# Microsoft Taught Models to Forget Their Own Thoughts. The Ghost Traces Stayed.

- slug: microsoft-taught-models-to-forget-their-own-thoughts-the-ghost-traces-stayed
- date: 2026-04-10
- category: Artificial Intelligence

Microsoft Research paper Memento teaches models to compress their own chain-of-thought mid-generation, cutting KV cache 2-3x and nearly doubling throughput. Key twist: erased reasoning blocks leave traces the model still uses.

---