Art Isn’t Dead, It’s Just Machine-Generated

•

Creative endeavors are more likely to be disrupted by generative AI long before systems building. Why? It is far easier to automate rather difficult creative tasks than relatively simple programming tasks. Code has to be correct. Art does not. 🧵👇

A product like GitHub Copilot, in its current form, can make coding more efficient, but it doesn’t replace the need for devs with programming knowledge. However, anyone can use a product like Stable Diffusion to produce high-quality and stunningly complete images in minutes.

Generative AI is commonly used as a programmer’s aid, but the productivity gains are modest. Recent studies have shown AI has improved developer productivity by 2x, but that’s on par with previous advances. The jump from assembly to C improved productivity by an estimated 2-5x.

On the other hand, generative AI models’ impact on creative output is extreme, and there is a massive difference in the economic calculus between code and image generation.

To create the below pictures, @appenz took a pre-trained image model and re-trained it on a few dozen photos of himself, costing around $.50 in resources. Once trained, generating images costs about $0.001 in compute and can be done in the cloud or on a latest-gen laptop.

This isn’t the end of artistic creativity. Similar to programming aids, generative AI will be adopted as a tool by artists, as both require some degree of user supervision.

Take our own cartoonist (and investment partner) @stuffyokodraws: Using 70 of her drawings, a model was able to generate images with an eerie level of mimicry. This can speed up her creative process, or help her figure out what to create next. Can you tell which are AI created?

When it comes to tasks like creative generation of content, the argument isn’t that computers are necessarily better than humans on a 1:1 basis. But as with many other tasks, when computers can produce complete work output, they just kill us on scale.

AI brings a massive improvement in economics, flexibility in being able to craft new styles and concepts, and the ability to generate complete or nearly complete works. It’s hard not to see a full phase shift in industries where creative assets are a major part of the business.

Read more from @appenz, @BornsteinMatt, @martin_casado, @stuffyokodraws and the a16z infra team about how generative AI makes art an extension of programming (and not the other way around) ↓ a16z.com/2022/11/16/creativity-as-an-app/

None of the above should be taken as investment advice; please see a16z.com/disclosures for more information.