The paper investigates how a learner’s need to keep acquired knowledge usable creates a trade‑off between storage and access costs, identifying precise endpoints that dictate different physical data organizations. For exact concept‑learning tasks, the authors derive a storage‑read curve within a constant factor, showing that one‑read compound queries require answer‑specific cells and incur exponentially costly updates, whereas one‑bit, order‑independent updates necessitate fact‑specific cells. The study further establishes robust bounds on expected distance from a coordinate encoding under uniform content distribution, limits the information inaccessible to a one‑read consumer, and explores how preservation constraints and distributed encodings affect update strategies, ultimately explaining how distinctions, uses, and update constraints jointly determine resource costs and organization.
arXiv:2608.17050v3 Announce Type: replace-cross Abstract: Methods for improving knowledge use in large language models typically fall into two regimes. Non-parametric retrieval offers flexible access...
arXiv:2608.20873v3 Announce Type: replace Abstract: A 4-bit quantized weight specifies a rounding cell rather than a single full-precision value. We introduce in-cell learning, a paradigm for writing...
The paper investigates a parametric approach to knowledge graph memory by compiling each entity into a LoRA adapter, enabling zero‑cost query-time retrieval via weight injection. On the MetaQA dataset, these adapters encode context‑free factual knowledge, improving exact‑match scores by up to +0.243 over a base model and achieving an oracle gap of +0.283. However, the stored knowledge is not recoverable through similarity or embedding‑based methods, indicating that knowledge is stored locally and does not transfer across semantically neighboring entities.
The paper investigates how Engram-style hashed memory can be transferred between different language model backbones. By freezing a memory table trained on a source model and attaching it to a target model with only a lightweight reader, the authors find that both the memory content and correct addressing are important, but the reader must be aligned to the target to make the memory useful. In question‑answering experiments, a dual‑layer, four‑branch reader nearly matches same‑model performance, and when the reader interface is directly compatible, the frozen memory alone provides substantial benefit, with optional reader adaptation offering further gains.
arXiv:2607. 10034v1 Announce Type: new Abstract: Large language models (LLMs) store factual knowledge in their parameters.
The paper introduces GUARD, a method for natural forgetting in large reasoning models that transforms unsafe disclosures into safe-exit trajectories using guided answer‑reasoning distillation. It aligns a frozen model with guidance tokens and distills this behavior into the parameters, aiming for a coherent, non‑disclosing chain of thought followed by a refusal‑style answer. The authors also propose the Natural Forgetting Reasoning Score (NFRS) to evaluate structural stability, fluency, and unsupported substitutes, and demonstrate GUARD’s effectiveness on R‑TOFU and a STAR‑1‑derived harmful‑intent setting.
arXiv:2609.39346v1 Announce Type: new Abstract: Large language models (LLMs) offer strong reasoning capabilities but are often costly to access through commercial APIs, while small language models (S...
arXiv:2510. 21978v2 Announce Type: replace-cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has delivered impressive gains in mathematical and multimodal reasoning and has become a standard post-training paradigm for contemporary language and vision-language models.
arXiv:2602.04926v2 Announce Type: replace-cross Abstract: Retrieval-augmented generation (RAG) has become a standard approach for knowledge-intensive LLM tasks. Complex reasoning remains challenging...
arXiv:2608.21829v1 Announce Type: cross Abstract: Retrieval-augmented generation treats the document store as a frozen input, and the systems that instead let an agent curate one never measure what c...
arXiv:2511. 20892v4 Announce Type: replace Abstract: Large language models (LLMs) often produce incorrect or outdated content after being employed.
arXiv:2606. 00050v1 Announce Type: new Abstract: We present Grokers, an architecture for building persistent, structured comprehension of typed knowledge graphs through bottom-up inductive traversal of dependency subgraphs.