LLM hallucinations come from the interaction of training incentives, probabilistic decoding, and context limits. This report explains how to evaluate factuality and design practical quality assurance.
The cleanest way to reason about LLM options is to ask whether you are changing weights, runtime context, or external actions. This report gives a practical selection guide.
A plain-language guide to tokenization, embeddings, self-attention, pretraining, inference, and context length in LLMs.
This report compares participant, linguistic intentionality, asymmetric communication, and grounding-skeptical accounts of LLMs.
AI coding should be treated as engineering for a higher probability of success, not merely as increasing the amount of work delegated to an agent.
We will discuss what will be gained and what will be lost when generative AI is used to turn tacit knowledge into explicit knowledge, based on the limitations of SECI and LLM.
We will organize LLM, symbolic grounding, Ontology, Graphiti, and MCP as materials for making decisions about the utilization of organizational knowledge.