Topics — The twenty-eight kinds of claim this anthology files against — the primary axis of both the practice registry and the reading list, and the same twenty-eight for each (ADR-026). The filing rule is the point: take a domain topic when the claim is ABOUT the domain; a claim merely discovered in one still takes its kind. The axis organizes the record, it does not bound it: content the list cannot place is evidence the list is short (ADR-059).
Agents and environments — systems that perceive, act and are changed by the consequences — percept–action loops, world models and the environments agents are trained in, multi-agent and collective behaviour, and what acting rather than only predicting changes about learning, memory and their costs.
6 of 737 LIT documents. Back to the full index.
| # | Title | Summary | Status |
|---|---|---|---|
| LIT-502 | MemGraphRAG: Memory-based Multi-Agent System for Graph Retrieval-Augmented Generation | Wu, Xiang, Tang, Chen, Zhang and Su (2026), ARXIV-2606.00610. A multi-agent GraphRAG system, filed for its pilot study rather than its system: across three published GraphRAG pipelines, graph expansion raises retrieval recall (GFM-RAG 84.3% against vanilla RAG’s 71.8%) and drops relevance (38.5% against 62.9%), and end-task accuracy falls with it. Their own filtering probe finds 40% of extracted triples can be deleted without cost. | Active |
| LIT-697 | AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society | Piao et al. (2025), ARXIV-2502.08691. The paper presents a simulator of about 10k LLM-driven agents in a realistic urban, social and economic environment. It claims that the simulator’s outcomes on polarization, inflammatory-message spread, UBI, hurricane mobility shocks and urban sustainability qualitatively match real-world empirical findings, which would make it usable as a testbed for social science. | Deferred — seeded from the abstract and a skim on 2026-09-25; not read in full |
| LIT-742 | The Computational Foundations of Collective Intelligence | Pilgrim et al. (2025), ARXIV-2509.07999. A conceptual framework with no new data or model. A collective differs from an individual in its computational resources, which are sensing, state (memory), processes and actions, and in the constraints that distribution imposes on them: coordination and cooperation. Wisdom of the crowd, collective sensing, division of labour and cultural learning are each read as one of those resource advantages being put to use. It is illustrated with golden shiners, ant nest-site choice and pigeon homing. | Active |
| LIT-756 | The Work Capacity of Channels with Memory: Maximum Extractable Work in Percept-Action Loops | Fiderer et al. (2025), ARXIV-2504.06209. A theory paper about agents modelled as finite-state hidden Markov channels interacting with an environment. It defines a channel’s work capacity, the best rate at which any agent can extract work from it under the second law. When the agent’s actions do not affect its percepts, the efficient agent is the maximally predictive one, as earlier results showed. Once actions feed back into percepts, that is no longer true. Theorem 3 gives an environment where, for every memory size, the work-efficient agents, the predictive agents and the agents that randomize their actions and forget them are disjoint sets. | Active |
| LIT-759 | From monoliths to modules: Decomposing transducers for efficient world modelling | Boyd et al. (2025), ARXIV-2512.02193 — a formal paper with no experiments. A world model is a transducer (a stochastic input-output machine with latent memory, generalizing POMDPs), and composing transducers is a Kronecker product of their operators. The paper inverts this. Two information measures, Intransducibility (with latents) and Acausality (observables only), are zero exactly when one block of variables can be a causal module driven by the rest. Peeling such modules off factors a monolithic model into a DAG of sub-transducers that can be learned and inferred separately. Minimal predictive models (ε-transducers) are closed under composition. | Active |
| LIT-787 | One Step Diffusion via Shortcut Models | Frans, Hafner, Levine and Abbeel, UC Berkeley (2024), ARXIV-2410.12557. A flow-matching network also conditioned on step size d, trained in one run: the flow-matching loss at d = 0, and a target for step 2d built from two of its own d-steps. On DiT-B at matched compute, one model gives 6.9 / 13.8 / 20.5 FID at 128 / 4 / 1 steps on CelebA-HQ-256 and 15.5 / 28.3 / 40.3 on ImageNet-256. Progressive distillation is better at one step (14.8 and 35.6) and cannot sample at many. The re-run consistency baselines are far below their published numbers. DiT-XL reaches 10.6 in one step. One run per cell. | Active |