Silicon Valley’s next trillion-dollar companies?

Computer programs can now create never-before-seen images in seconds.

Feed one of these programs some words, and it will usually spit out a picture that actually matches the description, no matter how bizarre.

The pictures aren’t perfect. They often feature hands with extra fingers or digits that bend and curve unnaturally. Image generators have issues with text, coming up with nonsensical signs or making up their own alphabet.

But these image-generating programs — which look like toys today — could be the start of a big wave in technology. Technologists call them generative models, or generative AI.

“In the last three months, the words ‘generative AI’ went from, ‘no one even discussed this’ to the buzzword du jour,” said David Beisel, a venture capitalist at NextView Ventures.

In the past year, generative AI has gotten so much better that it’s inspired people to leave their jobs, start new companies and dream about a future where artificial intelligence could power a new generation of tech giants.

The field of artificial intelligence has been having a boom phase for the past half-decade or so, but most of those advancements have been related to making sense of existing data. AI models have quickly grown efficient enough to recognize whether there’s a cat in a photo you just took on your phone and reliable enough to power results from a Google search engine billions of times per day.

But generative AI models can produce something entirely new that wasn’t there before — in other words, they’re creating, not just analyzing.

“The impressive part, even for me, is that it’s able to compose new stuff,” said Boris Dayma, creator of the Craiyon generative AI. “It’s not just creating old images, it’s new things that can be completely different to what it’s seen before.”

Sequoia Capital — historically the most successful venture capital firm in the history of the industry, with early bets on companies like Apple and Google — says in a blog post on its website that “Generative AI has the potential to generate trillions of dollars of economic value.” The VC firm predicts that generative AI could change every industry that requires humans to create original work, from gaming to advertising to law.

In a twist, Sequoia also notes in the post that the message was partially written by GPT-3, a generative AI that produces text.

Image generation uses techniques from a subset of machine learning called deep learning, which has driven most of the advancements in the field of artificial intelligence since a landmark 2012 paper about image classification ignited renewed interest in the technology.

Deep learning uses models trained on large sets of data until the program understands relationships in that data. Then the model can be used for applications, like identifying if a picture has a dog in it, or translating text.

Image generators work by turning this process on its head. Instead of translating from English to French, for example, they translate an English phrase into an image. They usually have…