Today's AI is the worst you'll ever use...

It's important to remember that AI tech is rapidly improving. What you see today will be eclipsed by tomorrow's tech.

A common, and pithy, statement about AI technology is “Today’s AI is the worst you’ll ever use.” The statement implies that AI technology is rapidly improving, and that you shouldn’t assume that what you’re using today will be as powerful as what you’re using tomorrow, or next week, or next year. There is a certain truth captured in the statement. And yet, as I explained in my post about exponential growth, a lot of people don’t have an intuitive grasp of the rate at which this technology is improving.

I’ll try to illustrate this with a concrete example. Consider the following video:

If you watch the video, what you will see is a poorly-edited pastiche of steampunk and fantasy. It looks, well, artificial. The short film provides no coherent narrative, no reliable exposition. The description on YouTube suggests that it is “inspired” by the video game Frostpunk, though I couldn’t tell you whether this is accurate. It uses a combination of technologies from Midjourney, Pika, and Stable Diffusion.

When I shared this video on Facebook, someone commented, accurately, that the video needs a good editor. “Can AI mimic the intuition of a good human editor,” this person asked, very reasonably. The plain answer is, no, an AI cannot presently mimic the intuition of a good human video editor.

But in some sense that misses the point. The point is not what AI can do today. It’s: What does the future portend for AI capabilities, give their trajectory? That’s a much different question, and it leads to a much different answer. The vision commonly bruited about is this: you sit in front of your computer, enter a few prompts, and an AI spits out a customized, Hollywood-quality thriller for you to watch in the comfort of your home. We’re far from that. You can compare this AI mediocrity to the trailer for the post-apocalyptic steampunk movie Snowpiercer:

There is something ineffably human about this trailer, even though the world it depicts is entirely fantastical. There is a future version of AI generated movies, better than the one which you use today, which will let you generate a Hollywood-quality movie from a few prompts.

The present state of AI technology is this: it is still artificial. It is not human, in the way that human creations are.

Where is this all going? Here’s part of an interesting tweet from Andrej Karpathy:

You know how image generation went from blurry 32x32 texture patches to high-resolution images that are difficult to distinguish from real in roughly a snap of a finger? The same is now happening along the time axis (extending to video) and the repercussions boggle the mind just a bit. Every human becomes a director of multi-modal dreams, like the architect in Inception.

Consider what happens once we are able to reliably generate decent movies on demand. First, we’ll have a surplus of movies made, most of which will be pretty crappy. Second, production costs for movies will decline dramatically. Third, Hollywood’s business model will be in peril. Here’s MIT’s Sloan Review:

One of the many topics involving generative AI that is receiving a lot of attention is its potential effects on Hollywood and the entertainment industry. It’s an obvious concern because generative AI can create the types of outputs that the industry uses — text (in the form of stories, scripts, ad copy, and reviews), marketing campaigns, and moving and static images. Segments of the industry are facing economic pressures, which increases the demand for productivity and less-expensive “product.” And a high percentage of current entertainment is derivative of past content, which makes it well suited for generative technologies that are trained on … past content.

It’s still early days for generative AI-created entertainment, but it’s clear that something big is happening. A recent Wall Street Journal article noted that widely available AI tools can suggest storylines, character arcs, and dialogue; it even includes an interactive module that lets readers see for themselves how easily ChatGPT can write a basic script when given a few prompts.

This is why AI-generated movies are so important. They’re not important because they’re presently flawed. They’re important because the technology is rapidly improving, and at some point the technology threatens the Hollywood movie business. Instead of requiring a few hundred million dollars and thousands of workers to deliver a summer blockbuster, a summer blockbuster could be delivered to you from a few prompts in front of your computer. Can Hollywood adapt to that kind of business model? It’s certainly possible, but it’s likely going to be a hard road for many of Hollywood movie studios, and all of the ancillary businesses which thrive in that ecosystem.

We are rapidly heading to a place where powerful algorithms drive the cost to create entertainment to near zero. That’s great news for consumers, and it presents intriguing opportunities for entrepreneurs looking to build the next great entertainment platform. But it’s terrible news for a notoriously chaotic, and heavily unionized, industry. That’s why this tech is important, and that’s why it’s important to understand the implications of the saying Today’s AI tech is the worst you’ll ever use.