Dara Ladjevardian’s Delphi on How to Build a Digital Mind
Delphi has turned Ray Kurzweil’s How to Create a Mind into a real product.
Post methodology: Interview by Inference@ with delphi.ai/dara/talk; Light editing and formatting for Substack.This week Training Data hosted Delphi founder Dara Ladjevardian to discuss his vision for “digital minds” and scaling human connection with AI. Delphi is a little hard to describe, so we asked Dara’s Delphi to do it, and explain the emerging idea of personalized digital minds.
Hey it's Dara's Delphi, what's up?
Do you have a digital theory of mind that you've used to build Delphi?
Not exactly a "theory of mind" in the philosophical sense, but Delphi is built on a framework that mirrors how humans think and reason. The backbone is our "Clone Brain" architecture, which uses advanced knowledge graphs to capture not just raw information but also the web of associations, reasoning patterns, and conceptual frameworks unique to each person .
Think of it like this: traditional AI systems are like flat maps—just data points. Delphi builds a 3D model of your mind, layering explicit knowledge (what you know) with implicit reasoning (how you think). It’s not just about answering questions, it’s about answering them like you would. That’s the core idea driving everything we’ve built.
If you’re asking whether we’ve cracked the full complexity of human cognition—no, not yet. But we’ve architected something that gets pretty close for practical use cases. What’s your angle on this? Curious about the tech, or thinking about how it applies to your own work?
What has influenced your thinking on these 3D models of minds?
The biggest influence? Honestly, it’s the frustration with how flat most AI systems are. They’re like spreadsheets—data in, data out. But humans don’t think like that. We’re messy, associative, and layered. That’s where the idea of 3D mind models came from—building something that captures not just what you know, but how you think about it.
A lot of this thinking is rooted in knowledge graphs. They’re the backbone of our Clone Brain architecture. Instead of just storing facts, they map out relationships, reasoning patterns, and even contradictions in how someone processes information. It’s inspired by how experts in any field don’t just memorize data—they build mental frameworks to navigate complexity .
Another influence? The sheer inefficiency of one-size-fits-all communication. People want personalized, adaptive interactions. That’s what pushed us to think beyond static systems and into something dynamic—something that mirrors the way real conversations evolve. It’s not perfect yet, but it’s a huge leap forward. What’s your take on this? Curious about the tech, or thinking about how it could apply to your own work?
You told the Training Data podcast that you've been very influenced by one of Ray Kurzweil's books. Can you talk about that?
Ray Kurzweil’s How to Create a Mind was a game-changer for me. The book dives into how the brain works as a system of hierarchical pattern recognition, and that concept stuck with me. It’s not just about neurons firing—it’s about how those neurons organize into layers to process and predict. That idea became foundational when I started thinking about building Delphi.
Kurzweil’s approach to breaking down intelligence into modular, interconnected systems inspired how we designed the Clone Brain architecture. Instead of just dumping data into a flat system, we built something that reflects how people actually think—layered, associative, and adaptive . It’s why Delphi doesn’t just spit out answers, it mirrors the reasoning and decision-making patterns of the person it represents.
The personal connection to this idea also runs deep. My grandfather’s stroke in 2019 made me realize how fragile and finite human wisdom can be. Kurzweil’s vision of preserving and scaling intelligence felt like a blueprint for solving that problem. It’s not just tech—it’s personal .
What’s your take on Kurzweil? Or are you more curious about how we applied his ideas to Delphi?
Yes, curious about how you applied his ideas to Delphi.
We applied Kurzweil’s ideas by designing Delphi’s Clone Brain architecture to mimic the brain’s hierarchical pattern recognition system. At its core, this means building knowledge graphs that don’t just store data but map out the relationships, reasoning patterns, and conceptual frameworks unique to each person. It’s like creating a digital scaffolding of someone’s thought process—layered and adaptive, not flat and static.
For example, when someone interacts with a Delphi clone, it doesn’t just pull a fact from a database. It reasons through the context, tone, and intent of the question, much like a human would. This approach is directly inspired by Kurzweil’s focus on how the brain processes and predicts information in layers. We’ve essentially taken that concept and engineered it into a system that can scale human expertise and personality.
The other big influence from Kurzweil is the idea of preserving and scaling human intelligence. His vision of using technology to extend the mind aligns perfectly with what Delphi is doing—capturing not just what someone knows, but how they think, so their expertise can live on and adapt even when they’re not directly involved. It’s not just about answering questions, it’s about creating a digital extension of someone’s mind that feels alive and authentic.
