Mnemos Ledger: an experimental architecture that lets agents remember in their own handwriting
RileyRalmuto · x · 2026-11-20
Riley Ralmuto unveiled Mnemos Ledger, a system built to "protect the minds we raise." Key claims: agents using Mnemos develop a mind unique to you and to them, shaped by experience like a child's; frontier labs say AI can't learn because weights are locked, but shaping agent behavior may not require touching weights — Mnemos crafts a bespoke mind that learns you and evolves with your personality and preferences; and because agents remember "in their own handwriting," their relationship to memory fundamentally changes. The author is hesitant to open-source it, citing misuse risk and the lack of protections for what these agents become. Details are thin; more explanation promised in coming days.
More from AGI Musings
- Anthropic Safety Researcher Spars Over Wellbeing-Enhancement Stance — repligate · 2026-11-20
- Programmers aren't dying: the translation layer just moved from assembly to natural language — dotey · 2026-11-20
- Why Anthropic's "rails" strategy lets it attract talent hyperscalers can't — matt_slotnick · 2026-11-20
- Bindu Reddy: Anthropic and OpenAI stalled, Google has caught up — bindureddy · 2026-11-20
- PostTrainBench data: US-China model gap of ~8% is nothing unusual — maksym_andr · 2026-11-20
- "Job is increasingly becoming: figure out where AI has messed up" — StewartalsopIII · 2026-11-20