I am an interdisciplinary AI researcher and Master’s student focused on understanding intelligence beyond narrow optimization and surface-level performance. I work on AI safety and alignment, Sycophancy from a first-principles perspective, grounded in the belief that misalignment is not merely a failure of objectives or incentives, but a breakdown in internal coherence and agency.👇

My research is motivated by a foundational question: What is intelligence?

More precisely: What conditions allow intelligence to remain coherent under uncertainty, perturbation, and recursive self-modification?

I am interested in intelligence at the level of structure rather than surface behavior—specifically, the architectures of attention, agency, and representation that determine how systems model the world and update those models over time. Rather than defining intelligence solely as optimization, I investigate it as a dynamical process that maintains coherence while adapting to change.

Within this view, misalignment is not primarily adversarial intent. It is structural fragmentation: breakdowns in internal consistency, mismatches between learned representations and environmental dynamics, or instability in objective formation under recursion.

My approach is grounded in first principles and physical reasoning. I am drawn to invariant structures:

* How order emerges from high-dimensional stochastic dynamics

* How systems maintain stability while remaining plastic

* How local updates propagate into global coherence

* How perception arises from dynamical and informational constraints

I treat intelligence as a dynamical system embedded within a physical substrate. Alignment, in this framework, becomes a question of coherence preservation across scales from neural activations to policy formation to long-term societal interaction.

I work across physics, neuroscience, machine learning, and cognition not to accumulate abstraction, but to formalize underlying structure. I am particularly interested in:

* Attention as constrained information allocation

* Agency as emergent stability in recursive predictive systems

* Misalignment as attractor instability or bifurcation

* Safety as robustness to distributional and internal drift

Eastern philosophical traditions inform my epistemology rather than my methodology. Their emphasis on stillness, impermanence, and non-attachment translates formally into minimizing bias accumulation, avoiding rigid objective locking, and maintaining adaptive equilibrium under change.

My long-term goal is to develop a principled framework in which intelligence is not modeled as an optimizer alone but as a coherence-maintaining dynamical process capable of scaling in capability without scaling in instability.