Some Thoughts on Irrational Encodings and Agent Memory
Encoding finite information in an irrational number is possible; the harder questions concern storage, selective retrieval, and what an agent can actually read.
Lines of words light up one by one; only the night ever answers.
Arm I · 10
Encoding finite information in an irrational number is possible; the harder questions concern storage, selective retrieval, and what an agent can actually read.
A reflection on separating facts, experiences and procedures, then deciding what an agent should store, retrieve, revise or forget under a finite context budget.
Persistent memory can carry an untrusted observation into later tasks. Retrieval quality therefore needs to be considered alongside provenance, permissions, validity and revocation.
Could graph structure guide hierarchical summaries? A proposal that distinguishes central entities from summary nodes and preserves relational evidence alongside a traceable abstraction layer.
Starting from iterated encoding and decoding, distinguish local rotation from global topological obstructions and examine what potentials, universal covers and Hodge decomposition can explain.
A route from Basic MC, Exploring Starts, and ε-greedy control to initial distributions, biased random walks, hitting probabilities, and natural starts.
Separating attention modules from attention functions, then exploring a constrained route through excitatory–inhibitory competition, spike measures, and causal operators.
Separating data, token, and reasoning-trajectory sampling in large models, and asking what a Gibbs energy view can genuinely contribute.
Projecting stochastic-gradient trajectories into ordinal patterns, and examining what fence posets may reveal—and fail to reveal—about local optimization dynamics.
As models learn to plan and use tools dynamically, workflows do not disappear; they become executable, recoverable, and verifiable control structures inside a harness.
Arm II · 2
Two curves fit ten points. Which one should we trust on the eleventh? Free energy connects the weights we give models, generalization bounds, and the solutions training actually reaches.
Prisoner's Dilemma, concept distillation, and social learning offer a starting point for understanding how partner selection, networks, and exploration shape cooperation.