Ruminations #14: My current set of beliefs from over 1000 conversations about AI

Everything from moats to hardware to research to product building

Hi all,

Last weekend I was contemplating my current set of beliefs about AI and AI products based on all of the conversations I’ve had over the last year. I realised that I met a bit over a thousand new people working in or on AI in 2023 - a terrifying realisation for someone who is much more introverted than they appear.

I decided to jot down all of the learnings that came to mind from those conversations.

I wouldn’t say it’s a complete list: I’m still in the process of mining my thoughts.

Regardless, I would love to hear your perspectives on the learnings and opinions shared in the write-up. Let me know if it is helpful!

---

---

OpenAI’s latest big release, Sora

I was pretty surprised when I saw this release.

Experts I speak to have been saying for the last year that video is the next big area of expansion for multi-modal foundation models. They’ve also said that it’s extremely technically difficult because of the computational intensity of working with video.

What’s especially difficult, I’m told, is creating a model that simulates the world, which is what’s needed to create a highly effective video generator model.

This is because you essentially need the model to develop an understanding of physics. Arguably some of the best computational models for physics exist in video game engines like Unreal Engine.

On that note, I have been thinking increasingly about the overlap between gaming and AI, as I think they fuel the progression of each other’s field (and no greater example is the fact that the GPUs used to train and run today’s foundation models were originally designed in large part for video games). I’ve read a lot of Matthew Ball’s writing on the crossover between gaming and technology but I would love to hear from others on the overlap of AI and gaming and what they’re reading in that space.

Anyway, I will share more as I speak to folks about it but in the meantime check out their release video below.

---

Content we’ve enjoyed

All In Podcast - Vision Pro: use or lose? Meta vs Snap, SaaS recovery, AI investing, rolling real estate crisis

From SP Partner, Yonatan:

Good discussion on who would be the winners and losers in AI and where to invest, circa 40 minutes in.

With a reply from SP Partner, Philippe:

I agree, very interesting podcast: the “picks and shovels” companies and the ones that own proprietary data can make lots of money. Open source models will crash the value of models to zero (although their utility value goes to infinity).

P.S. from Casey: I know the hosts of this show can be grating and their political persuasions can be polarising, if the politicised parts are not your jam just skip forward. I use Snip’d to quickly find the parts of podcasts that are interesting to me.

Earnest Analytics - AI Tools Sales Growth Report

My colleagues who run our listed fund shared this with the team. Fascinating to see OpenAI’s share shown this way, though it’s not surprising. ChatGPT is shown in green, Jasper AI is in purple.

It’s unclear how much of OpenAI’s growth in share came from an expansion of the overall use of AI Tools versus taking users from other players (I suspect it’s both but mostly the former given how large OpenAI is relative to Jasper’s initial size).

Forbes - AI 50

It’s worth checking out the Forbes AI 50 if you want to get a sense of what others are building in AI. You can filter by what they’re building, which is useful if you want to check out potential competitors.

---

That’s all!

Casey