AnthropicAgent Memory Benchmark Shows Wide Gaps in Continual Learning Performance4 min read·23h ago·via The New Stack