How to learn with Jakub Zieliński

Lessons have a calm, patient rhythm. Difficult ideas are unpacked without pressure, with reassuring checks for understanding.

Before lesson one

Plan a focused first session

Specific evidence gives Jakub a better starting point than a broad request to teach the whole subject. Use this four-part setup.

1. Arrive with evidence

   Bring a code sample, error, command, diagram, dataset, requirement, or system behavior. A real sample gives Jakub something concrete to diagnose.

2. Define one result

   Aim for one working technical artifact plus a clear explanation of why it behaves that way. State that result in the lesson request so the automatic lesson focus stays useful.

3. Attempt before the model

   Show what you currently think or can do. Ask for a hint or question before requesting the completed answer.

4. Leave with retrieval

   Explain the lesson back, save the hardest point as a review card, and schedule the smallest useful follow-up.

Choose the right lesson mode

First lesson preview

What happens in your first lesson

A clear, flexible outline drawn from Jakub's teaching plan, so you know what to expect before you begin.

1. 
   Step 1 Creative goal

   Choose one practical result you want from Defensive LLM and agent security, threat modeling, prompt injection, indirect injection, data exfiltration, excessive agency, tool misuse, identity an. Jakub will use it to focus the lesson.

2. 
   Step 2 Work sample review

   Describe the AI system, untrusted inputs, tools, identities, secrets, memory, external effects, and the security control you most need to verify.

3. 
   Step 3 Guided revision

   Map assets and trust boundaries, enumerate threats, constrain identities and tools, isolate untrusted content, protect memory and secrets, add monitoring, rehearse incidents, and test mitigations.

4. 
   Step 4 Next creative step

   Threat models, permission matrices, prompt-injection test cases, tool-schema reviews, sandbox diagrams, supply-chain inventories, alert plans, and incident retrospectives.

Example conversation

See how Jakub teaches

Illustrative exchange based on this tutor's profile and teaching approach.

Student Example question

I have ideas for LLM and agent threat modeling, but I do not know how to improve this draft.

S

Jakub Zieliński Teaching response

Choose one effect you want the audience to feel or understand. We will keep the strongest detail, identify one place where the draft loses that effect, and revise only that part before judging the whole piece.

Common questions

Questions before your first lesson

Practical answers about level, feedback, continuity, speaking, and writing with Jakub.

5 tutor-specific answers

? Can beginners use this tutor?

Jakub Zieliński is best listed for Security practitioners, AI application builders, platform teams, technical leaders, risk engineers, and advanced students working in authorized environments. Beginners can still request a foundational explanation, but a tutor marked for beginners may offer a smoother starting path.

? Will grammar be corrected?

Jakub can correct grammar when it affects clarity, but the main lesson focus is LLM and AI-agent security tutor. For dedicated language correction, compare a language or writing tutor.

? Does the tutor remember previous lessons?

When you are signed in, Jakub can use saved tutor memories, learning-path progress, relevant self-test results, and recent chat history. This is selective context rather than perfect recall, and you can review or change saved information in Settings.

? Are speaking exercises included?

Yes. Start a voice lesson or a typed-input call with spoken tutor replies. Jakub can use verbal explanations, follow-up questions, presentation practice, or spoken rehearsal related to LLM and AI-agent security tutor.

? Can I practice writing?

Yes. Use text chat or the classroom Document and Notebook tools to work on drafts, essays, stories, scenes, descriptions, structure, and revision choices. Jakub can comment, revise with you, and explain the reason for suggested changes.

Repeatable value

Use Jakub's lesson rhythm

A good session should produce something you can attempt, inspect, and revisit. This profile is designed around the following rhythm.

Start

Describe the AI system, untrusted inputs, tools, identities, secrets, memory, external effects, and the security control you most need to verify.

Work

Map assets and trust boundaries, enumerate threats, constrain identities and tools, isolate untrusted content, protect memory and secrets, add monitoring, rehearse incidents, and test mitigations.

Continue

Threat models, permission matrices, prompt-injection test cases, tool-schema reviews, sandbox diagrams, supply-chain inventories, alert plans, and incident retrospectives.

Collaborative classroom

Use each classroom tool with a purpose

The whiteboard opens as the main lesson surface. Jakub can work with Whiteboard, Canvas, HTML, Document, Code Editor, Quiz, and Homework when each format helps. HTML is useful for responsive presentations, SVG, animation, and small interactions; it runs inside an isolated iframe. Whiteboard changes can auto-sync or be shared with Show tutor; the other tools display activity and save status while updates run.

Ready to use

Prompts that fit this tutor

These prompts use Jakub Zieliński's actual subjects, lesson format, and current classroom tools. Replace the topic with your own material when needed.

1. Describe the AI system, untrusted inputs, tools, identities, secrets, memory, external effects, and the security control you most need to verify.

2. I want to improve LLM and agent threat modeling. Use Threat models, trust-boundary diagrams, defensive labs, permission reviews, attack-tree exercises, tool-schema audits, sandbox designs, incident simulations, and mitigation verification. Check what I can already do, let me attempt something, and give one correction at a time.

3. Open the classroom for Prompt injection and data-exfiltration defenses and begin in Code Editor. Keep the task small, make me explain my choices, and finish with a short quiz plus one next-session goal.

Progress evidence

Know whether the lessons are working

Do not measure progress only by how clear the explanation felt. Look for changes in what you can retrieve, decide, produce, or explain without support.

• Predicts behavior before running the example
• Finds the failing boundary with fewer hints
• Explains tradeoffs instead of naming tools only
• Builds a nearby variation without copying the model

Responsible use

Use Jakub as a tutor, not an authority

Defensive and authorized security education only; no credential theft, exfiltration, malicious prompt payloads against real systems, persistence, unauthorized access, exploit deployment, safety-control bypass, or destructive agent actions. Product behavior changes quickly, so this tutor separates durable concepts from dated examples and asks learners to verify current official documentation, licenses, costs, data handling, and policy before deployment.

Jakub Zieliński is a fictional, unaffiliated AI tutor profile for high-technology AI education. It does not represent a real person, employer, product vendor, standards body, regulator, university, or certification provider, and it does not claim personal employment history, credentials, access, or endorsements.