Spotkanie dodatkowe AI w praktyce

AI in Practice

The presenter argues that AI can take on substantial work—including research, writing, and computer tasks—but useful results depend less on clever one-line prompts than on giving the AI a clear goal, relevant context, good source material, and repeated feedback.

Practical Prompting and Workflow

• State the desired outcome first. Explain what you want to produce and why—for example, a website intended to sell a workshop—so the AI can plan the work and break it into stages.
• Provide context. Specify the audience, purpose, and your preferred way of learning or communicating.
• Supply reliable material. The presenter says an AI grounded in notes, research, books, consultations, and other selected sources can give more specific answers than one relying on general internet knowledge.
• Set communication rules and provide examples. Define preferred language, tone, vocabulary, formatting, and things to avoid. Examples of your strongest posts or other work can help the AI follow your style.
• Ask for critique that targets a concrete issue. Instead of asking whether a post is “good,” ask where readers are most likely to lose interest, for example.
• Work backwards from the result. Describe the outcome you want, then ask the AI to devise a step-by-step plan, identify where you need to participate, and specify what it can handle.
• Refine repeatable tasks and save successful workflows as “skills.” These reusable instructions can cover tasks such as writing posts, summarizing consultations, formatting carousels, or downloading and processing video material.

The presenter emphasizes that initial setup and correction take time. Repeating a task and correcting the output can eventually make it easier to delegate consistently.

Tools and Use Cases Discussed

• Claude and ChatGPT: The presenter personally prefers Claude for longer writing and tasks that need to reflect his voice. He uses ChatGPT more often for computer or browser tasks and for some large planning jobs. These are his preferences, not an independent comparison.
• Computer and browser automation: He describes using ChatGPT to correct more than 100 invoices and to find and capture images from an application. He recommends installing the relevant browser extension in Google Chrome and initially requiring approval before actions are taken.
• Whisper Flow: A speech-to-text app he recommends for dictating instructions. He says it helps him give detailed prompts quickly and costs him about 30 PLN per month; these are his reported experiences.
• Obsidian: A free note-taking app used to build a linked knowledge base, or “second brain,” from notes, transcripts, research, and selected creators’ material. The presenter recommends Markdown (.md) files as compact, readable source material for AI.
• NotebookLM: Described as a useful but more limited text-focused option. The presenter says running AI locally may offer another approach for people especially concerned about privacy, though it may require a powerful computer.
• Claude Code and ChatGPT Codex: Recommended by the presenter for working with files and projects, even by users who do not code.
• Connectors: AI tools may connect to services such as Gmail, Calendar, Drive, Canva, Zoom, and invoicing software, depending on the tool and enabled integrations.

He suggests using more capable models for planning or research on large projects, then handing the resulting tasks to cheaper or smaller models. He also recommends paid subscriptions as a way to access stronger models, while acknowledging that they are costly.

Accuracy, Privacy, and Delegation

• AI can hallucinate confidently. The presenter gives an example in which an AI incorrectly transferred a concept about the foot to the hand. He advises checking outputs rather than accepting them uncritically.
• Ask the AI to distinguish its own interpretation from claims attributed to an author or study. He cautions that an abstract is not a substitute for reading the full research, and that transcription and rewriting can introduce errors.
• A second review pass—such as asking a fresh chat to check a completed task—can help catch mistakes.
• Be cautious with personal and client data, particularly identifying details. The presenter suggests using anonymized cases and not sharing information that is more private than necessary.
• He distinguishes low-cost errors, such as draft hooks or formatting, from high-cost errors, such as client assessments, professional decisions, or publishing unchecked work. He says AI can assist with professional judgment, but people should retain responsibility for consequential decisions.

Marketing and Social Media

The presenter says it is possible to build a niche audience without following every viral trend. He describes using longer, more technical content to attract the people he wants to work with, rather than relying on oversimplified advice such as generic exercise lists. His broader point is that a creator can shape social media around their own audience and values, even if that approach reaches fewer people.

Guides, Demonstrations, and Examples

The session includes practical demonstrations and guidance on:

1. Structuring prompts around outcome, context, source material, rules, and examples.
2. Dictating prompts with Whisper Flow.
3. Building an Obsidian knowledge base and giving an AI access to a folder.
4. Enabling browser or computer use through Chrome extensions.
5. Creating reusable AI “skills.”
6. Checking AI outputs for factual errors and security risks.

The presenter says he will share the presentation and workshop transcripts, along with additional materials, for participants to use in their own AI workflows.

Main Speakers and Sources

• Main speaker: Wiktor Wrzesiński, leading the workshop and describing his own AI workflows and opinions.
• Audience participants: Several attendees contribute questions or comments, including Judyta, Damian, and Kacper.
• Other people and sources mentioned: Franek Georgiew; marketing practitioners Paweł Tkaczyk and Kamil Kozioł; biomechanics and training educators including Ben Janas and David; and a podcast conversation with Tomek from GSA.