The TED AI Show Episode 1, A16Z Scout & AI Quick Hits (I/O, OpenAI and more...)

My new podcast with TED, a call for early-stage founders, and a summary of the latest in AI

Hey Creative Technologists,

It’s been a minute! The last few weeks have been a whirlwind for me, and I’ve got some exciting news to share. In this edition:

1. My new podcast with TED + Ep 1 on Deepfakes

2. A16Z Speedrun + call for early-stage founders

3. AI quick hits (i/o, gpt4-o, wonder dynamics, and celebrity lookalikes)

Things are getting meta with - project Astra ‘watching’ OpenAI’s GPT-4o announcement. But the race is clear — OpenAI wants to be omnipresent — literally mediating your interactions with the physical & digital world.

TED2024 was a blast, and my new podcast with TED is now live!

First and foremost, I’m the host of the new TED AI Show. Our goal is to be your weekly guide to live and thrive in a world where AI is changing everything.

And episode 1 is live! We go deep into deep fakes and explore how AI is

Check it out: The TED AI Show — available wherever you get your podcasts.

I also had the opportunity to co-host session 2 at the TED conference in Vancouver with the one and only Chris Anderson. It featured the likes of Vinod Khosla, Dr. Fei-Fei Lei, Thomas Dohmke, Helen Toner and more. Definitely one of those bucket list moments.

A16Z Speedrun & call for early-stage founders (I’m signing scout checks!)

It’s been a blast to partner up with the a16z games. Thanks to everyone who applied to the upcoming Speedrun accelerator. Areas of interest include:

1. Visual 3D Creation Tools

2. Spatial Computing Apps

3. Real World AI & Digital Twins

4. Generative AI Experiences

But don’t worry! If you missed the deadline, and are building something cool in the areas above (or know someone else who is), I’m signing early stage scout checks. Please hit me up: bilawal@metaversity.us

Quick Hits

1. Google i/o — Agentic multimodal AI infused into almost everything

It was a blast to kick it with my (and likely your) favorite AI creators in Mountain View for Google’s annual developer conference.

I also had a chance to get hands on with Veo and the team behind it. It’s Google’s Sora competitor (more on that in another edition), but the real star of the show was Project Astra — which is another step closer to JARVIS:

2. GPT-4o — multimodal in & out brings us one step closer to JARVIS

   Speaking of Project Astra, OpenAI stepped up their game (one day ahead of Google) with new variant of GPT-4o that’s just as compelling. Rather than duct-taping specialized models for text + audio + vision, it’s multimodal from the ground up, just like Google Gemini. And of course, Microsoft showcased how it’ll make Copilot better:

3. Autodesk acquires Wonder Dynamics - AI-based VFX & 3D Tool

   In an interesting twist, one of the most interesting VFX/3D startups has been acquired by Autodesk — with mixed reactions from the community. My thoughts on it in the post below:

4. Banger AI Paper: Create a 3D model from a single image, set of images or a text prompt in < 1 minute 😮‍💨 — CAT3D

   The Google 3D research team strikes again! Looks much better than ReconFusion too. Here’s hoping there’s a code release. Meanwhile, you can play with samples here.

   

5. The AI training data conversation is reaching a tipping point

   When at VFX conferences, I learn everyone from small to large studios are using generative AI, but are too afraid to talk about it publicly. At the heart of this issue is an as yet open ethical question around training on copyrighted data scraped from ‘publicly available’ sources.

Meanwhile, Hollywood and celebrity heavy hitters might be the reason the tech industry finally start talks about this practice of training on ‘publicly available content’ — and whether it’s truly fair use.

It’s an interesting situation. Are we living in a world where celebrities will be compelled to take the money to allow AI companies to use their likeness — lest they be ripped off and replaced with eerily similar replicas?

Not to dissimilar from the pressure for platforms with unique data (eg. Reddit) to do licensing deals with AI companies. They’re getting trained on either way, so they might as well get paid for it.

---

That’s it for this edition! I'm excited to continue exploring these topics in future posts and on the TED AI Show. Particularly stoked for next week’s episode, which is gonna be an exclusive. For those in the US — I hope you have an awesome memorial day holiday. See y’all in the next one!

Cheers,

Bilawal Sidhu