Welcome back to the AI For Humans Newsletter!

Have you seen those videos of people playing Geoguessr? It’s that web game where you have to determine from a single screen photo where in the world the photo was taken. The game had a HUGE moment on the internet a few years ago and it’s become a very competitive sport.

This week, I read a fascinating essay from Scott Alexander at Astral Codex Ten about how OpenAI’s new reasoning model o3 has gotten very good at Geoguessr. Scott goes through a bunch of examples of his own photos and shows that while it might not be able to determine where a random field of grass might be, it’s still impressively accurate overall.

The actual point of Scott’s essay (btw, Scott is one of the authors behind that AI 2027 speculative fiction work) is to explore what it feels like when AI can do things that feel superhuman. And, if the AI pundits are to be believed, there may be lots of those things directly around the corner. And it got me to thinking…

How Should We Feel When AI Does Something Better Than Us?

I started spinning on all the things AI already does better than me but none of those things really feel like my thing. I’m a decent writer and producer of media (whatever that is now) but I have never pretended to be a mathematician. So when o3 helps with my teenage daughter’s calculus homework I’m not weirded out at all. That seems like something a computer should be good at, right?

Then, I read Derek Thompson’s essay in the Atlantic this week arguing that AI might already be taking entry-level jobs from college grads and I started to get properly freaked out. What if eventually AI is better than me and you at everything?

What the hell Gavin? What’s the point of all this?

Sorry. I’m not trying to scare you or push for UBI (though, honestly, we should be talking about it). But I do want you to think about what makes you uniquely you, and how to double down on that right now. Maybe even rethink what you want to do with your life, no matter your age.

For me, this probably involves more creative writing and making things. After I’ve done that, I know that I feel a sense of joy and that time hasn’t been wasted.

There is an optimistic message here. We’re still early. AI might be accelerating fast, but the human ability to care, create, and wonder hasn’t been automated yet. And maybe that last part is the real key. Curiosity is the one thing we do instinctively that no model truly understands and, it’s entirely possible, might never fully get.

So the best move now? Lean into what lights you up. Ask better questions. Follow your weird hunches. That’s where the future lives.

More on Thursday!

- Gavin (and Kevin)

Catch up on Meta’s new AI app & see why their new voice model might need therapy 👇

We 💛 This - Runway’s Reference Tool

Runway has been crushing it as of late and their new Character Reference tool is a ton of fun. From weird isometric video game versions of your favorite images to bringing 3D models into AI video, it’s one of the best ways to try to get character consistency across the board.

Like all generative media, it’s not perfect. In each set of four images, I’m lucky to get two useable ones but the useable ones are pretty great. The slot machine aspect is there in full effect so just be prepared to spend some credits on getting what you want.

If you want to dig in more on how to get the best out of Runway’s tools, we highly suggest you follow their awesome community manager Timmy who shares a ton of learnings and will often answer question if you’re running into problems.