The author introduces SelMem, an early-stage experimental project written in Rust that explores whether LLM agents can develop persistent behavioral divergence through a reconstructive memory system. This system allows agents to forget, distort, and consolidate experiences over time, potentially leading to different internal narratives even with identical histories.
- The project is currently in a very early stage with no serious benchmarks yet.
- It aims to evaluate how selective memory architectures influence agent personality development.
- Feedback is specifically requested regarding evaluation methods and comparisons to existing memory architectures.
The author considers this work important for understanding the long-term effects of memory mechanisms on LLM behavior, though it remains a preliminary exploration rather than a finalized product.