433: Twitter SNAFU, Returns on Curiosity (ROC), Nvidia, HBO & Netflix, Google Reader, Anthropic’s Claude, Fentanyl, and Dune: Part 2

"because you never 'solve' humans"

 You don’t have a right to the cards you believe you should have been dealt with. You have an obligation to play the hell out of the ones you’re holding.

–Cheryl Strayed

I’d love to sit there and read a good book 💚 🥃

🛋️🤖🧑‍🎨 AI can be quite good at interior design *if* you know where to go in latent space to pull out designs that appeal to you. For example, I really like the image above. It was part of an experiment that turned out better than I expected!

Thinking about that process made me reflect on the different kinds of creativity involved with generative image models:

The first rule is that you don’t get to make *exactly* what you want, unlike if you drew or painted something (not that it’s easy to get exactly what you want that way either, but you have more direct control).

Instead, it’s more like you’re trying to pull images out of a dreaming mind that generates all kinds of strange, beautiful, scary, interesting, and random things.

You can steer the AI more or less in the direction you want by tweaking the words in your prompt and by adjusting the parameters (more or less chaos/randomness, fewer or more steps in the diffusion process, changing aspect ratios, etc) and prompting directly with images, but it’s all probabilistic and fuzzy and you won’t get the same picture twice even with the same prompt.

A big part of the process is iteration and curation. Start with the seed of an idea and then create lots and lots of variations, then pick the very best one. Rinse and repeat.

Recently I’ve been experimenting with even more advanced techniques like multi-prompting, multi-image prompting, and prompt-blending. It’s powerful stuff and I’m a newbie barely scratching the surface!

In short, this creation process is largely about imagination and curation vs the actual physical execution of the idea. It unbundles the two and unlocks ideas from people who otherwise couldn’t have executed on their inspiration.

🙏 The simplest way to build an audience online, whether on Twitter, a podcast, a newsletter, or YouTube, is to focus on one thing.

The reason is simple: Any time someone decides to follow you because of X and then you tell them about Y instead is an occasion for them to be turned off and leave.

If you only focus on X, over time you can get known for X and attract everybody else who is into X.

But if you cover X, Y, and Z, it’s a bit like the conjunction fallacy. The more things you add, the more specific your niche becomes and the fewer people will be interested in all those things.

I’m well aware of this, which is why I’m so happy to have found you — and that you have found me.

I feel like I have been playing this game in ‘hard’ mode. But hey! I can’t help it, I can’t stick to just one thing, and I care more about finding my ‘tribe’ of omnicurious (did I just coin a new word?) than having the biggest audience.

Quality over quantity. It’s a lot more fun to hang out with you because you get it!

---

🏦 💰 Liberty Capital 💳 💴

What are you looking at? This is fine

🐦 Non-Users Barred from Reading Tweets as Elon Musk Enforces Rate Limits on Twitter ✋🙈🛑

As I’m sure you know because everybody has been talking about it — unless you live in the real world and were touching grass this weekend, in which case, kudos to you! — Elon has rate-limited all of Twitter + made being signed in mandatory to even just look at Tweets or bio pages.

On its face, it’s a bit bonkers since Twitter makes money by showing ads, so it’s all about impressions, inventory, and engagement...

Musk assured us that this was a temporary measure to combat “data scraping and system manipulation.”

It’s possible that this is the real reason, but because he has made it very hard to take him at his word, we can also speculate that maybe they had a capacity crunch and just had to rapidly and drastically reduce server load (maybe they couldn’t pay some of their cloud bills, or were optimizing capacity and cut too much).

There’s even a theory that a bug in the Twitter app has made them accidentally DDoS themselves…