Q&A #1. Prompting 101. NotebookLM brief overview.

00:05 – 04:50 — Daria's landing page build story Daria shares how she initially struggled with Claude design for a website, then discovered 21st.dev (a components library), copied prompts into Lovable, and ended up with a landing page her bosses said looked better than the company's main site. Took about a couple of days. 04:50 – 05:30 — Transition to Prompting 101 Anna moves into the lecture portion since there are no more questions. 05:30 – 09:00 — Prompting fundamentals & Anthropic's approach Disclaimer that the fastest way is often to dump your thoughts and ask AI to write the prompt, but today she'll cover Anthropic's structured method. Context is what matters most. Opus 4.7 needs more precise prompting than earlier models. Iteration and testing are core. 09:00 – 12:30 — Playground demo & temperature Walkthrough of Anthropic Workbench and OpenAI Playground. How temperature works (0–0.3 for predictable/code, higher for creativity but more hallucination risk). Choosing models, thinking modes, token limits. 12:30 – 16:00 — Anthropic's prompt structure Layered approach: role + high-level task → context → examples. Walking through the detailed structure guide Anna built. Examples matter a lot. 16:00 – 18:30 — Thinking step-by-step Less critical with mature models, but still valuable for product development and processes with defined stages. Swedish insurance form example from Anthropic. 18:30 – 19:30 — XML tags for organization Helps AI scan context in a more token-efficient way, especially for longer prompts. 19:30 – 22:00 — Preventing hallucinations Telling Claude to say "I don't know," using low temperature, "think before answering" instructions. Acknowledgment that hallucinations are inherent to transformers. 22:00 – 23:00 — System prompts vs. regular prompts When to put context in the system prompt (anything true across every thread, like a recurring form structure or your professional role). 23:00 – 23:45 — Extended thinking Useful for complex tasks; can over-engineer simple ones. 23:45 – 28:50 — NotebookLM demo & live AI coach prompt update Anna shares her prompting guide built from Sander Schulhoff's content (Learn Prompting founder). Walkthrough of flashcards, quiz, audio overview features. Live test: feeding her existing AI coach prompt + the guide + Anthropic source, asking for improvements. Notes Mike Taylor's insight about avoiding ChatGPT memory for precision tasks. 29:09 – 33:10 — Daria's Q&A: which prompting approach is better? Daria asks whether free-form prompting (just describing the situation) or structured prompting is better. Anna's answer: free-form works for most things, but for product development and token-sensitive tasks, the structured approach plus testing both versions is worth it. 33:10 – 34:04 — Closing Two-week break for May holidays, recording to follow. Brief check-in on whether Daria used the Week 1 reading guide.