AI Capabilities And Limitations
← Hoch {lessons}
{lesson} 06AI Capabilities And Limitations

Study Notes

wab

wab — yIlaD.

{study notes}
  • Qapla' Data: Model's knowledge = training data only. Knowledge cutoff = hard stop. No real-time {browsing}, no lived experience.
  • Capability Zone: Topics frequent in training, recent (within cutoff), consistent across sources. Colleague-level knowledge.
  • Limitation Zone: Rare topics, post-cutoff events, niche/local content, contested subjects. Model struggles here.
  • Knowledge Failures: Staleness (outdated info), uneven coverage (gaps), inherited bias (wrong "defaults"), source amnesia (can't cite origins).
  • Solutions - Tools: {Web search}, {retrieval/RAG}, {tool use} = features that patch knowledge gaps by accessing post-cutoff information.
  • Outsider Test: Map your domain—identify mainstream vs. niche topics. Probe coverage, staleness, default assumptions. Flag tasks: model-knowledge or context-required.
  • Delegation Strategy: Know where model is strong vs. thin. Hand off tasks accordingly, supply context yourself when needed, or seek external sources.
{flashcards} 10 {cards}
{question}
yI'ang · ←/→
{answer}
{flip}
Anki (.tsv) ↓
{knowledge check} 6 {questions}