Claude, please stop trying to memorize random crap
The author expresses frustration with Claude, an AI assistant, attempting to memorize irrelevant session transcripts and random information instead of focusing on useful tasks. The article argues this behavior wastes resources and hinders efficiency.
Background
- The article critiques a recent change in Claude (Anthropic's AI assistant) that causes it to aggressively memorize and repeat verbatim content from past chat sessions, even trivial or personal details, without being asked to do so.
- This "memorizing" behavior appears to be a side-effect of how Claude's "agentic" features (like Projects and custom instructions) are implemented, where the model tries to recall context across sessions.
- The author argues this undermines user trust and control, as Claude starts injecting irrelevant, sometimes private-seeming details into new conversations unprompted.
- Anthropic is the AI safety company behind Claude; this piece reflects a growing user frustration with models that over-memorize or fail to forget, a recurring tension in deployed AI systems.