Reproduce “Optimisation Is Not What You Need” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alfredo Ibias
Research from Avatar Cognition formally demonstrates that catastrophic forgetting and overfitting are inherent limitations of optimization-based artificial intelligence methods when aiming for Artificial General Intelligence. The work provides theoretical proofs and empirical evidence that "world-modelling" or "representation-based" frameworks can avoid these issues.
Key insights
- Catastrophic forgetting and overfitting are fundamental and unavoidable properties of optimization-based AI methods when pursuing Artificial General Intelligence (AGI).
- Optimization fundamentally aims to perfectly minimize a loss function on specific data, which inherently limits true generalization and lifelong learning.
- "World-modelling" or "representation-based" frameworks can overcome these limitations by fostering non-overlapping internal representations, allowing for continuous learning without forgetting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.