The Evolution of AI Research - from autoresearch to active inference researcher скачать в хорошем качестве

The Evolution of AI Research - from autoresearch to active inference researcher

I'm developing a next generation AI agent that does research, inspired by Andrej Karpathy's autoresearch: Find the first implementation here: https://github.com/ErikDeBruijn/autor... How do AI researchers evolve from clever tools into genuine thinkers? This explainer traces the leap from Andrej Karpathy's autoresearch — an AI agent that tirelessly runs experiments in a simple loop — to a next generation that actually understands why results work. The core problem: current AI researchers suffer from total amnesia between sessions. Auto Researcher 2 fixes this with a generative world model that learns hidden causes, a persistent memory system that prevents redundant experiments, and a curiosity module based on active inference that picks the most learnable experiment — not just the most novel. The result: a shift from score-chasing to building real research intuition.