AI rewards deep knowledge: better questions come from what you already know

Many people now ask ChatGPT or Claude whenever a question comes up at work, even about complex legal provisions or accounting standards. So do we still need to keep expert knowledge in our heads? The author's answer is yes. AI rewards people who can ask good questions, and good questions come from deep knowledge. This guide is for professionals in accounting, legal, HR and management roles, including foreigners working in Korea.

Your question sets the ceiling

Take accounting. An employee with basic knowledge asks an AI: "How do I account for trade receivables?" An expert asks: "Under IFRS 15, how should I identify performance obligations and allocate the transaction price in a bundled contract that includes variable consideration, and what practical approach works?" The same tool gives very different depth.

IFRS 15 (Revenue from Contracts with Customers) applies to annual periods beginning on or after January 1, 2018. Korea adopted it as K-IFRS 1115 (기업회계기준서 제1115호, 고객과의 계약에서 생기는 수익). Its five steps are: identify the contract, identify the performance obligations, determine the transaction price (including any variable consideration), allocate that price to the obligations, and recognize revenue as each obligation is satisfied. Only someone who knows this structure can ask the expert's question.

Legal work is the same. Asking for "points to check in a contract" gives a generic list. Asking how to choose the governing law and set the dispute resolution clause in an international contract, in light of recent case law, gives a far more useful answer.

A podcaster who prepares like a student

Dwarkesh Patel hosts the Dwarkesh Podcast. He was 23 when Mercury's Meridian magazine profiled him in February 2024, and his guests have included Meta CEO Mark Zuckerberg, Google DeepMind CEO Demis Hassabis and Ilya Sutskever, then OpenAI's chief scientist.

His method is preparation. According to the Mercury profile:

• For his interview with DeepMind's chief, he said he would have read most of DeepMind's papers from the previous couple of years.
• He talked to a dozen AI researchers to prepare for one interview.
• He implemented a transformer, the model design behind today's language models, before interviewing Sutskever.
• His aim is to cover the essentials of an undergraduate course on a topic before he interviews an expert in it.

As a result, his questions go well beyond the surface.

Learning compounds

The author's central point is that learning works like compound interest. An accountant with solid fundamentals adapts quickly when a new standard arrives. Legal staff who understand the structure of the Civil Act can connect a new court decision or amendment to what they already know. HR staff studying a labor law amendment can use AI to lay out the differences from the old rule and test how it would apply in practice.

The online publication Every described Patel in July 2024 as using AI tools to "compound his own intelligence". Two of his habits stand out:

• AI as a memory aid: he asks a language model to turn material into question-and-answer pairs and to quiz him, a form of spaced repetition inspired by researcher Andy Matuschak, whom he had interviewed.
• AI as a map of the unknown: when Claude had no good answer to a question, he took it as a sign that the question was worth asking his guest, the geneticist David Reich.

The author gives a workplace version: if an AI cannot give a consistent answer on how ESG management relates to financial performance, that topic needs deeper research of your own.

Study methods that work

Learning research supports two habits in particular. A 2013 review by John Dunlosky and colleagues in Psychological Science in the Public Interest rated practice testing and distributed (spaced) practice as the most useful study techniques. Rereading, highlighting and summarizing were rated low.

The author's suggestions:

• Space your review. An accountant can revisit key standards each week and apply them to real cases. Legal staff can review important court decisions regularly to follow how interpretations change.
• Use questions and answers. Ask AI to turn your notes into a Q&A set, then answer it yourself. An HR manager introducing a new system might ask: "How does this link to our current performance review?"
• Revisit core terms daily. A finance or management support employee can review budgeting principles and investment decision criteria until they become second nature.

Using AI to think better

• Give context. Share the background, the contract terms and the rule involved. An accountant asking about a transaction should describe the deal, the contract conditions and the relevant standard.
• Check your reasoning. After reviewing a contract, ask the AI whether your interpretation holds and what risks you may have missed.
• Connect fields. An HR manager improving the culture can combine ideas from psychology and management with AI's help.
• Build your own framework. Collect what you learn into a consistent way of making decisions.

Growing as an expert

AI now handles search and basic analysis, but that does not lower the value of expertise. An accountant can review routine entries faster and spend the time saved on strategic financial analysis. A legal expert can let AI draft the first version of a contract and focus on risk analysis. An HR expert can draft basic policies with AI and spend more time on culture and talent development.

In the author's words, AI does not replace knowledge. It makes people who have knowledge stronger, because deep basics let them ask sharper questions and judge the answers.

Checklist

1. Before asking AI, write down what you already know and what you need to find out.
2. Review key concepts on a schedule rather than rereading once.
3. Turn your notes into questions and test yourself.
4. Give AI full context, and check its answers against the original source.
5. Treat weak or inconsistent AI answers as research leads.
6. Build your own decision framework from what you learn.