OpenAI New Investments + Kickstarter AI Drama

Here's what is on the docket today:

• OpenAI Funds 4 Startups

• Unstable Diffusion kicked off of Kickstart

OpenAI invests in 4 startups

OpenAI announced a 100 million dollar investment fund in November, promising that their investments would “get early access to future OpenAI systems.” Judging from the massive power that OpenAI’s API has- just look at Google’s response to ChatGPT’s release- this seems like it will be a massive edge for any startup they fund.

The startups they invested in are called Descript, Mem, Harvey, and Speak- but what do they do?

Descript is a startup focusing on AI-powered video-editing founded by Andrew Mason- former CEO/founder of both Groupon (once worth as much as $16B) and detour.com (acquired by Bose).

Essentially, it’s a video editing software, reimagined. Their core feature is the ability to edit videos in a way similar to google documents- simply delete the transcribed words you don’t want in the video, and the relevant clip will be cut out. It also seems you can arrange visuals in a manner similar to Google Slides. In terms of AI capabilities, the features I find most exciting are the following: built-in audio enhancement, filler word removal (get rid of those annoying “um”s, and overdub (create a clone of your voice).

It’s so far raised $100M and seems like an appealing product to members of the growing creator economy, who may not be as technically savvy as traditional video editors.

Mem focuses on a broader issue- workspaces. They call themselves the “first self-organizing workspace.” Founded by two Stanford grads turned product managers turned founders, Mem promises to do it all.

The core features seem to be surfacing relevant content, helping you write notes better, etc. Normally I’d be dubious that anything related to a “second brain” would be useful, but with the rapid progression of AI tools, I could see Mem greatly enhancing knowledge workers’ productivity if they manage to pull off the technology.

Speak’s core product is an AI tutor chat bot which can have open-ended conversations with learners on any number of topics, and provide real-time feedback on pronunciation, grammar, vocabulary, and more.

Harvey, still in beta, has the unique business plan of using generative AI to make lawyers’ legal workflows more efficient. It was founded by Winston Weinberg, previously an antitrust litigator and Gabriel Pereyra, previously a research scientist at DeepMind.

What exactly it means to be an OpenAI-funded startup is as of yet unclear, but one thing is for sure- I wouldn’t bet against any of them; the early knowledge about/compatibility newer models may be an existential advantage.

Just look at AI behemoth JasperAI, a successful company who fine-tunes and prompts OpenAI’s GPT-3 to write engaging copy- they were caught off guard by the sudden release of ChatGPT, which can (for free) perform much of the same functionality as Jasper. For a company as large as Jasper, most recently valued at 1.5 billion dollars, the threat that the changing landscape of AI poses is as of yet uncertain.

Unstable Diffusion Kicked off Kickstarter

In other news, Kickstarter has banned an AI project amid backlash and announced a new policy when it comes to AI projects- namely, focusing on removing AI projects which use work from uninformed creators in their training models.

This move comes as a broader wave of concern has arisen concerning AI-generated art- specifically due to the fact that many of these models get their training data without asking all their sources for permission.

The AI project in question goes by the name “Unstable Diffusion,” a reference to the popular open source image generator Stable Diffusion. You see, a couple weeks ago, Stable Diffusion launched a version 2.0 to the welcoming arms of the AI art community, but some creators were saddened by the fact that it didn’t include nude images, celebrities, tec. In the dataset.

And so, fans of AI content generation banded together to form the kickstart for Unstable Diffusion, where they hoped to create an image generation model unfettered by the typical constraints of for-profit or public facing companies- allowing them to freely use mature content in their training models.

This caused a tsunami of backlash.

Despite this, Unstable Diffusion continues to ask for funding on its website through Stripe.

Thank you for reading

This was Steel Man's first article, and we hope to continue to bring you the most recent and important AI news- every Monday, Wednesday, and Friday.