March 26 2024

Hey Creators,

Welcome to this week's installment on the inner workings of Create, the easiest way to code in natural language™️. Let's dive in right away, because we're stoked on multi-model generation and want to get back to playing around with it.

Jim Carrey types furiously on a computer — This has been Marcus all week. We swear this is his happy face.

🦾 MULTI-MODEL GENERATION IS STRONG...AND GETTING STRONGER 🏋️

Guys, the multimodal multi-model generation is really starting to hum. If you were an early Create adopter, you were probably frustrated by the speed or quality of output. Similar to my girlfriend when we started dating (before she properly trained me).

But now, different models for every version means you can get better balance on the all important speed vs quality ratio.

a venn diagram comparing speed, quality, and cost — Oh, to live in the utopic middle of a venn diagram...

In addition to our other models, we've added Anthropic's Haiku and Sonnet models, and they are lightning fast. That changes how it feels to build.

side by side comp of GPT-4 and Haiku building a netflix style page — GPT-4 vs Haiku. Almost 2x the speed!

Check out the full video here.

a man in a suit and tie with the words have you had enough? cause there's more

😮 THEY'RE ALSO INTEGRATIONS 😏

That's right. These models are Integrations, too. Beyond creating your app with these models, you can have your app use these models by pressing "/" like any other integration. This is the fastest way to build tools using these APIs.

a screen shot of a web page showing models as integrations

As we mentioned last week, different models are good at different things. Use a fast model like Groq or Haiku when you need near instant results. Use a big model like Opus or GPT-4 when response quality really matters.

Neo says he knows Kung Fu — Whichever LLM they used on Neo, it was FAST and QUALITY.

Let us know over on Discord what you're doing with the new models, or if there's anything we can do to help!

FASTER IMAGE GENERATION WITH STABLE DIFFUSION

We’ve upgraded the Stable Diffusion model we’re using for image generation and hey mama, this thing is fast.

sample credit card background image generator — Maybe we should have used The Flash for this demo...

We're talking, real time image repopulation fast, Superman flying around the Earth to turn back time fast, me inhaling a Philly cheesesteak after I finish this elimination diet fast.

Dhruv found a cool hack that you can see HERE. Take a peek at the spec to see how you can get Create to generate images for your app in the blink of an eye by linking img tag + alt text.

😛 1 THING WE LIKED THIS WEEK 🤪

Some people use twitter for silly Dune memes:

LA Dodgers hat but with Lisan Al Gaib — BRB off to buy this hat.

And some people use twitter to preach the gospel:

a tweet about training Haiku with examples — If you don't follow Matt, consider this your sign from God.

Claude 3 Haiku is an absolute game-changer. Super fast, high caliber output. And you can accomplish this in Create as well. Just give Haiku a prompt with 10 examples and 💣BOOM💥.

And if, like us, you're obsessed with Haiku, Matt had another great tweet last week that seems to mirror what we're seeing. It seems the best workflow is to start with a large high quality model (Opus) and then use a smaller, faster one (Haiku, Groq) to ideate.

You guys seeing any cool applications? Hit us up on Discord.

🤢 1 THING WE HATED THIS WEEK 🤮

Uhm, we went viral in Japan? We went viral in Japan!

To be clear, we do not hate this. In fact, we LOVE IT!

Twitter user @SuguruKun_ai posted a thread showing off many of the features that make Create awesome - "talking" to AI to generate real time UIs, using different LLMs, simple prompts or images to produce robust apps, you name it, my guy was tweeting about it.

Japanese thread extolling the virtues of Create — Does this mean we can get a sneak peek of the new Demon Slayer season?

Our user graphs had an enormous spike. But we had NO idea why. And that's where the Hate comes in.

We searched our own name on Twitter and got...no results. So we did what any smart company that doesn't know what the hell is going on does -- we asked our users. And wouldn't you know? They told us.

Shouldn't Twitter search work if you look for your own name. It felt like everyone was staring at us and we didn't know if something was in our teeth or we just looked really hot that day (we looked really hot). So Elon, I know Twitter has other issues, but could you look into that?