AI doesn't know how to forgive and cannot forget
Summary
An in-depth look at AI memory architectures, detailing four memory layers (weights, context window, retrieval store, logs), how forgetting happens in machines, and why there is no forgiveness operation in current systems. The piece argues that forgetting is an active, adaptive process in humans and that memory in AI is read-only, with implications for privacy, data retention, and how we deploy AI systems.