Motivation
World Learning Algorithm
Original publication date: May 11, 2025
Relevance: This is the theory that inspired the current ‘Theory Learner’ prototype defined in the ‘Aim of Science’ essay. It defines a distinct Critical Rationalist (agreement-based) learning algorithm and a concept for optimizing an LLM to generate and observe better, more useful theories of the physical world.
Explanatory World Model Builder
Original publication date: Dec 26, 2024
Relevance: A proof-of-concept automated learner designed around a Popperian framing of science. The system learns by automatically querying an LLM that updates a Python knowledge file; objects in that file are composed, computed, and passed back to the LLM to evaluate consistency with reality. It gradually builds a formal model with maximum consistency against the knowledge encoded in the LLM, demonstrating agreement-based selection of better theories of the world. This prototype did not learn useful knowledge on its own, which led to the World Learning Algorithm theory.
Project GitHub: https://github.com/tiplur-bilrex/ewm_learner
Applied Fallibilism Concept
Original publication date: Dec 3, 2023Original essay on Urbit
Relevance: A blog series where I first tried to define and contrast Popper’s framing of science, his critique of inductive probability, and the progress of probabilistic machine learning. I then proposed a toy Popperian science-learning prototype that led directly to the Explanatory World Model Builder proof of concept.
