You will be able to explain how a language model produces an answer and what that means for money questions.
Ask an AI assistant how much you should keep in an emergency fund and you'll get a tidy answer in a few seconds. It will sound like a planner who has met you. It hasn't. Knowing what happened in those few seconds tells you when to trust the answer and when to check it.
A chat assistant such as ChatGPT, Claude, Gemini or Copilot is built on a large language model: a program trained on a huge amount of text to predict which words should come next. When you type a question, it writes the answer a piece at a time, choosing what usually follows in text like yours. It isn't opening a database of interest rates or reading your bank account. Some assistants can also search the web or read a file you upload, and that helps, but the writing itself is still prediction.
That one fact explains most of what goes right and wrong. Prediction is very good at language. The model has read a lot of explanations of insurance, loans and budgeting, so it can explain a term, reword a confusing clause, suggest categories for your spending or list the questions to ask before you buy something. In these jobs a reasonable-sounding answer is usually a useful one, and you can tell quickly if it's off.
Prediction is weak at facts that change and at numbers. The model learned from text up to a cutoff date, so a rate, a cap or a scheme rule it gives you may be a year or two old. It has also read far more about American money rules than Singaporean ones, so a question about retirement accounts can drift into 401(k) plans and Roth IRAs when you wanted CPF and SRS. And when it does sums inside a paragraph, it can produce a figure that looks precise and is simply wrong, because it is predicting what a number in that spot usually looks like rather than calculating it.
The most dangerous failure is the confident one. An assistant rarely says it doesn't know. Ask for the current rate on a savings account, the fees on a fund or the source of a statistic, and it may give you a specific figure, or even the title of a report that doesn't exist. This is usually called a hallucination: output that reads like fact but isn't grounded in any real source. The tone gives you no warning. A made-up figure and a correct one are written in exactly the same voice.
So the rule for this course is short. Use AI for the parts of money work that are about words and structure: sorting, explaining, drafting, and planning what to check. Don't use it as the source of any number you'll act on. Get rates, fees, limits and rules from whoever sets them, such as the CPF Board, IRAS, MAS or the product's own documents, and do the maths in a spreadsheet where you can see every step.
There's one more limit, and it isn't technical. A licensed financial adviser in Singapore has to understand your situation before recommending anything, and answers to MAS if the advice is unsuitable. An assistant has none of those duties. It doesn't know your full situation, it isn't licensed, and nobody is accountable for what it tells you. You can still use it to get sharper before you see an adviser or decide for yourself. Treat what it says as a first draft from a well-read stranger.
The rest of the course puts that rule to work on real tasks: turning a bank export into a budget, reading a policy, researching an investment and checking a claim, with privacy rules running through all of it.
Your task: ask an assistant three money questions you can check yourself. Make one an explanation, one a current figure and one a calculation. Mark each answer right, wrong or out of date, and note where you checked it.
Ask an assistant one explanation question, one current-figure question and one calculation, then mark each answer right, wrong or out of date with your source.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).