The coolest part about working with models at a core level is constructing how memory is stored and retrieved.
Imagine, for a moment, that you could see the strands of thought that comprise how you perceive the world. It’s not a single thread but a web of interlocking prompts, experiences, beliefs, and reactions woven together over a lifetime.
Much like a growing model, you began as a clean slate. Some were fortunate to have exceptional teachers or stable families; others were not so lucky, shaped instead by scarcity, broken systems, or poor nutrition, each variable influencing how they interpret the world and make decisions.
Now imagine being able to design your own mind, to take a snapshot of your worldview, biases, and blind spots, and see them laid bare like code ready for debugging. To walk back to the origin point of a prejudice, an assumption, or a trauma, and retrain it. That’s the frontier of AI: not generating text, but reconstructing how we think.
Most people only see the surface, the party tricks of AI, the emails, the viral videos, the illusion of intelligence. But the real work is in metacognition: teaching systems not just to store and retrieve, but to understand why something was stored, what context shaped it, and when it should be revised.
We’re building machines capable of reflecting and correcting their biases, while our societies remain unable or unwilling to do the same.
That’s the paradox. We automate learning but neglect wisdom. We train fairness in code but tolerate injustice in life.
If AI evolves as a mirror of humanity, it will not just reproduce our brilliance; it will also expose our blindness. The question isn’t whether machines can think like us. It’s whether we can learn to think about ourselves with the same precision we demand from them.