🤠 Be a friend's hero today - share this newsletter and brighten their inbox.

THE INTERVIEW

Can you introduce yourselves and give us your best elevator pitch for your AI product?

We're Chris Kruger and Steffen Holm, co-founders of JARS AI. Our product is an interactive AI entertainment platform where users submit ideas for episodes, and AI characters act them out in real-time.

It's like a fusion of Roblox, Twitch, and The Sims, built from the ground up with AI. What makes JARS unique is the real-time social experience that allows users to collaboratively create content.

What inspired you to create JARS AI, and what's the most surprising insight about human behavior you've gained during its development?

Our key realization was the importance of prioritizing a fun, creative experience over ultra-realistic graphics. We view content creation as a shared, weird, and iterative process, and our aim is to provide tools that allow users to build upon each other's work.

The most surprising insight we've learned from building JARS is how people are still finding their footing when using Gen AI tools. Almost without fail, there is a phase of trying to break things and push boundaries before interacting with the core features of JARS. We're really excited about this behavior, though, as we think the most interesting use cases will probably come from people using JARS in ways we never intended.

Can you share a surprising user benefit and the quirkiest feature request that impacted your JARS’ development?

We were pleasantly surprised by the high entertainment value users find in JARS, with some spending hours daily on the platform. This makes sense now as it is just another form of entertainment that helps spark joy, but in the early days, we were just set on making a weird AI show platform with no idea how people would respond. 

The quirkiest feature requests we've been getting are for features that help users take more ownership of the content they make on JARS. One example is the feedback we received that led to the creation of "Avajars," - which allows users to make custom actors for their shows. Tools like these are huge as they unlock "exponential creativity" and lead to fun edge cases for these custom characters.

For example there is a character "JIM OILY" that started because of a rendering bug in our Avajar creation flow that resulted in him being a named character in our universe. He now has a whole backstory and community lore, which has been really funny to see grow!

How did you balance conversational AI with an intuitive interface, and what's the most unusual dataset you've used for training?

We're actively transitioning from rapid feature development to a more design-forward approach, trying to make JARS more intuitive for a broader audience. As we're entering the next phase of our growth, we're excited to be focusing on being more design-forward to align more closely with our core hypothesis of "making a fun, accessible experience is what allows someone in this space to win vs. just functional milestones."

Honestly, our most unusual dataset relates to toxicity management. Being a UGC platform where anyone can submit episodes has resulted in a robust dataset supplied by a wide array of bad actors. We've spent an unexpected amount of time internally building features that relate to trust and safety–ensuring that we can filter out hate speech of course, but also unique ways that people try to troll JARS specifically. 

This helps us design better user experience guardrails but it does point to the insight that Gen AI is still new enough that "trying to break it" is just as valid of a use case today as it is to use a product for its intended purpose. It will be interesting to see how this changes as Gen AI usage gets more ubiquitous and basic stress testing becomes less novel.

What's the best analogy you've used to explain JARS AI to non-tech people, and what's the funniest mistake your AI has made, and what did you learn from it?

We describe JARS as "personalized low fidelity TV generated by AI." Our main challenge is ensuring AI characters understand their context fully. Occasional quirks, like characters referring to themselves in third person are frustrating, but also present an opportunity to use the surplus of information coming out of our LLMs to create more engaging content.

Thank you for taking the time to chat with us. What can our readers look forward to from JARS AI?

We're excited about JARS' community-driven development! We'd, of course, love to invite your readers as users of JARS, but we also want to encourage bringing as many voices to our development roadmap as possible. Please join our Discord, share ideas, and help us shape the future of this weird corner of the internet!

Website 👇️ 

THE END

I'm Kevin, a former marketing technology director on a mission to provide informative reviews and interviews that respect your time. Please subscribe or share if you like. Thank you for reading.

Was this newsletter useful? Help me improve it! Click below to vote.

• 👍 It was informative. Thanks

• 😐 Meh - was ok.

• 👎 Not interesting to me.