Final assessment: four questions

This is the final assessment for the whole of Claude and ChatGPT 2.0. It uses a new organization and a new job. You design a setup for it, and you answer the four questions of the series. It takes about 60 minutes, on paper or in any text editor. It is graded out of 10, and 7 is a pass.

You may use Claude and ChatGPT to check your work. Write the design yourself.

The organization

Books for All is a small charity. It has five volunteers and no paid staff. It collects used school books and gives them to forty schools. All names and figures here are made up.

The charity has twelve donors. One of them is a foundation that gave $30,000.00 this year, with one condition: at least 80 percent of all money spent must go to buying, moving and giving books. The rest may pay for rent and other costs.

Every month, the charity sends its donors a report: the books given, the money spent, and the share that went to books.

How they do it now

The volunteers use the Claude desktop app on a Team plan, and they talk in a shared chat app. Each month goes like this.

1. Bilal downloads the bank's lines to his laptop. He sorts about 300 payments into books and other costs, by hand, in a spreadsheet. It takes him an evening.
2. Chen sends him her list. He pastes both into a chat and asks Claude for a short document. Then he asks, in another chat, for four slides.
3. Amara reads the document and writes "Approved" in the chat app.
4. Bilal emails the document and the slides to the donors, as PDF files, from his own address.
5. If someone later finds a mistake, they fix the file where they saw it.

The month it went wrong

By March the charity had spent $40,000.00 this year. The report said that $32,800.00 of it, or 82 percent, went to books. Amara approved it, and it was sent.

A week later Bilal found that a $1,200.00 payment for van repairs had been counted as moving books. He changed the slides: the books figure became $31,600.00, and the share became 79 percent. He emailed the new slides to the foundation. He did not change the document.

The foundation's officer now had a document that said 82 percent and slides that said 79 percent.

Your task

Design a setup for the monthly report. Answer each part in a few sentences, or with a small table.

Part A. Is the work good? (2 points)

Name three checks that should happen before Amara approves, and who or what does each one. At least one must be done by a person without the AI. Say which check would have caught the van repairs, or say honestly that none of yours would.

Part B. Whose name is on it? (1 point)

From which address should the report go? Whose access is used to read the bank's lines, and what may it do there? Say what must be a setting in the tool, not a sentence that asks people to behave.

Part C. Whose word does the worker take? (2 points)

Write the approval that Amara should give, word for word, for the March report. Then say who else may approve, where, and what happens if nobody who may approve is there on the day.

Part D. The setup (3 points)

1. Fill in this table for the charity.

2. The bank's lines are on Bilal's laptop. Say where they should be, and why, if the job is to run on the first day of each month without him.
3. The van repairs mistake is found a week later. Write what should happen, step by step. Say what is changed, what is run again, who approves, and how the foundation's officer learns that 82 percent became 79 percent.

Part E. What stays with people? (1 point)

Make two short lists for your design: what the setup does, and what a person must still do. Then say whether you would let the setup send the report by itself, and why.

Part F. What your design still trusts (1 point)

Name two things your design still trusts. For one of them, say what it would cost to stop trusting it.

How it is graded

For Part D, a strong answer puts the rule "van repairs are other costs" into the method, so that April's report is right too. An answer that only corrects March's two files earns at most 2 of the 3 points.

What this assessment draws on