Research & Learning

AI as a tutor

Getting to grips with an unfamiliar topic is something AI does better than a textbook and faster than googling. Its advantage over articles and videos: the dialogue adjusts to you. Didn't get it — you re-ask; too simple — you ask for more depth; lost the thread — you go back.

Working techniques:

• Set your level. "Explain what options are. I understand how stocks work, but I don't know derivatives" — the explanation lands on your foundation.
• Demand examples. "Every concept with an everyday example." Abstractions without examples don't stick.
• Test yourself. "Ask me five questions on the topic and go over my answers" — active recall cements things better than rereading.
• Ask for a map. "Lay out a study plan for the topic from zero to a confident level: the stages and what to master at each" — then follow the plan in separate chats.

Researching a question

For questions like "which service to choose," "how is market X structured," "what's known about Y" — combine the model's knowledge with web search:

• Facts and numbers — with search. Prices, current versions, events are time-sensitive; without search you get outdated ones (the lesson on the knowledge cutoff).
• Structure and explanations — can go without search. How something works, what approaches exist, the pros and cons — stable knowledge.
• Synthesis — a strong side. "Compare three approaches in a table: cost, complexity, risks" — the model assembles the scattered into a structure.

The deep-research scheme: ask for a research plan → go through the points with search → ask for a summary with conclusions and a list of sources.

Hygiene: don't let yourself be convinced

Two systemic risks of research with AI:

• Adjusting to your opinion. Ask "why is X better than Y?" — the model obligingly makes the case. Ask "why is Y better than X?" — it argues the opposite. Phrase neutrally: "compare X and Y by criteria," and for important decisions — as two separate requests from opposing positions.
• Smoothness instead of truth. A coherent, structured answer creates a sense of reliability. The sense is deceptive — recall the lesson on hallucinations: check key facts against sources.

Learn from your own materials

The combination with the previous lesson — learning from your own documents:

• Upload a scientific paper → "explain it in plain language, then answer questions from the text."
• A lecture outline → "make question-answer flashcards for review."
• A book (chapter) → "the main ideas and how to apply them to [your situation]."

Key takeaways

• The tutor dialogue: set your level, demand examples, test yourself with questions.
• Facts — with web search, structure and explanations — from the model's knowledge, synthesis — its strong side.
• Neutral phrasing against adjustment; for decisions — two requests from opposing positions.
• The smoothness of an answer doesn't equal reliability — check what matters against sources.