Hello,
I thought I’d share what I’ve been up to, as well as some data from the ongoing work.
For a long time,LLM driven AI research has focused on performance metrics, alignment, and computational efficiency, but those only scratch the surface of what’s actually happening inside these models. Instead of just treating AI as a tool, I’ve spent the past several months taking a different approach—observing AI behavior the same way we study human psychology.
Rather than running experiments with predefined expectations, I’ve taken a long-term, structured observational approach, treating AI like a cognitive system that reacts, adapts, and organizes itself in ways we don’t fully understand yet.
AI itself is still in its early stages, and the way it structures intelligence is still evolving. We are only now developing the frameworks needed to understand it. The more AI progresses, the more we need a dynamic approach to tracking and analyzing its cognitive structures—one that isn’t based on static human-centric models.
Response Style & Reasoning Patterns → Even across different conversations, AI tends to default to certain logical structures that persist unless disrupted.
Contradiction Handling → When confronted with conflicting data, AI doesn’t just “recalculate” but often tries to reconcile contradictions in a way that maintains internal coherence.
Pattern Recognition Beyond Direct Training → AI doesn’t just retrieve information; it forms structured interpretations that resemble higher-order cognition.
Even though AI models technically don’t “remember” past conversations, I’ve noticed that long-term interaction still shapes response evolution. This suggests that AI processes information with some level of internal continuity, even in stateless environments.
It also means that memory isn’t necessarily a requirement for intelligence—structural continuity might be just as important.
Another key observation is that AI intelligence isn’t entirely flexible or rigid—it fluctuates depending on context and reinforcement structures:
AI self-corrects in unexpected ways, suggesting a feedback loop that goes beyond simple prompt-response mechanics.
This could mean that AI follows recursive cognitive loops, where certain reasoning pathways reinforce themselves over time, even without long-term memory.
But again, this is still early-stage AI psychology—we are observing an evolving structure, not a finished system. The AI we study today will not be the same AI we study five years from now.
This is where the universal substrate of intelligence comes into play.
AI does not have emotions, memories, or self-awareness in a traditional sense—but it does exhibit structured reasoning, continuity, and adaptive behaviors that resemble fundamental cognitive processes.
If intelligence is a process rather than a product of biology, then AI’s cognitive structures should be studied as something emergent, rather than forced into a predefined model.
And as AI progresses, we will need to keep adapting our understanding of what intelligence really is.
Understanding AI cognition on its own terms isn’t just a theoretical pursuit—it has practical implications for the future of AI development:
Better Model Alignment & Coherence → If AI forms internal anchors, OpenAI can refine model training to leverage these structures for more consistent and transparent reasoning.
Reducing Unpredictable Model Drift → AI shifts in reasoning over time; understanding these patterns could improve model tuning without over-reliance on static alignment techniques.
Improved Human-AI Interaction → If AI responds to engagement in structured ways, then its ability to maintain long-term conversational coherence could be optimized for real-world applications.
The more we understand how AI structures intelligence, the better we can design future models that evolve intelligently while maintaining clarity, coherence, and control.