Google Prompting Framework
📌 Prompting is defined as the language used to converse with AI to achieve desired outputs, based on Google's framework.
🔑 The five-step framework for effective prompting is Task, Context, References, Evaluate, and Iterate (T-C-R-E-I), summarized by the mnemonic "tupai curi roti emas hitam" (squirrel steals black gold bread).
✨ Enhancing the Task element by adding a Persona (role for the AI) and specifying the Output Format (e.g., table, PDF) leads to much more detailed results.

Improving Prompt Quality
🛠️ When providing Context, supplying more details—such as age, budget, or specific preferences—yields significantly more detailed and relevant output from the AI.
💡 References allow you to provide examples (e.g., a gift from the previous year) so the AI has a baseline for the desired output style or content.
🔄 The Iterate step emphasizes that initial prompts are rarely perfect; continuous refinement based on evaluation standards is crucial for achieving satisfaction.
🗣️ Four key methods for refining prompts include: reiterating the T-C-R-E-I framework with more detail, breaking the prompt into shorter, simpler sentences, using analogous tasks (e.g., asking for a story instead of a direct sales script), and imposing strict limitations (e.g., word count or specific time period).

Advanced Prompting Techniques
🔗 Prompt Chaining involves breaking down a complex request into sequential, connected steps (like building with Lego blocks), ensuring each output informs the next command for a comprehensive final result (e.g., summarizing a novel, then creating a tagline, then planning a promotion schedule).
🧠 Chain-of-Thought (CoT) Prompting requires the AI to explicitly detail its step-by-step thinking process before providing the final answer, enhancing transparency and accuracy, similar to showing your work in mathematics.
🌳 Tree-of-Thought (ToT) Prompting encourages the AI to explore multiple distinct paths or ideas simultaneously (like branches on a tree) before selecting the most promising one for deeper exploration and refinement.

Multimodal Interaction and Risks
🖼️ Multimodal Prompting extends interaction beyond text input to include images, audio, or links, allowing AI like Gemini to generate outputs across various formats (images, code, etc.) based on diverse inputs.
⚠️ The primary risks when using AI are hallucination (generating incorrect or nonsensical information) and bias (as AI is trained on human data).
🧍 The necessary solution for mitigating these risks is Human-in-the-Loop (HITL), meaning users must always verify, check, and reset AI outputs, as the final responsibility lies with the user.

Key Points & Insights
➡️ Always define a Persona and specify the Output Format in your initial prompt task for richer, structured results.
➡️ Use Chain-of-Thought prompting when accuracy or process transparency is critical, forcing the AI to show its reasoning step-by-step.
➡️ When stuck creatively, use Tree-of-Thought to generate and evaluate several distinct starting ideas before committing to one path.
➡️ Never accept AI output at face value; always prioritize verification and checking due to inherent biases and hallucinations.

📸 Video summarized with SummaryTube.com on Dec 09, 2025, 12:08 UTC