Making Sense of LLMs - Where's the value at/add?

This is the second installment in a series about LLMs. You can find the first article here:?Making Sense of Large Language Models - An Introduction.

Following the news about LLMs has been exhilarating lately. Hardly a week passes without some AI model mastering a new feat that would have sounded unbelievable just a?few months ago. As they say, any sufficiently advanced technology is indistinguishable from magic. But magic has nothing on economics, and the cold, hard reality of the market will kick in sooner or later. Mostly sooner. In other words, someone will have to pay for our magical AI toys, and we have yet to figure out who it is going to be. Spoiler, it will be the consumers, but the details are still TBD.

Let's back up a little. Before the internet revolution, the way to win as a company selling physical goods was by controlling the creation and distribution of those goods. Think newspapers with iconic staff writers and trucks hauling paper nationwide. All the glory and power of the?fourth estate.

Digitalization threw a big monkey wrench into the gears of this machinery. Not only is online publishing essentially free of charge, it also reaches a global audience faster than any printing press can start rolling. Dominating the creation and distribution of goods was like a water moat that defended one's business. The internet power-plowed this ditch in, with abundant content and zero marginal cost of distribution.

If you can get everything instantaneously at the press of a button, your biggest problem becomes knowing what you want and finding it. Companies like Google and Facebook quickly jumped into this new world. They became the field guides to our blossoming online lives. And by guiding our attention and converting it into advertisement, they created trillion-dollar industries (for more on the topic, see Ben Thompson's?Aggregation Theory).

Chatting with the newest LLMs gives the impression that we are witnessing the creation of another trillion-dollar industry. And investors are desperate to figure out what shovels to sell during this algorithmic gold rush: "There are, of course, the standard moats:?

• Scale moats ('I have or can raise more money than you!')
• Supply-chain moats ('I have the GPUs, you don't!')
• Ecosystem moats ('Everyone uses my software already!')
• Algorithmic moats ('We're more clever than you!')
• Distribution moats ('I already have a sales team and more customers than you!') and?
• Data pipeline moats ('I've crawled more of the internet than you!').?

But none of these moats tend to be durable over the long term" (Who Owns the Generative AI Platform?). For now, cloud infrastructure providers and AI chipmakers have the firmest footing in the market.?

As exciting as money-making is, the implications of the AI revolution are far more significant than who can charge for what. The cost of digital content creation across all domains will fall dramatically and ultimately converge toward the electricity costs needed to run LLMs. If the last sentence does not both excite and scare you, read it again.

The legal world still tries to shoehorn the square building blocks of digital rights management and intellectual property through the round holes of the Civil war era legislature. Without going into details, making sense of algorithms that Frankenstein together all of the world's text, imagery, videos, and music will be an?uphill battle. A mad scramble across the board, and we haven't even touched the topics of monopolies and regulation, which I will tackle in an upcoming blog.?

Right now, nobody knows how it will shake out. But we can assume a few core tenets. First, digital content will become radically cheaper and easier to create.?

Second, we will be hit by a tsunami of content. This content will often be subpar initially, but the overall tide of improvement will lift even the smallest vessels of fringe taste from the sandy shores of mediocrity. If you can name it, it can be willed into existence. Everyone with access to the internet will yield magical powers.?

Third, the lines between genuine human creation and robotic imagination will get increasingly blurry and hard to delineate. Think tentpole movie-grade CGI but not just for Hollywood productions but for anything. Like vinyl, there will be a newfound interest in the artisan and the "real." Most consumers, though, will move on and not look back.

Fourth, text prompts are clumsy interfaces and only the first-order approximation of an engineer's idea for engaging consumers. Be prepared to interact increasingly with LLMs hiding behind robots, art, toys, or industrial machinery.

My predictions are sufficiently vague to hopefully stand the test of hindsight's 20/20 vision. If nothing else, I hope this blog makes you want to get involved yourself. Big innovations are a force of?creative destruction. They are also equalizers as we all start anew. Or, as Sheryl Sandberg?said, "if you're offered a seat on a rocket ship, don't ask what seat! Just get on."

This is the second installment in a series about LLMs. You can find the third article here: Making Sense of LLMs - Data gets to be the new oil again!