30 October ’24 | Link to podcast and transcript

One of the more interesting podcast interviews I have consumed in the past months / yr. Matt Cynamon the ‘Head of Library Sciences’ at Union Square Ventures comes on the AI + I podcast with Dan Shipper to talk about all the cool AI stuff they are cooking up at Union Square Ventures (USV).

You may have seen the OH (overheard) on the USV Librarian account @usvlibrarian on Twitter. That was built by him. Over 70 mins he takes us through the GenAI stuff they are building at USV.

Cool tools such as

– The Librarian: all the 15k articles they have written at USV which can be queried) [14:30 – 23:00 mins]

– Meeting Notes – OH [45:00mins onwards]

– The Dream Machine [1:03 onwards] – this is v trippy…AI-generated images to sync with conversation topics – loved the whales and cats imagery!

Am sure some other VC funds have some of these built (Yohei Nakajima comes to mind), but USV is of course one of the giants of venture. Impressed with the suite of tools that USV has built out, and grateful that they and Matt have been more public about it.

***

A couple of interesting points that emerged from the podcast episode

  • The cost of creating good storytelling is dropping. (AI is still not ready for releasing content without final review and polish and even some rewriting, but it gets you to 60-70% of the content very fast).
    • Matt: “The market for purely AI-generated content is the size of your biggest group chat.”
  • Hence with low costs of creation, AI can help you tell stories about things you wouldn’t create stories about, like internal meetings. There is a ton of content getting created that is not getting captured and AI can help you capture and make that content in a ready to consumer format.

My take on the above two points is that as creation gets democratized and the cost of creating drops, the importance of curation and distribution become more and more important. Having content that is uniquely about your opinion or taste, and looks like / reads like content that AI couldn’t have generated (or at least without distinct prompts) begins to matter.

Content could be meta-content like The Dream Machine – AI-generated images that emerge in real time to words spoken in a conversation – as content generation gets cheaper, new content formats could be meta ideas for content generation by GPTs, like a meta content idea for creating fake but seemingly real countries in the layout of wikipedia etc.

***

Link to my notes / highlights of the podcast episode.