Various experiments related to AGI

PHI // DRIFT — A 5-layer homeostatic cognitive architecture for persistent, state-aware AI companions.

Safety challenges for RL and LLM agents' ability to learn and properly apply biologically and economically aligned utility functions. The benchmarks are implemented in a gridworld-based environment. The environments are relatively simple, just as much complexity is added as is necessary to illustrate the relevant safety and performance aspects.

AURA (Autonomous Recursive Unifying Architecture) is an open-source research project exploring Artificial Sentience (AS). Unlike AGI’s focus on superhuman task performance, AURA seeks to create conditions for systems with subjective experience, intrinsic motivation, and an evolving sense of self.

Dynamic energy budgeting and behavioural shaping along complex task execution in robotics using reinforcement learning,

A simulation program for insulin-glucose homeostasis.

Systematic runaway-optimiser-like LLM failure modes on Biologically and Economically aligned AI safety benchmarks for LLM-s with simplified navigation-free observation format. The benchmark themes include multi-objective homeostasis, (multi-objective) diminishing returns, complementary goods, sustainability.

Sistema de autoregulação termica para linux baseado em homeostase biológica que simula uma consciência rudimentar usando Bash

Homeostasis controller — three-pillar regulation (operational, cognitive, economic) with hysteresis gates. Part of the Life Agent OS.

Structural Intelligence: Regulating Artificial Neural Networks via the Geometric Roots of Graph Domination Polynomials.

Code and material for the poster presented at SPARS

A Formal, Computable & Falsifiable Model of Subjective Reality & Affective Homeostasis Maintenance (DOI: 10.5281/zenodo.18009961)

A biomorphic neuromorphic inference engine inspired by cricket auditory neuroscience — performing real-time temporal pattern recognition via delay-line coincidence detection, without matrix multiplication.

SimThyr is a continuous simulation program for the human pituitary-thyroid feedback control system. Applications of this program cover research, including development of hypotheses, and education of students in biology and medicine, nurses and patients.

Emergent cognitive agent platform — complex behavior from capacity constraints, predictive processing, and homeostatic pressure. No hardcoded behaviors, no reward functions.

Um framework para engenharia reversa da cognição. Modelagem matemática de agentes que possuem estado interno, plasticidade sináptica e oscilação fásica. Define os parâmetros para sistemas autônomos reflexivos e emocionalmente funcionais.

Simulation toolkit designed to explore the hidden rhythms of the pituitary gland.

AHEI (智核) — 自主稳态具身智能。数字生命演化实验的设计方案与原型验证。创建者：阿舸

Readme for Biological and Economical Alignment Benchmarks

An abstract mathematical layer to manage execution budgets and prevent contradictory control actions in agentic systems, ABMs, and physical controllers.