Finding the missing AI startups
The novelty of AI is undeniable. Semantic processing of information has always been where computers didn’t just fall short, but utterly floundered, and now they don’t.
This raises the simple question: Where is the profit? All the major players in the field seem to be burning cash, and since this time around, it’s not just building, but actually using the stuff that costs actual money, it’s not clear whether they will be able to simply flip the big profit lever once adoption is wide enough.
A few explanations satisfy: Perhaps it’s not yet found - it is after all hard to really work the consequences of a technology from seeing it. Perhaps it’s hard to take advantage of, since doing so requires rearranging a great number of things around the new technology.
Maybe so. I believe there is another: They arrived before the technology. Consider two major contenders for making the computer talk smart: Meta and Google. Google famously invented the transformer, and insisted that they essentially had the tech behind ChatGPT in house but simply didn’t know how it would hit. I don’t doubt them - they produced the first case of modern AI psychosis in Blake Lemoine.
They had this technology not on a lark, but because it was useful to them. Not in its current form, and not in isolation, Google, Facebook, Instagram and YouTube all work by algorithmically matching a person and pieces of semantic content which might appeal to them, be it websites, posts, or video. They work with moderation, hugely automated.
Influenced by the “cultural technology” view, I think of LLMs essentially as maps, in the sense that they depict something real - culture, the internet, the world - not wholly, but representing them, but also maps as in video game maps, spaces people explore and play in. Spaces where, sometimes, someone falls down a deep, dark hole. How is YouTube any different?
Often, ramshackle guides are given, in the forms of the porous, shifting, ramshackle personas the models are given. Their agendas are doubly unclear - one, because the technology is little understood, and second, because the goals of the companies trying to steer you through them are, similarly, obscured. YouTube optimized for watchtime and pulled people down extremist rabbit holes. Facebook optimized for engagement, and, by proxy (though not unknowingly), for anger.
The latter requires the former. You cannot guide people to further video’s they’ll watch, posts they’ll read, and so on if you cannot tell which pieces are alike in what ways. I’m not sure what the state of the “guides” at LLM companies is, by the way. Neither do you, and perhaps, neither do they. But there we have it: Why aren’t there more AI startups making serious money? It’s because the low hanging fruit is already gone.