How to Ask Better Questions in the AI Age

TL;DR

Asking the right question is becoming more valuable than memorizing answers because AI can perform research and solve many assigned problems. Strong AI users act like leaders: they set direction, identify overlooked issues, ask focused follow-ups, and use interactive summaries to reach the information they actually care about.

Transcript

Let's start with this $20 billion rumored $20 billion partnership that you have with Nvidia. Can you talk about the structure of the deal and how it came about? >> So the most interesting part about it is the call um where the idea was first floated was about 3 weeks before money was in the bank. >> Oh, so Johnson moves fast. >> Of course, that's h... Read More

Key Insights

• The defining skill of the AI age is asking the right question because AI can retrieve information, conduct research, and solve assigned problems. The wording and direction of a question shape what the system investigates and therefore determine the usefulness of its output.
• Effective AI users increasingly resemble leaders because they direct work instead of personally completing every task. Their responsibility is to assess what they hear, establish priorities, and ask the overlooked or uncomfortable question that guides AI toward a valuable answer.
• Interactive learning with AI is more targeted than consuming static content because follow-up questions progressively narrow the subject. Ross compares this process to the game of 20 Questions, where carefully chosen questions distill a large possibility space into the information someone truly wants.
• A personalized daily brief can replace an overwhelming stream of information with concise headlines, supporting research, and relevant links. Ross initially received extensive text, then recognized that AI already held the context and could summarize it while remaining available for deeper follow-up questions.
• GPUs and LPUs are complementary because language-model token processing contains matrix multiplications with different bottlenecks. Compute-constrained operations can run on GPUs, while memory-throughput-constrained operations can run on LPUs, allowing the combined system to address varied performance limitations.
• AI-to-AI communication makes response speed especially important because an AI system does not benefit from waiting as a human might tolerate waiting one or two seconds. Faster processing lets agents initiate research and use other tools concurrently while continuing their primary work.
• Agentic AI works by allowing AI systems to use other AI systems and tools. An agent can begin a research job about tools it may need while using another tool, creating expanding chains of activity that make execution speed and coordination increasingly important.
• Authentic leadership can take many forms because the basic requirement for a leader is having followers. Ross argues that new founders often imitate advice that does not fit them, even though leadership offers many valid approaches and should remain true to the individual.

Questions & Answers

Q: Why is asking the right question important in the AI age?

Asking the right question is important because AI can retrieve information, perform research, and solve many assigned problems, but it still needs direction. The initial question determines what the system examines and strongly influences the resulting output. Human value therefore shifts from remembering answers toward identifying worthwhile problems, exposing overlooked assumptions, and framing precise questions that guide AI toward useful conclusions.

Q: How does AI change the role of individual contributors?

AI moves individual contributors toward a leadership role because they can delegate research and problem solving to AI systems instead of completing every task personally. Their work becomes setting direction, considering the information returned, and asking the next important question. Like strong leaders, they must identify issues that others missed or were afraid to raise and guide execution without supplying every answer themselves.

Q: How can follow-up questions improve learning with AI?

Follow-up questions let a user narrow broad information into the specific details that matter most. Ross compares the process to 20 Questions, where each question reduces uncertainty about what someone has in mind. Unlike a static article or podcast, an interactive AI exchange can respond to curiosity, clarify unclear points, and continue drilling down until the user reaches the desired information.

Q: How did Jonathan Ross improve his personalized daily brief?

Ross initially had AI produce a large amount of text based on his interests, interactions, and research about current events. He then realized the system already possessed the relevant context, so he changed the output to a concise set of headlines with links. Instead of reading everything in advance, he could ask follow-up questions whenever a headline deserved deeper investigation.

Q: Why do GPUs and LPUs work well together for LLM inference?

Language-model token processing requires many matrix multiplications, but those operations do not share a single bottleneck. Some are constrained primarily by computation, while others are constrained by memory throughput. Ross says compute-constrained work can be placed on GPUs and memory-throughput-constrained work on LPUs. Combining the architectures therefore addresses different limitations across the performance curve instead of relying on one strategy.

Q: Why does inference speed matter more when AI uses AI?

Humans may accept waiting one or two seconds after entering a command, but an AI system spends that delay idle even though it generates tokens and processes tasks much faster. Agentic systems can also ask other AI systems to research tools or perform supporting work. Faster inference reduces waiting across these interactions and allows interconnected tasks to proceed more quickly.

Q: How could payments affect the growth of AI agents?

Ross says agent payments are still early because existing payment systems are not designed for autonomous AI activity. If agents could make micropayments within a budget assigned by a human, the number of payments could rise sharply. His phone-number project illustrates the limitation: obtaining services required proving he was human instead of simply letting an authorized agent spend within a preset budget.

Q: What leadership lesson does Jonathan Ross give new founders?

Ross defines the first principle of leadership simply: a leader has followers. He argues that leadership can take infinitely many forms, much like investing can use different stages, instruments, and markets. New founders make a mistake when they copy advice that does not fit their nature. A workable leadership style must attract followers while remaining authentic to the person practicing it.

Summary & Key Takeaways

• Jonathan Ross argues that the transition from the information age to the AI age changes the skill that matters most. Schools trained people to remember information and answer questions, but AI can retrieve knowledge and perform research. Human value increasingly comes from choosing the question that directs AI toward a useful result.

• Ross connects effective AI use with leadership. Good leaders do not personally produce every answer or complete every task. They consider available information, establish direction, and ask important questions that others overlooked or hesitated to raise. As AI handles more execution, individual contributors increasingly become leaders responsible for directing AI systems.

• Interactive AI can turn information consumption into an active process of inquiry. Ross changed his personalized daily brief from a large body of text into headlines supported by research and links. He can then ask follow-up questions, narrowing broad information until he reaches the details that matter most to him.