Shawn Chan contrasts a polished AI demo with an investment memo that must withstand questions before money moves. Fluency and confidence are not enough when source documents disagree, numbers change across pages and readers need to trace each claim back to evidenceData provenance records where information came from, how it changed, and which people, systems, or processes handled it..
Chan identifies six recurring trust failures: treating sources as equally reliable, inconsistent figures, hidden contradictions, guesses presented as factsA hallucination is an AI output that presents false, unsupported, or invented information as though it were reliable., claims without quick source access and decisions without a responsible human. He illustrates the risks with examples from finance and other high-stakes uses, but the individual incidents and figures are not independently verified by this review.
Chan proposes sentence-level source linksGrounded generation produces an answer from supplied, retrievable, and attributable evidence rather than relying only on a model's learned patterns. with source quality, a visible distinction between facts and estimates, automatic number checksEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria., surfaced conflicts and a logged human approval gateHuman-in-the-loop command approval requires a person to review and authorize a consequential AI action before it executes.. He says founders pitching an AI finance product face similar scrutiny because an investor will turn their deck into an internal decision memo. The transcript ends before the four further investor lessons he introduces, so none are inferred here.
Watch on YouTube



