As Claire Grouze wrotes, "The semantic layer is presented as the holy grail to make your analytics agent reliable." This is the 'knowledge base' we suppose that intelligence and reasoning works from, filled with ontologies, causal principles, and the like. The stuff we think it takes to 'understand' a domain. But the semantic layer is brittle and context-dependent. When used on its own in this test, it refused to respond to most questions. Even optimized - at the cost of much more work and slower processing speed - it still answered few queries. And, as Grouze writes, it doesn't eliminate hallucinations - it just moves them. This sort of finding is why I think teaching recognition skills is more important that teaching facts and principles. Via Isin Pesch.
Today: Total: [Direct link] [Share]
