Researchers probe how AI models form representations
Mechanistic interpretability is giving researchers a closer look at internal processes in large language models and how conceptual representations form.
Researchers probe how AI models form representations
Studies report features representing concepts and relationships. Grokking describes a shift from memorizing examples to generalizing rules. These findings do not prove human-like understanding.
Mechanistic interpretability is giving researchers a closer look at internal processes in large language models and how conceptual representations form.
