FAQ on Google's Prompt Engineering Course

What is prompting, according to the course, and what's the core framework for designing
effective prompts?
Prompting is the process of providing specific instructions to a generative AI tool to receive new
information or achieve a desired outcome, which can be text, images, video, sound, or code.
The course emphasizes a five-step framework (summarized as "Tiny crabs ride enormous
iguanas"):
Task: Clearly define what you want the AI to do.
Context: Provide relevant background information to improve the AI's understanding.
References: Include examples or past work to guide the AI.
Evaluate: Assess the AI's output to determine if it meets your needs.
Iterate: Refine the prompt based on the evaluation to achieve better results. The course also
mentions the importance of incorporating a persona (a role for the AI to embody) and specifying
the desired output format.
What are the four iteration methods suggested in the course for refining prompts when the initial
results aren't satisfactory?
The course suggests four iteration methods, remembered with the mnemonic "Rahen saves
tragic idiots":
Revisit the Prompting Framework: Re-examine the initial prompt using the five-step framework
("Tiny crabs ride enormous iguanas") to identify areas for improvement (e.g., adding more
context, references, or a persona).
Separate into Shorter Sentences: Break down complex prompts into simpler, more manageable
sentences to avoid overwhelming the AI.
Switch to an Analogous Task: Reframe the desired outcome as a similar, more relatable task to
elicit more creative or relevant responses.
Introduce Constraints: Add specific limitations or requirements to narrow the focus and generate
more targeted results.
What is multi-modal prompting, and how does it differ from traditional text-based prompting?
Multi-modal prompting involves using different types of data (modalities) as input for AI models,
including pictures, audio, video, and code, in addition to text. The fundamental prompt
engineering framework ("Tiny crabs ride enormous iguanas") remains the same, but you need
to be more precise about specifying the input and output modalities and providing appropriate
context. For example, you might input an image and ask the AI to generate a social media post
featuring it.
What are hallucinations and biases in AI, and how does the course recommend mitigating these
issues?
Hallucinations: These are instances where an AI tool provides outputs that are inconsistent,
incorrect, or nonsensical.
Biases: These are prejudices an AI tool can pick up from the human content it is trained on, and
can be related to gender, race, etc.
To minimize these issues, the course recommends a "human in the loop" approach, which
involves carefully checking and verifying the outputs generated by AI tools to ensure accuracy
and fairness.
Module 2 of the course focuses on design prompts for everyday work tasks. Can you give an
example of a prompt for writing an email, and explain why it is effective?
Example Prompt: "I'm a gym manager, and we have a new gym schedule. Write an email
informing our staff of the new schedule. Highlight the fact that the MWF Cardio Blast class
changed from 7:00 a.m. to 6:00 a.m. Make the email professional, friendly, and short so that the
reader can skim it quickly. Here's the new schedule: [link to schedule]."
Effectiveness: This prompt incorporates several key elements:
Clear Task: It explicitly states the purpose of the email.
Context: It provides background information about the schedule change and the sender's role.
Format: It specifies the desired tone (professional, friendly) and length (short).
References: It includes a link to the new schedule for detailed information.
What are some examples of prompts for data analysis and building presentations, as covered in
Module 3?
Data Analysis: "Attached is a Google Sheet of store data. How can I create a new column in
Sheets that calculates the average sales per customer for each store? Give me insights into the
relationship between daily customer count, items available, and sales based on the given data."
Presentations: The video suggests using prompts to help create presentation outlines, generate
slide content, or even suggest visual elements. Example: "Create an outline for a presentation
on [topic], targeting [audience]. Include key talking points and suggestions for visual aids."
Describe the advanced prompting techniques of prompt chaining, Chain of Thought, and tree of
thought, as covered in Module 4.
Prompt Chaining: This involves guiding AI through a series of interconnected prompts to
progressively build complexity and refine the desired outcome. The output of one prompt
becomes the input for the next.
Chain of Thought: This technique asks the AI to explain its reasoning in a step-by-step manner,
similar to showing your work in math. This helps understand the AI's thought process and
identify areas for improvement. Prompts end in, "Explain your thought process".
Tree of Thought: This method allows exploring multiple reasoning paths simultaneously, like
branches on a tree. This is useful for complex problems that require brainstorming and
evaluating different options.
What are AI agents, and how does the course suggest creating them? What are the two types
of Agents covered in the course?
AI agents are specialized AI tools designed to help with specific tasks and answer questions in
a targeted manner. The course provides a guideline for creating AI agents:
Assign a Persona: Define the role you want the agent to embody (e.g., "act like a successful
personal fitness trainer").
Provide Context: Give as much detail as possible about the scenario and the desired
conversation.
Specify Interactions: Define the types of conversations or interactions you want to have with the
agent.
Set a Stop Phrase: Define a phrase that will end the interaction.
Provide Feedback Mechanism: Include a method for the agent to provide feedback or areas of
improvement after the conversation.
The two types of agents covered in the course are:
Agent Sim: A simulation agent that simulates scenarios, such as interviews or role-playing
exercises.
Agent X: An agent for expert feedback that provides personalized tutoring or consulting on any
topic.