Businesses are building on open-source AI

Posted by @bfirsh

To cut to the chase: we’ve raised a $40 million Series B led by a16z. Let me explain why.

Last year, Stable Diffusion was released. It was an open-source image generation model that caught the imagination of tinkerers. An explosion of forks were created: inpainting, animation, texture generation, fine-tunes.

At the start, it felt like a toy. It was just people tinkering around and seeing what was possible.

But soon, side projects started to turn into real products. Indie hackers like Pieter Levels and Danny Postma made apps that generate profile pictures, redecorate your house, and create professional headshots. They’re now real businesses making over $1 million annual revenue as solo developers.

The growth in tinkering and building since then has been astonishing. In the last year and a half, 2 million people have signed up to Replicate, and 30,000 of them are paying customers. We’ve seen people build autonomous robots, real-time drawing apps, and language model command-line interfaces.

We thought it was just going to be startups and indie hackers, but side projects turned into startups which turned into growing startups. Hack projects inside large companies turned into real products.

Unsplash are labelling all the images in their catalog using BLIP. BuzzFeed are… turning your pets into plushies. Character AI, Labelbox, and thousands of other businesses are deploying models on Replicate.

Open-source language models

Although we got our start in generative images, language models are where we see most of our growth. The week Llama 2 came out was our biggest week of growth ever.

Language is the medium of thought, and the medium of business. We’re seeing more and more businesses switch from closed to open-source language models. You can fine-tune models like Llama 2 and Mistral to perform better than GPT-4 on particular tasks with a much smaller, cheaper model.

The move to open-source is partly practical, but it’s also partly philosophical: people want to own the model weights for their product, in case a proprietary model like GPT-4 changes or gets shut down.

Lots more people are going to be building with open-source language models, and there are some major improvements to language models on Replicate coming soon to support them all.

What’s next?

There are roughly two orders of magnitude more software engineers than there are machine learning engineers (~30 million vs. ~500,000). By building good tools, we think it is possible for software engineers to use machine learning in the same way they can use normal software.

But we’ve only reached a tiny fraction of those software developers and businesses. That’s why we’re raising a $40 million Series B. It’s led by a16z, and joined by NVentures (NVIDIA), Heavybit, Sequoia, and Y Combinator.

We’re going to be spending the money on:

  • Making it better for businesses: We’ve added features like organizations and deployments to make Replicate work better at scale. Hands-on support and lots more features are in the pipeline.
  • Making it faster: We’re going to make predictions much faster on our most popular models and they’ll have no cold boots. We’re also working on faster cold boots across the board.
  • Better language models: We’ve got some big updates coming soon.

Open-source is open for business. 😎