https://cognaptus.com/tags/llm-systems/ Recent content in LLM Systems on Cognaptus Hugo -- 0.145.0 en-us Tue, 01 Sep 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-09-01-not-every-layer-deserves-the-same-cache/ Tue, 01 Sep 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-09-01-not-every-layer-deserves-the-same-cache/ RippleKV shows that long-context quality can improve when a fixed KV-cache budget is allocated according to model-specific layer sensitivity rather than divided uniformly or by depth. https://cognaptus.com/blog/2026-06-05-no-cluster-is-an-island-scaleacross-explorer-and-the-geography-tax-of-ai-training/ Fri, 05 Jun 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-06-05-no-cluster-is-an-island-scaleacross-explorer-and-the-geography-tax-of-ai-training/ How scale-across AI training turns model architecture, parallelism placement, scheduling, and long-distance networking into one business-critical optimization problem. https://cognaptus.com/blog/2026-03-25-relays2s-when-ai-stops-waiting-its-turn/ Wed, 25 Mar 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-03-25-relays2s-when-ai-stops-waiting-its-turn/ RelayS2S shows how real-time voice agents can start speaking quickly without giving up the stronger reasoning of cascaded ASR-LLM systems. https://cognaptus.com/blog/2026-03-24-from-prompts-to-policies-how-digital-twins-are-quietly-rewiring-enterprise-ai-agents/ Tue, 24 Mar 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-03-24-from-prompts-to-policies-how-digital-twins-are-quietly-rewiring-enterprise-ai-agents/ A mechanism-first reading of DT-MDP-CE, a framework that turns messy enterprise agent traces into offline-learned policies for more controllable context engineering. https://cognaptus.com/blog/2026-03-22-act-while-thinking-when-ai-agents-learn-to-multitask-finally/ Sun, 22 Mar 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-03-22-act-while-thinking-when-ai-agents-learn-to-multitask-finally/ A mechanism-first reading of PASTE, a speculative tool-execution system that reduces agent latency by predicting not only which tool comes next, but also how its arguments can be derived safely. https://cognaptus.com/blog/2026-03-21-reflection-in-the-dark-when-prompt-optimization-forgets-to-think/ Sat, 21 Mar 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-03-21-reflection-in-the-dark-when-prompt-optimization-forgets-to-think/ A mechanism-first reading of VISTA, a multi-agent prompt optimization framework that turns reflective prompting from blind rewriting into auditable diagnosis. https://cognaptus.com/blog/2026-03-13-agents-with-memory-turning-execution-logs-into-institutional-knowledge/ Fri, 13 Mar 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-03-13-agents-with-memory-turning-execution-logs-into-institutional-knowledge/ A mechanism-first reading of trajectory-informed agent memory, showing how execution logs can become structured operational guidance rather than decorative vector-store clutter. https://cognaptus.com/blog/2026-03-02-when-less-proves-more-the-case-for-minimalist-ai-theorem-provers/ Mon, 02 Mar 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-03-02-when-less-proves-more-the-case-for-minimalist-ai-theorem-provers/ A mechanism-first reading of AxProverBase, showing why feedback, memory, and lightweight search may matter more than architectural ornament in verifiable AI workflows. https://cognaptus.com/blog/2026-02-28-template-thinking-why-your-next-ai-agent-should-steal-from-cognitive-science/ Sat, 28 Feb 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-02-28-template-thinking-why-your-next-ai-agent-should-steal-from-cognitive-science/ A practical reading of how cognitive models and classic AI algorithms can serve as reusable templates for designing interpretable, task-fit language agents. https://cognaptus.com/blog/2026-02-01-fademem-when-ai-learns-to-forget-on-purpose/ Sun, 01 Feb 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-02-01-fademem-when-ai-learns-to-forget-on-purpose/ FadeMem shows why scalable AI agent memory may depend less on storing everything and more on governing what should fade, merge, or survive. https://cognaptus.com/blog/2026-01-22-many-minds-one-solution-why-multiagent-ai-finds-what-single-models-miss/ Thu, 22 Jan 