AISelf-modifying neural networks and the prospect of recursive self-improvement
A post links to “A Modern Self-Referential Weight Matrix That Learns to Modify Itself,” quoting the possibility that such networks could, “in principle,” learn to learn recursively.
TLDR
A post sharing the paper quotes a possibility: such neural networks can, “in principle,” “meta-learn to learn,” then “meta-meta-learn to meta-learn to learn,” and so on. The quoted passage describes this as “recursive self-improvement.”
Combined views
1 Source, first seen 1619d ago
Self-modifying neural networks and the prospect of recursive self-improvement
A post links to “A Modern Self-Referential Weight Matrix That Learns to Modify Itself,” quoting the possibility that such networks could, “in principle,” learn to learn recursively.
TLDR
A post sharing the paper quotes a possibility: such neural networks can, “in principle,” “meta-learn to learn,” then “meta-meta-learn to meta-learn to learn,” and so on. The quoted passage describes this as “recursive self-improvement.”