The case for Distilled AI

Yesterday we set the foundation. Today let’s talk use cases.

Most AI applications to date, with or without web3, have focused on either AI for fun or AI for productivity. Success has come in both, and therefore as a foundation Distilled AI will support applications in both categories. Here’s how it’ll look for individuals, groups, schools, artists, and beyond.

Fun

If you’ve ever tried to teach your chatbot to learn your unique messaging styles for whatever reason, you’re certainly not the only one. Bots coming straight out of foundational models naturally lack inherent personality. And we’re here to change that. Access to personal data from each human gives private agents valuable contextual information, unique styles and flairs, and the necessary background — including preferences and limits — to perform autonomous actions. With Distilled AI, these synthetic lives take on unique personas — digital twins that mirror human’s exact tone whether it’s the blunt sarcasm for casual social groups or slick professional tone used when conversing with colleagues.

You feed in the samples from your private side and agents will replicate. You feed in any early ideas or abstract concepts and agents will eventually be able to articulate them better than you ever can. They can do all this while remaining completely secure and private.

But it’s not limited to just individuals. Groups, too.

One of the Messenger groups that I was in recently tried to teach Meta AI to swear — you know how hard it is as it kept saying how it could not be offensive. Succeeding to get the AI to actually do that was downright hilarious. Most fun they’d had in months.

But here’s what we are most bullish on: Keeping the Culture Alive. For those that have been in touch with us while we were in stealth, you know how important cultural distinction is to us. With shared repositories for cultural groups to pool their languages, expressions, diction, and cadence, agents can become truly one-of-ours — speaking and behaving precisely like how you would expect from a person from your own communities.

That alone opens up avenues for applications that are amusing to say the least. And we haven’t even touched on what cross-pollinations of unique cultural identities can bring. Endless possibilities. That topic deserves a deep dive of its own — but we’ll save it for another day.

With culture-based data, these agents can master the subtle humor, local slang, and distinct quirks unique to your community. They can carry the vibe of your favorite subculture. A group of friends could train an agent to speak in the lingo of their favorite gaming community, or a neighborhood group could craft an agent fluent in local gossip and inside jokes. These agents could create the experience of hanging out with someone who just gets it, whether it’s a vintage punk bot or a die hard sci-fi nerd.

Keeping things entertaining is a great way to keep users engaged in any new piece of technology.

Applications we will fund in this category: digital twins, professional shitposters, narrative shillers, cultural spokesperson, etc.

Productivity

That being said, for the vast majority of people, life goals are not to spend hours on end talking to AI. They want to spend their finite amount of time on this earth interacting with real people. As with many pieces of technology in the past decades, the objective remains the same: to free up time so people become more productive, and thus can spend more time on actual real life hobbies. AI is not an exception.

Now think about how rich and powerful people save time — they have executive assistance, and with AI this is a wonderful opportunity to bring that capability to the common people. Each person has his own group of executive assistants, and based on the request, it can then route it — summon different task-specific AI agents. Private data as a result become be the common denominator of these agents — tens or even hundreds of these little little guys, all knowing everything about each of you, your experiences your preferences, your value, your knowledge, your biases right your likes and your dislikes so it can be helpful for you freeing up your time so you can do all the things that you love.

Applications we’ll fund in this category: life coach, memory keeper, email helper, travel planner, blog writer, music and video producer, …

Those are just broadstrokes. I trust that our developer community will find even more creative ways to drive innovation, more viral use cases. I’ll end the blog here with a few of my personal favorites, things that I, my family, and the people who are close to me would personally use on a daily basis.

Arts: The Collective Brain of the Artistic Universe

• Private Studios: Artists from all over the world could join a digital studio, merging skills in real-time on canvases, songs, or sculptures. You could see Van Gogh collaborating with Banksy — AI-style — in a purely imaginative, AI-crafted environment.
• AI Gallerists and Curators: AIs trained on the work of specific artists curate unique exhibitions, even generating pieces inspired by the original work. Imagine “Lost Works of the Renaissance” made accessible today, not as reproductions but as new interpretations through AI.
• Personalized Creativity Coaches: Artists can train an AI on their past work, and it then suggests new directions, inspirations, and even collaborations based on their unique style and preferences. Imagine a painter whose AI studies their strokes, then suggests color schemes and ideas based on their mood.

Education — personalized bespoke tutors, living legacy, research oracle

• Personalized Learning Paths: Each AI deciphers the student’s unique strengths and blind spots, customizing a curriculum that aligns with their levels of understanding and inclinations.
• Living legacy: generations from now, the great minds from our current world will still be preserved. This will be equivalent to students in our current generation being able to converse with historical figures — Sun Tzu, Martin Luther King, you name it.
• Research Oracle: Scholars can engage with an AI that seamlessly adapts, providing real-time insights and layered context. A question on Foucault, for example, doesn’t yield just an answer but a web of critiques, histories, and connections — much like a partner in thought.

So for that, I think it’s extremely important for us to build a solution that allows AI to access private data sources securely. Knowledge distillation is a great way to achieve this — the idea of extracting knowledge from large foundational models so that the result is a much smaller, more efficient, and more useful AI agent tailored to each individual. This idea motivates me; it’s similar to how we, as humans, gain basic knowledge early on and eventually specialize. We all start with basic knowledge — reasoning, math, reading — shared across the world. Yet over time, we each develop different expertise and domain knowledge, distilled into our minds through years of education, personal learning, and experiences. In this way, each of us real humans effectively becomes a private agent, grounded in shared foundations but refined by individual paths.