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-01-22-many-minds-one-solution-why-multiagent-ai-finds-what-single-models-miss/ A mechanism-first reading of why multi-agent LLM systems can improve results without adding information: they factor constraints across agents and stabilize solutions that single update dynamics may not reach. https://cognaptus.com/blog/2026-01-21-when-retrieval-learns-to-breathe-teaching-llms-to-go-wide-and-deep/ Wed, 21 Jan 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-01-21-when-retrieval-learns-to-breathe-teaching-llms-to-go-wide-and-deep/ A mechanism-first reading of ARK, a training-free knowledge-graph retriever that lets LLMs control when to search broadly, when to traverse locally, and when to stop. https://cognaptus.com/blog/2026-01-16-one-agent-is-a-bottleneck-when-genomics-qa-finally-went-multiagent/ Fri, 16 Jan 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-01-16-one-agent-is-a-bottleneck-when-genomics-qa-finally-went-multiagent/ A mechanism-first reading of GenomAgent: why specialized multi-agent orchestration improved genomics QA accuracy while cutting tool-use cost. https://cognaptus.com/blog/2026-01-12-speculate-smarter-not-harder-hierarchical-decoding-without-regret/ Mon, 12 Jan 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-01-12-speculate-smarter-not-harder-hierarchical-decoding-without-regret/ A mechanism-first reading of Hierarchical Speculative Decoding, a lossless verifier that improves LLM inference speed by accepting more draft tokens without changing the target distribution. https://cognaptus.com/blog/2026-01-11-vibe-coding-a-theorem-prover-when-llms-prove-and-break-themselves/ Sun, 11 Jan 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-01-11-vibe-coding-a-theorem-prover-when-llms-prove-and-break-themselves/ Why Isabellm’s real lesson is not autonomous AI reasoning, but verifier-gated system design for domains where being plausibly right is still wrong. https://cognaptus.com/blog/2026-01-07-infinite-tasks-finite-minds-why-agents-keep-forgettingand-how-infiagent-cheats-time/ Wed, 07 Jan 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-01-07-infinite-tasks-finite-minds-why-agents-keep-forgettingand-how-infiagent-cheats-time/ A business-focused reading of InfiAgent, showing why persistent file-based state may matter more than ever-larger context windows for long-horizon AI agents. https://cognaptus.com/blog/2025-12-29-when-bandits-get-priority-learning-under-scarce-tiered-capacity/ Mon, 29 Dec 2025 00:00:00 +0000 https://cognaptus.com/blog/2025-12-29-when-bandits-get-priority-learning-under-scarce-tiered-capacity/ A mechanism-first reading of MSB-PRS, a bandit framework for allocating stochastic capacity when high-priority tasks must be served first. https://cognaptus.com/blog/2025-09-06-cheap-thrills-hard-guarantees-bargaining-with-llm-cascades/ Sat, 06 Sep 2025 00:00:00 +0000 https://cognaptus.com/blog/2025-09-06-cheap-thrills-hard-guarantees-bargaining-with-llm-cascades/ BARGAIN shows how LLM cascades can reduce expensive model calls while preserving finite-sample guarantees on accuracy, precision, or recall. https://cognaptus.com/blog/2025-07-17-beyond-search-rags-awakening-to-enterprise-spreadsheets/ Thu, 17 Jul 2025 00:00:00 +0000 https://cognaptus.com/blog/2025-07-17-beyond-search-rags-awakening-to-enterprise-spreadsheets/ A mechanism-first look at why enterprise RAG needs row-aware tables, hybrid retrieval, reranking, and feedback loops before it can safely answer internal business questions. https://cognaptus.com/blog/2025-07-14-chunks-units-entities-rag-rewired-by-cuerag/ Mon, 14 Jul 2025 00:00:00 +0000 https://cognaptus.com/blog/2025-07-14-chunks-units-entities-rag-rewired-by-cuerag/ Clue-RAG shows that better enterprise retrieval may come less from bigger knowledge graphs and more from using the right semantic layer at the right moment. https://cognaptus.com/blog/2025-04-14-when-streams-cross-wires-can-new-ai-models-plug-into-old-data-flows/ Mon, 14 Apr 2025 00:00:00 +0000 https://cognaptus.com/blog/2025-04-14-when-streams-cross-wires-can-new-ai-models-plug-into-old-data-flows/ A mechanism-first reading of enterprise compound AI architecture, showing why streams, registries, planners, and coordinators matter more than another chatbot wrapper.