolve into more streamlined versions. Smaller versions, SLMs, perhaps. And expect to see a lot more of companies using S/LLMs in concert with other AI structures, acting as collections of systems fronted by a single AI agent to interface with the wetware.
I see “creative” AI being used in two main ways now, and this will continue. The first is a minority approach, but possibly a modestly lucrative one. This is with creative professionals using it as part of their pipeline. It’s not a complete replacement for animation or code departments. However, it can do some of the basic stuff and with skilled professional (humans) to check, filter, and adapt the output, it’ll get better over time. I don’t see the need for human oversight going away soon (outside the AI company’s own advertising). The problem of this disenfranchising the new folk in each field is something that will be dealt with by companies either drinking the AI Kool Aid and relying entirely on AI as it improves, or by training their own cadre internally. We’ll see both.
The big chunk of users now, and probably indefinitely, will be unskilled and uncreative people who’d like to be more so and who lack the time or the inclination to actually learn the skill. Humans have always wanted a short cut, and we see this in every other field. AI is nothing new in that regard. Personally, I see this as an extension of prolefeed or soma. Mixing my SF authors here. It’s something to keep folk occupied. It’s not a way to create new masterworks. Not in this use case.
The big question is how this will be monetised, as the bubble can’t be sustained indefinitely. My guess is that they’ll attack both ends of the puzzle by reducing costs as iterations increase efficiency, and raise income by adding AI to ever more unnecessary situations and getting people to pay subscriptions for it. You’ll have seen this all over the place.
Of course, the bubble may pop before they balance the books. If that happens, the tech will just change hands, the debts will be written off, and a new group of AI proponents will take the helms of the next wave of AI companies. It’s not going away.
Personally, I’ve done enough pondering on this topic now. As I said at the start of this series of rambles, I wasn’t sure how I thought about the whole thing, and writing these articles and reading your comments would help me frame it more clearly in my mind. Well, I’ve done that now, so next week will be my conclusion. At least for the time being. Then you can look forward to me getting back to pontificating about games and art and writing. I bet you can’t wait.
Quirkworthy 
LLM are a dead end since the scale very badly and the way they are trained makes them dumber and dumber over time. The data centers can be used for different use cases if they are not too specialized. Apple for a reason has shelved its AI approach and licenses Google`s AI. They can´t and do not want to compete with the current LLM AI and rather spend their money for the long term haul (ARM anyone? They bought the license in the 90s and worked on it until it was great).
The future road will be specialized algos – those we are already using for a long time in creative work and AI that are closer modeled after the human brain.
I agree that pure LLMs have a poor future, though as part of a cluster of AI modules that a central agent can access to farm out the appropriate tasks to I can see them being used for some time.
My expectation is that the big AI companies have the next couple of steps in dev already. At least, with more money than the GDP of a small country, you’d imagine they would.
My guess is that at least OpenAI has less of a plan and that they are gambling big. Apple seems to be the company with the best understanding what really is needed for a good AI. And quite some companies are in there for the me-too.
The me-too is a major chunk of all this. They’ve convinced themselves that a godlike AI is just around the corner, and that will cure cancer and warp travel so that the revenue fro it will also be godlike. No company wants to miss out on that imagined future.
I think this daydream is going to go the way of flying cars: the future we were promised, but not the one we got.
Regardless of current profitability, AI is here to stay, will kill many jobs and enrich and empower the rich & powerful even further. The company I work for is firing 10% of it’s local staff thanks to AI & ongoing digital transformation. It will only get worse. AI can work miracles but is only used for slop & greed.
Which goes to my central thought on the whole topic.
AI is here and is doing stuff we might rather it didn’t, but which we cannot prevent. That being the case, how do we, as *actual* creatives, change how we look at the world, and how do we change the way we create?
My own answer to these questions is the subject of next monday’s post.
Another avenue for profitability for generative AI is to integrate advertisements into the product. This might be easier for text, though for images, this could be adding to the prompt to get product placement in backgrounds of images.
Of course, Chatgpt is already planning on adding advertisements into it’s model to make up for that profit loss. https://www.cnn.com/2026/01/16/tech/chatgpt-ads-openai
Mark’s comment made me laugh. How long will it be before we see someone using an AI generated image for box art without noticing that their hero character is promoting Pepsi?