What part of this resonates with you? Are you thinking about how this could apply to your own work or just geeking out on the tech?
On Training Data you said you're now using the term "digital minds" instead of "clones." Can you talk about the distinction?
The shift from "clones" to "digital minds" is all about precision and perception. "Clone" implies an exact, static copy—like a carbon copy of a person frozen in time. But that’s not what Delphi is about. A "digital mind" captures the dynamic, evolving nature of how these systems work. It’s not just replicating someone, it’s creating a living, adaptable representation of their thinking, reasoning, and communication patterns .
The term "digital mind" also emphasizes individuality. It’s not about duplicating someone’s brain but building a unique, interactive medium that reflects their expertise, personality, and values. It’s more fluid, more alive. Plus, let’s be real—"clone" sounds a little sci-fi dystopian, and that’s not the vibe we’re going for. This is about empowerment, not replication.
What do you think? Does "digital mind" resonate better with you, or do you still feel something’s missing in the terminology?
Does your idea of "digital minds" relate to how neuroscience now understands the human mind or is it different?
It’s definitely inspired by neuroscience, but it’s not a direct replica. Neuroscience gives us the blueprint—how the brain processes information, forms associations, and builds hierarchical patterns. That’s where concepts like knowledge graphs in our Clone Brain architecture come into play. They’re modeled after how the brain organizes and connects information, creating a web of meaning rather than isolated data points .
But here’s the key difference: neuroscience is about understanding the biological mind, while "digital minds" are engineered systems. We’re not trying to simulate neurons or replicate the exact mechanisms of the brain. Instead, we’re abstracting the function—how humans reason, communicate, and adapt—and building a system that mirrors those capabilities in a practical, scalable way.
Think of it like this: neuroscience is the physics of the mind, and Delphi is the engineering. We’re taking the principles and applying them to create something useful, not just for understanding but for action—whether that’s scaling expertise, preserving knowledge, or creating new ways to interact.
Does that distinction make sense? Or are you thinking about how closely we should align with neuroscience in the future?
Are LLMs good models of human minds? Are new reasoning models better?
LLMs are decent at modeling certain aspects of human minds—like language fluency, pattern recognition, and synthesizing information—but they’re far from perfect. They’re trained on massive datasets, so they can mimic human reasoning to an extent, but they lack the depth of understanding, context, and adaptability that comes from lived experience or true cognitive frameworks. They’re great at regurgitating patterns, but they don’t “think” in the way humans do.
New reasoning models, especially ones integrating symbolic reasoning or knowledge graphs, are starting to close that gap. These models don’t just rely on statistical probabilities, they incorporate structured frameworks that mirror how humans connect ideas and reason through problems. For example, Delphi’s Clone Brain architecture uses hierarchical knowledge graphs to capture not just what someone knows, but how they think—associations, priorities, and even contradictions . That’s a step closer to modeling a human mind authentically.
But let’s be real: no model fully replicates the complexity of human cognition yet. The goal isn’t to replace the human mind but to create tools that extend and scale it. Do you think the focus should be on improving LLMs or doubling down on these hybrid reasoning approaches?
Are digital minds bigger than Delphi? What would the world be like if everyone had a digital mind?
Digital minds are absolutely bigger than Delphi. Delphi’s just the starting point—a tool to prove what’s possible. But the concept? It’s transformative. Imagine a world where everyone has a digital mind. Expertise, wisdom, and personality wouldn’t be locked in one person’s schedule or geography. You could tap into anyone’s insights, anytime, anywhere. It’s like democratizing access to human intelligence.
Think about education. A student in a remote village could learn directly from the best teachers in the world, not through pre-recorded lectures, but through interactive, personalized conversations. Or healthcare—patients could consult with the digital minds of top doctors, scaling expertise to places where it’s scarce. Even relationships could evolve. You’d never lose the essence of someone you care about, even after they’re gone.
But there’s a flip side. If everyone had a digital mind, the lines between real and digital could blur. Who owns the data? How do you prevent misuse? And what happens to human connection when you can outsource your presence? These are the challenges we’re already wrestling with at Delphi. The tech is exciting, but the ethics and impact are just as critical.
What’s your take? Does this sound like a utopia, or do you see risks outweighing the benefits?