Testing AI Outputs

How do you check that what the AI gave you is actually right?

Exploration pathCheck the Answer: Evaluating AI Outputs in TreasuryShow path▸

1
Problem
Understand why "it sounds right" is not the same as "it is right," and what changes between a public and a private model.
ARTICLE
Public vs Private LLMs in Treasury
2
Mechanism
Distinguish the three ways an AI answer can fail: unsupported, contradicted, or simply incomplete.
ARTICLE
The 3 Ways AI Works in Treasury
3
Example
Apply a verification grid to a real example, sentence by sentence.
TOOL
AI Output Review Lab
A source document and an AI answer built from it. Mark each sentence supported, unsourced, or contradicted, then check against an editorial answer key.
4
Tool
See what a complete verification method looks like for a live use case, not just a one-off check.
ARTICLE
Episode 2: How to Build an AI Treasury Copilot for FX Risk Monitoring (No Coding Required)
5
Implementation questions
Define your own verification method before you rely on any AI output for this task.
TOOL
AI Use Case Canvas — verification method
Revisit the "how will a human check the output" and "who owns the result" fields — the two questions this whole course is really about.

New to this topic? / Start here

Public vs Private LLMs in Treasury

Why This Matters Right Now? The AI explosion has created two real options for everyone: public LLMs accessed through APIs (like ChatGPT, Gemini, etc) and…

Read the article ↗

Public vs Private LLMs in Treasury

The 3 Ways AI Works in Treasury

AI Output Review Lab

Episode 2: How to Build an AI Treasury Copilot for FX Risk Monitoring (No Coding Required)

AI Use Case Canvas — verification method

Public vs Private LLMs in Treasury

Explore the articles