1. Introduction
2. Team Structure3. Process Flow
3.1. Module Learning & Experience Design
3.2. Individual Flow Design
3.3. Narrative Design & Skills Mapping
3.4. New Assets Creation
3.5. Annex: World Building - Learning & Experience Design
3.6. Annex: World Building - Story Design
3.7. Process Flow Conclusion4. Caveats
4. Caveats5. LLM Prompt Engineering Techniques
5.1. Use The Latest Model
5.2. Zero-shot Prompting
5.3. Few-shot Prompting
5.4. Chain-Of-Thought Prompting
5.5. Structuring Prompts
5.6. Describing Prompts
5.7. Editing Prompts
5.8. Extending Responses
5.9. Multiple Users Collaborating6. Text-to-Image Prompting Engineering Techniques
6.1. General Techniques
6.2. Photography
6.3. Architecture
6.4. Various Aesthetic Styles
6.5. Product & Material3.3. Narrative Design & Skills Mapping
The Narrative Designer will be responsible for the process of designing dialogue and skills mapping while the Learning Designer will finalize the score.
Refer to the section ‘Multiple Users Collaborating’ on how the Narrative Designer can get ChatGPT up to speed by sharing sufficient and relevant information that the Learning Designer has developed thus far. The Narrative Designer would then be able to start developing the dialogue.
The Narrative Designer uses the CPD to design the dialogues.
3.3.1. Best Path Dialogue with Skills Mapping for All Flows
The methodology will vary depending on your topic, your level of expertise, your familiarity with the subject matter, the number of characters in a conversation, etc.
We can have ChatGPT generate all or part of a script or we can roleplay with it line by line
There are strengths, weaknesses, and variations to each and both outputs are still a first draft.
3.3.1.1. Beats of a Conversation
We can ask ChatGPT to generate an outline of the conversational flow. It does the best job of this if we provide it contents defined in the Experience Requirements Document (ERD).
In this module, we have [insert total number] different conversational flows.
Let’s create flow [X] of the module.
Narrative description: ###
[Insert flow narrative description]
#####
Conversational goals: ####
[Insert flow conversational goal]
####
Learning objectives: ###
1/ [Insert flow learning objective]
2/ …
###
Behaviors that the learner will demonstrate: ###
1/ [Insert flow key learning point]
2/ …
####
Skills: ###
1/ [Insert Skills]
2/ …
###
Return with, “Added to LXDbot’s memory” and wait for more information about the emotional arcs of the participants in this conversational flow.
List the beats of the Flow [X] conversation and describe them in detail.
Learner’s emotional arc: ###
[Insert learner’s emotional arc]
###
[Insert character name]’s description: ###
[Insert character’s description]
###
[Insert character name]’s emotional arc: ###
[Insert character’s emotional arc]
###
[Insert character name]’s personality traits: ###
[Insert character’s personality traits]
###
3.3.1.2. Beat Script Generation
We can use ChatGPT to generate a partial script for individual beats of a conversation using the following prompt:
Generate a script for Flow [X] using the beat.
Beat: ###
[Insert the beat description]
###
Abide by the following constraints: ###
1/ Both the learner and [Insert character name] should each speak at least three times per beat.
2/ Learner dialogue should demonstrate the following skills: “””[Insert target skills]”””
3/ The maximum character count for each independent line of learner dialogue is 150 characters.
4/ [Add additional constraints as needed]
###
Continue generating dialogue using the beat.
Beat: ###
[Insert the beat description]
###
3.3.1.3. Whole Script
Take advantage of the content we have already generated at this point, including character profiles and 5 act story arcs.
Give Chat GPT guide rails. Iterate by revising the same prompt.
Write a dialogue continuing the conversation for a learning module about [insert topic] using the following information.
Character profiles: ###
[Insert character profiles, and their relationship with each other and the learner]
###
Conversation goal: ###
[Insert a 1-2 lines description of the objective of the conversation]
###
Story arc or conversation beats: ###
[Insert story arc or conversation beats]
###
Module description, learning objectives, learning points, and skills: ###
[Insert module description, relevant learning objectives, learning points, and skills]
###
Rewrite this into a script with only dialogue. Do not include prose.
Adhere to these guidelines: ###
1/ The learner must speak a minimum of [insert number] times
2/ Each of the learner’s responses must contain no more than [insert number] of characters.
3/ Each of the characters must speak at least [insert number]
###
Try adjusting the parameters to focus on one or two sections before proceeding. There are limits, and taking a bite-sized approach can often be better.
Try adding specifics. “Critical project” is vague, “AI-powered Refrigerator” or some fictional product is better. Try using ChatGPT to generate more specific ideas for existing plot points.
For example:
Generate [insert number] of ideas for innovative, high-tech products that a mid-sized company could sell directly to consumers.
For each of the [insert number] of products, include the product name and a 2-3 sentence description.
Iterate, and try including skills. It will not always assign them correctly, but it will give an indication of what it is trying to accomplish and can spark new ideas.
3.3.1.4. Role-Play Exchange
Make sure we clearly establish parameters. Don’t be afraid to interject and stop generation if it misunderstands.
Play the role of [Insert character name] in a conversation with me until I tell you to stop.
