This report explains why hallucinations happen from training objectives and inference behavior, compares benchmark evaluation with practical evaluation, and outlines what RAG, verification, citations, and tools can and cannot fix.
ai-systems
View All africa-history ai-market ai-systems enterprise-ai-platforms developer-tools knowledge-systems philosophy-knowledge learning-science data-infrastructure tech-news semiconductor-memory macro-finance middle-east geopolitics country-profiles uk-politics uk-history uk-culture japan-politics real-estate ancient-history cultural-anthropology music-culture global-popular-music japan-religion welfare-policy global-religion
A practical guide to the difference between pretraining, fine-tuning, prompting, RAG, tool use, and agents, framed by what each one changes and when to use it.
A practical introduction that breaks LLMs down into tokenization, embeddings, Transformer self-attention, pretraining, next-token prediction, inference, and context length.
We will organize the basics of the SECI model and tacit knowledge, and summarize from a deployment perspective what generative AI is good at and what tends to distort summarization, translation, and knowledge management.
A practical report that organizes the design theory, implementation decisions, and major alternative services for long-term memory that combines Graphiti's temporal knowledge graph and MCP.
An integrated report that connects the history of AI research, LLM, symbolic grounding, Ontology, Graphiti, and MCP to practical decisions for utilizing organizational knowledge.