29 startups in AI I'm watching closely after OpenAI's meltdown
29 startups in AI I'm watching closely after OpenAI's meltdown
On the eve of ChatGPT's birthday, there's a whole new class of billion-dollar-plus companies. Plus, Together AI makes a big splash with a round led by Kleiner Perkins.
Author’s note: Double feature today! We’re covering a lot of startups and a mega funding round for Together AI.
The anniversary of ChatGPT’s launch, which effectively ushered in a new era of AI and created a zeitgeist as large as the iPhone, feels like it crept up a lot faster than we’d expect.
A lot has changed since its launch. ChatGPT didn’t just spawn an entire new class of products with an industry-altering technology. It seems to have also inspired a whole generation of developers that are racing to build open-source equivalents—and companies right on top of those. And the infrastructure layer for deploying, managing, and accessing these models is only getting more complex.
You could easily rattle off the names of fifty startups that are all barreling into the unicorn club, from foundation model developers like Cohere to applications on top of emerging modalities like ElevenLabs. And there is of course OpenAI at the center of all of this, doing its best to beat everyone on price while offering some of the best products available—if you’re willing to use an API.
But not everyone is willing to use an API. And as we come up on ChatGPT’s birthday we’re moving on from the “holy shit” moment of AI into the “okay, so what do we do with this” phase of AI. And there is a whole universe of use cases waiting to be tapped by a whole new generation of startups.
And the timing is also kind of bizarre given what’s played out over the past few weeks with OpenAI CEO Sam Altman’s dramatic ouster and return. At a time when OpenAI looked like it was building momentum on all sides, it now faces an even steeper climb than before to convince enterprises to build tools on top of its APIs.
So, I figured as we come up on the anniversary of one of the most consequential launches of the past decade, I’d just put out my “watch list” of companies that I’m following pretty closely relative to others in the industry. The list has emerged from a mix of a lot of reasons, ranging from their presence hyped industries, quiet growing traction, buzz among investors, and expert opinions from sources—as well as products I’ve found to just generally be really helpful and easy to use.
I tell people that the only way to really keep up in the industry is to talk to as many people as you can and try to be everywhere at once (like Harrison Chase, somehow). And even still it’s hard to keep up. But the one commonality among the startups (and a handful of projects) on the list is that they come up consistently—whether that’s in idle chatter at an event, conversations with sources, or going viral on Product Hunt Hacker News.
Generally speaking, the startups on this list exclude the obvious ones that have cemented a place as a fixture in AI. That includes companies in the infrastructure layer like Hugging Face, but also companies deploying AI at a considerable scale at the app layer (like a Notion). And in general I try to skew as early as possible. If there is a larger company on this list that falls in that category, it’s more because of the open question around its ultimate fate or due to some recent developments that have impacted it.
This list is specifically focused on AI and companies in the immediate orbit of the industry, and excludes a number of other emerging foundational technologies, like MotherDuck, that I’d classify more along the lines of “big data” companies—which the other half of Supervised. (I’ll be putting out my watch list for big data companies sometime later in December.)
With that, let’s get to the list, which is lengthy and in no particular order
Startups I’m following closely in AI right now
GGML: Founded by Georgi Gerganov, ggml pioneered a file format for quantized (shrunken) language models that can be run on less powerful edge devices. GGML (and the more modern GGUF) is the company behind Llama.cpp, a C++ package that allows developers to run a variety of quantized open source models on edge devices like a laptop.
Guardrails: Run by founding Predibase engineer Shreya Rajpal, Guardrails places an explicit wall between a language model and its output. Most recently it launched tooling around putting up string-based guardrails as well. If a model doesn’t output something in the correct format, it ships it back to the model and basically says “hey, try again” with some additional instructions.
Voyage AI: While most consider Voyage AI—and really any embedding startup—late, one has finally arrived. Voyage AI provides a real potential competitor to OpenAI’s ada-series embedding tools. There have been superior embedding models than ada-002