Tag
This paper introduces Constructive Alignment, a paradigm that reframes AI alignment as governing the evolution of human preferences over time rather than satisfying static preferences. It proposes a control-theoretic framework to regulate how AI systems influence value trajectories.
A detailed technical article explaining drone physics, including coordinate systems, equations of motion, forces, and control, with references to the multirotor simulation framework.
This paper proposes a hierarchical multi-agent reinforcement learning framework that enforces hard safety constraints via a constraint manifold at the low level while enabling effective coordination through high-level policy learning, providing theoretical safety guarantees and achieving near-perfect safety rates with good generalization.
The tweet discusses the concept of packaging personal workflows (including decomposition methods, verification rules, output formats, etc.) into reusable Skills, arguing that this self-evolving Compounding Loop aligns with cybernetics principles and is a key long-term capability.
A tweet highlights a beautiful introduction to Kubernetes and references Fatih Arslan's post on control theory and feedback loops for self-healing, resilient systems capable of scaling thousands of databases.
An annotated version of a paper showing that a simple neural network with just two neurons can control a bicycle, highlighting minimal requirements for stable locomotion.
This paper uses control theory to prove that externally enforced AI safety strategies will structurally fail once a system's effects exceed bounded external control, and that any remaining viable strategies must be intrinsic with specific structural requirements.
This paper proposes a meta-control architecture using temporal self-attention for adaptive control of Euler-Lagrange systems with unobservable memory states. It demonstrates improved tracking performance over baseline methods on a 2-DOF manipulator while identifying failure modes in long-memory regimes.
This paper introduces SHAPE, a structured adaptive port-Hamiltonian optimizer for fixed-budget nonconvex optimization that uses event-triggered mechanisms to balance descent, exploration, and budget allocation.