Content generation >> content creation

Or how to become a creator by pressing a button


It's getting warm outside and big tech is chasing the heat. All big tech players rush to integrate AI capabilities into their offerings. I offered this as a prompt to DALL-E and I got the above. Youtube, Snapchat, Instagram, WhatsApp are about to be redefined (or are they?). It feels like a reaction response to the AI frenzy everyone is in and personally I have yet to see any big tech come up with a wow use case. So buckle up to have a chatGPT within all your platforms.

What do we have for you today?

📺 Youtube announces their plan to integrate generative AI capabilities into their platform and allow users to “virtually swap outfits” or “create a fantastical film setting”. I’m all …eyes 👀

🧞 Snapchat introduces new AI feature sharing a vision where all of us will talk to family, friends, and AI. Time to make some AI friends - they rarely talk back and will stay with you forever

💡 Shall all builders think more about the opportunities that “active learning” offers? For certain use cases, it may help bridge the gap between prototype results and the accuracy required in the real world

The future of content (co-)creation


Content creation is about to be shaken from the ground up with the new AI tools. It’s easy to imagine how the content creation pyramid as we know it (1% creators, 9% people that interact, 90% viewers) is about to be redefined. The rationale is simple: creating content takes time and energy. With AI, you can create new content at the click of a button.

It’s already happening with image generation, slideshow presentations and I’ve previously written about how co-creation will redefine learning.

The latest big announcement comes from Youtube. The new Youtube head honcho, Neal Mohan has announced Youtube is planning to integrate generative AI capabilities into their platform, teasing that creators could try "virtually swapping outfits” or "creating a fantastical film setting through AI's generative capabilities”.

The implications on the ecosystem can be big. I see a new wave of tooling that will appear to support creators. We only need to look at Capcut’s success to see how good tools can become great businesses. I’m very excited to watch this space.

Is it a tool or is it a person?


Open AI’s ChatGPT is positioned as a productivity tool. It doesn’t really have a personality - you don’t want to "hang out" with chatGPT. You use it in secret when your boss wants you to do something.

Snapchat just released an AI feature for its premium user base that on the surface is not too different than chatGPT. In fact, it actually is a more restricted version of chatGPT. Given the younger audience, Snapchat wants to ensure the chatbot adheres to the company’s trust and safety guidelines.

The way they positioned it though is different. “My AI” as they call it, is a persona (with a proper alien face and its own alien Bitmoji). Snap’s CEO Evan Spiegel sees a future where we all chat to AI as it's the most normal thing: “the big idea is that in addition to talking to our friends and family every day, we’re going to talk to AI every day,”.

I’m expecting Clubhouse now to launch AI integration and get a bunch of AIs and humans in a room to discuss stuff. Might not even be a bad idea.

Are you doing active learning or not?


In traditional generative ai models, a lot of the improvements in accuracy come as a result of the model being released in the world, capturing feedback, and improving as it is being used. It’s similar to how we learn as humans, learning from our mistakes and through trial and error.

Eric Landau argues the future of generative AI will be powered by active learning. With active learning, models learn from pre-labeled data and then explore unlabeled data. It’s a form of semi-supervised learning, where you only need part of your data to be labeled.

Eric argues that deploying this technique can enable companies to bridge what he calls the “AI production gap”, a situation that often occurs when the accuracy of the model outputs required are just a bit off from what’s required in a real-life scenario.

Perhaps active learning can help get us to results closer to what we have in mind when we provide prompts to models. But as history has shown us, we’re not always after the “best” results. We might be hooked more by the gap between what we expect and what we get (hello old friend dopamine). Also, with variable pricing models, maybe the incentives of companies are for models to always be just a tiny bit off, right?

Gen AI Deals that make your eyes (and mouth) water 💰

What caught our eye? 👀Founded by Abhay Parsanis, Adobe’s former CTO, Typeface is a platform trained on the ChatGPT and static diffusion model that can generate personalized blogs, Instagram posts and websites for companies.

What caught our eye? 👀Robin AI utilizes Anthropic’s large language model to draft and edit legal contracts with the goal to help companies reduce costly lawyer fees.

What caught our eye? 👀In the words of CEO Daniel Weiner: “Autobound’s AI suggests hyper-personalized content to the specific individual you’re reaching out to. Think ChatGPT, but for sales emails. We’re making it simple for any seller to write an A+ sales email instantly, so they can focus on revenue-producing activities like actually talking to buyers.”

New developments to spam your #random Slack channel 💬

Things to learn when you need a raise

Want access to the Generative AI investors database? Share and subscribe to AI In the Middle (below) to receive it! ⬇️

Before you go

If we could see history through the lens of a selfie, it would look like this.

And scene. That’s all for this week, we’ll see you Monday. Thanks for reading, sharing, and subscribing. Have something to share? Slide in our DMs.

— Calin Drimbau (@calindrimbau)