Thinking - And Re-Thinking - Context In AI

Blitzy’s Jake Sortor argued that for AI—especially in defense—context is decisive, not just the size of an LLM’s token window. He said intelligence failures such as misclassification and “update failures” come less from lack of data and more from systems being unable to correctly classify and refresh their understanding of surrounding information. Sortor warned that “more context is not always better,” since excessive material can bury important signals and harm performance. He called for intentional “context engineering,” including deliberate choices for retrieval and filtering, and warned that treating context as an automatic byproduct of larger models could repeat past intelligence mistakes. He framed context design as a national security imperative.