I'm beginning the conversation.
Conversation Description: ###
[Insert a description of the conversation]
###
Character’s profile: ###
[Insert character profile]
###
My role as a learner: ###
[Insert a description of the learner’s role whom we are playing in this conversation]
###
Context of the physical environment: ###
[Insert a description of the physical environment that the characters are in]
###
Thanks for meeting with me, Charles. It's Monday, and it already feels like a long week.
Charles: I'm always here for you. Monday or not, we'll tackle whatever challenges come our way. So, what's been weighing on your mind?
Well, I just got wind of our main competitor partnering with a generative AI company.
Tell me more about it.
I just heard Daring Robotics partnered with Generating the Future. It's more of that Gen AI hype that's all over the news. If I'm being honest with myself, I'm worried Phoenix Robotics is falling behind.
3.3.2. Learning Designer, Development & SME Review
The Producer will arrange for the Learning Designer, Developmental Editing Lead, and relevant Subject Matter Experts (SMEs) to review the material simultaneously.
These reviewers will have a 48-hour window to offer their feedback. Should they fail to do so, the Producer reserves the right to move forward without their input.
After receiving the reviewers' feedback, the Producer will instruct the Narrative Designer to implement the required modifications.
3.3.3. Stakeholder Approval
Once the Narrative Designer has made the necessary changes, the Producer will arrange for the final review by the final stakeholder who is usually one of the Learning Experience Leads.
At this point in time, we will no longer be accepting any change requests to the best path and will proceed with the dialogue branching process.
3.3.4. Branching with Skills Mapping for All Flows
3.3.4.1 Suboptimal Response Brainstorm By Eliminating Skills
First begin with providing ChatGPT the skill point awarding rubric.
How would you rewrite the dialogue if it didn't demonstrate the skill "[Insert skill name]" or "[Insert skill name]"?
Dialogue: ###
[Provide a dialogue response with associated skill values
Example: The dialogue option "Of course. How about I start by summarizing the issue, then we can discuss solutions?" demonstrates the following skills and awards the following skill points "Verbal Communication" +2 points; "Collaboration" + 2 points; and "Other awareness" +1 point. ]
###
The Skill Point Awarding Rubric awards points for skill demonstration.
The criteria for +1 skill point is: ###
Demonstration of the skill was:
- Not Precise
- Heavily Formulaic
- Not Situationally Appropriate
###
The criteria for +2 skill points is: ###
Demonstration of the skill was:
- Fairly Precise
- Somewhat Formulaic
- Fairly Situationally Appropriate
###
The criteria for +3 skill points is: ###
Demonstration of the skill was:
- Precise
- Adapted
- Situationally Appropriate
###
The definitions for the facets are: ###
- Adaptedness: How natural the demonstration is. In contrast to a prepared or formulaic demonstration.
- Situational-Appropriateness: How well-suited the demonstration of the skill is to the situation.
- Precision: How accurate the demonstration was.
###
3.3.4.2 Variations Of A Decision Response Option
For dialogue branching, we try asking for alternative learner responses.
Can you give 5 different versions of this dialogue that only score [insert skill point value] for [insert skill name].
Dialogue: ###
[Insert dialogue]
###
Rewrite 3 alternative versions of Statement X that respond to Statement A in a poor way that damages rapport and doesn't support clear communication.
Statement X is a response to Statement A. Statement X uses the skills of empathy and active listening.
Statement X: ###
[Ex. Hiroshi, I appreciate your input. Could you provide more details about the issues your team is facing? I want to make sure I understand your perspective.]
###
Statement A: ###
[Ex. We are working on it. Some issues need attention, but we will manage.]
###
Alternative 1:
Hiroshi, that's not very helpful. We need specifics, not just vague statements about "issues." Please try to be more clear next time.
Alternative 2:
Well, Hiroshi, it seems like you're not really interested in sharing the details. If you don't want to talk about the issues, why even mention them?
Alternative 3:
Hiroshi, I don't have time for beating around the bush. Just tell us what the problems are so we can move on. We can't afford to waste time with ambiguity.
Specifying the consequence of those lack of skills seems to be helpful. Results will likely need to be finessed - mistakes can be a bit ham-handed.
3.3.5. Same Cycle of Reviews as Best Path
The reviews follow the same cycle as that of the Best Path.
3.3.6. Scoring Finalization
The Learning Designer will finalize scoring without the aid of ChatGPT.
The learning module is now ready for performance editing to complete the module.
← Previous
Next →
On this page:
- 3.3. Narrative Design & Skills Mapping
- 3.3.1. Best Path Dialogue with Skills Mapping for All Flows
- 3.3.1.1. Beats of a Conversation
- 3.3.1.2. Beat Script Generation
- 3.3.1.3. Whole Script
- 3.3.1.4. Role-Play Exchange
- 3.3.2. Learning Designer, Development & SME Review
- 3.3.3. Stakeholder Approval
- 3.3.4. Branching with Skills Mapping for All Flows
- 3.3.4.1 Suboptimal Response Brainstorm By Eliminating Skills
- 3.3.4.2 Variations Of A Decision Response Option
- 3.3.5. Same Cycle of Reviews as Best Path
- 3.3.6. Scoring Finalization