Holaaa! 👋🏻

We all know that building a successful business is more than access to great AI technology.

Fast access to AI tech → can help you build a successful company + some cash.  

Defensibility → helps you stay there.

When it comes to building defensibility around AI-powred apps everyone is a bit confused.

Let’s look at some examples from current apps:

📻 Improving existing models with user data

Even if you're using the same data, you can still build unique models - using your own user data. When you grow bigger, you collect more user data, so you can train better models, which can result in better results and more sales.

A great example of this is what Lexica is doing. They might figure out how to get the hands right…

🍭 Building a brand + sticky features

A strong brand can create a psychological barrier to entry for new competitors. Think of what Lex is doing - they are using the same GPT3 model, but they developed a clean, user-friendly document editing page that helps users write faster. Their core feature is the ability to type “+++” every time you need AI assistance to finish your content. Some users have said they find themselves using +++ in other apps because they’re so used to the feature.

Additionally, Lex has introduced a graph of your writing streaks, which may further incentivize users to stick with the tool.

📍 Focus on a niche market & problem

It’s always great to build a product around your domain expertise and industry insight that a few people have. Adding AI early into that will help you fine-tune big models for a specific use case. 

A good example of this is what Descript is doing - they have great experience/knowledge in the video-creation market, and they’re now adding GPT3 to create even better and more useful video editing software.

🛼 Fun Stuff

• Notion opens the waitlist for Notion AI - a writing assistant that can help you write, brainstorm, edit, summarize, and more. This is further proof that AI tech is on its path to becoming a commodity, and you need to think about defensibility around the tools you create today.

• Descript announced a $50M Series C, led by Open AI Startup Fund to create the most simple video editing software - as easy as editing a document.

• Artists felt that Deviant's new AI tool "stole" from them. So, Deviant made it possible for artists to opt out of databases that are used to train these AI models.

• Canva introduced a text-to-image generator. Bigger companies are watching 👀

• Galactica launched today - a large model for science that can summarize academic papers, solve math problems, generate Wiki articles, and write scientific code 🧬Subscribe now

🤓 Technical Stuff

• Researchers released new large collection of data(630k) of audio+ text pairs, and used them to pretrain a model to better understand audio inputs. This are huge news because we can learn so much by mixing audio and written text, as we talk and write differently.

• New multi-flow model called Versatile Difussion that supports image-to-text, image-variation, text-to-image, and text-variation. Try it here. I'll be testing it out over the next few days and will share what I learn in the next issue

🧱 What are you building?

If you're using AI in your work or projects, I'd love to hear about it! Please reply to this email and let us know what you're up to. We may feature your work in our newsletter.

❤️ If you like The Prompt, and want to support my work:

• Share The Prompt with a friend, and invite them to subscribe here.

• Book an ad in The Prompt ( reply to this email if you’re interested)