Some critical considerations when creating an AI knowledgebase:
1. Confidentiality: What data can be shared publicly, vs. internal-only, vs. with a particular customer/prospect/partner, or only with an enumerated list of individuals? How do you segregate or redact your knowledgebase? "You cannot share this pre-release info unless we have an NDA in place."
2. Temporal authority: what documents are pre-release, and should be known internally but not *yet* externally? What docs are obsolete and should be deprecated for use? How does the AI know what is current? "This is no longer true, but was true 2017-2023. Keep it around for historical purposes..."
3. Factual authority: How does the AI know that this doc is from an actual, factual customer case, or a working, in production, product capability/feature, versus a hypothetical user story or a not-yet-in-production feature request?
This is where specialized training, fine-tuning, and evaluation become *vital.*
What’s a knowledge base? And why do you need one?
#coding #programming #aicoding