ChatGPT 4: A Bargain for Scrum Practitioners?

TL; DR: ChatGPT 4: A Bargain for Scrum Practitioners?

When OpenAI released its new LLM model GPT-4 last week, I could not resist and signed up for $20 monthly. I wanted to determine whether ChatGPT 4 is superior to its predecessor, which left a good impression in recent months; see my previous articles on Scrum, Agile, and ChatGPT.

I decided to run three comparisons, using the identical prompt to trigger answers from the new GPT-4 and previous GPT-3.5 models. Read on and learn what happened. It was not a foregone conclusion.

🇩🇪 Zur deutschsprachigen Version des Artikels: ChatGPT 4: Ein Schnäppchen für Scrum-Praktiker?

🗞 Shall I notify you about articles like this one? Awesome! You can sign up here for the ‘Food for Agile Thought’ newsletter and join 45,000-plus other subscribers.

🎓 Join Stefan in one of his upcoming Professional Scrum training classes!

📖 Get notified when the Scrum Anti-Patterns Guide book is available!

Join your peers on April 5, 2023: Hands-on Agile #50: The Product Community of Practice w/ Petra Wille.

Setting up the Session

I ran the session on ChatGPT on Wednesday, March 15, from 5:40 am CET to 6:30 am CET on the models GPT-3.5 and GPT-4 after upgrading to ChatGPT Plus.

I did not apply Grammarly to ChatGPT’s answers.

The Initial ChatGPT Prompt to Explore the New Mode GPT-4

I used the following initial prompt to prime ChatGPT: “Are you using the same training data as GPT-3.0 or 3.5?”

ChatGPT-4.0:

As a GPT-4 model, I am an updated version of the previous GPT models. While I share some similarities with GPT-3 and GPT-3.5, there are differences in the training data used for each model.

The primary difference is the time period of the data. GPT-3 and GPT-3.5 were trained on data available up until around 2020 and 2021, respectively. In my case, as GPT-4, my training data extends until September 2021. This allows me to have a more up-to-date understanding of the world, including newer events and developments.

Additionally, improvements in data curation, preprocessing, and filtering techniques have likely been made between versions, leading to differences in the quality and breadth of the training data. While the core data sources might be similar, such as books, articles, and websites, the data sets themselves would not be identical across these models.

The Experiments

Now that we have a better understanding of the model’s background let’s run three experiments to figure out which model is superior. I will run three comparisons or A/B tests, using the identical prompt to trigger answers from the new GPT-4 and previous GPT-3.5 models:

1. The first experiment compares the Scrum Guide 2017 to the Scrum Guide 2020
2. The second experiment is designing a Retrospective exercise
3. The third experiment is to help a new Scrum team to design a Definition of Done.

Learn more about the new GPT-4 model from OpenAI’s GPT-4 Developer Livestream of March 14, 2023, or the OpenAI GPT-4 Technical Report.

Cannot see the form? Please click here.

Experiment 1: Comparing the Scrum Guide 2017 to the Scrum Guide 2020

Prompt: I want you to act as a Scrum Master. Tell me: what is the difference between the Scrum Guide 2017 and the Scrum Guide 2020?

ChatGPT-3.5:

As a Scrum Master, I’d be happy to answer your question about the differences between the Scrum Guide 2017 and the Scrum Guide 2020.

The Scrum Guide is the official guide to Scrum, and it provides a framework for implementing and improving Scrum practices. The latest version of the Scrum Guide was released in November 2020, and it contains several updates from the previous version released in 2017. Here are some of the key differences between the two versions:

1. Focus on the Scrum team: The Scrum Guide 2020 places more emphasis on the Scrum team, rather than just the Development Team. The Product Owner and Scrum Master are now considered equal members of the Scrum Team.
2. Empirical process control: The Scrum Guide 2020 highlights the importance of empirical process control and the three pillars of transparency, inspection, and adaptation.
3. Updated language: The language in the Scrum Guide 2020 has been updated to be more inclusive and gender-neutral.
4. Simplified framework: The Scrum Guide 2020 has removed some of the prescriptive language from the previous version to make the framework more flexible and adaptable to different contexts.
5. Refinement instead of grooming: The term “Backlog Refinement” is now used instead of “Backlog Grooming” to emphasize the continuous improvement aspect of the practice.

Overall, the changes in the Scrum Guide 2020 are intended to make the framework more accessible and adaptable to a wider range of contexts while still maintaining the core principles and values of Scrum.

Comment: These answers are irrelevant, or hallucinations: Scrum Master and Product Owner are now “equal members of the Scrum?” No mention of commitments, namely the Product Goal? Highlighting the importance of empirical process control? That has been at the core of Scrum from its beginnings. The answer is embarrassing.

ChatGPT-4.0:

As a language model, I can provide you with the key differences between the Scrum Guide 2017 and the Scrum Guide 2020. Here are some of the most notable changes:

Less prescriptive and more flexible: The Scrum Guide 2020 is shorter and more focused on the Scrum framework’s core concepts. It has reduced the prescriptive nature of some practices, allowing teams more flexibility in adopting Scrum according to their specific context.

Introduction of the Product Goal: The 2020 version introduces the concept of a Product Goal, which is a long-term objective for the product. It provides guidance and direction for the team, and all the Sprint Goals should align with and contribute to the Product Goal.

Simplified team structure: The Scrum Guide 2020 eliminates the distinction between development team members and other roles. It now refers to the whole team as the “Scrum Team,” consisting of the Scrum Master, Product Owner, and Developers, all of whom share accountability for delivering a potentially releasable increment each sprint.

A greater focus on commitment: The Scrum Guide 2020 emphasizes the importance of commitment for each artifact. The Product Backlog has the Product Goal, the Sprint Backlog has the Sprint Goal, and the Increment has the Definition of Done. These commitments help provide clarity and focus for the Scrum Team.

Self-managing instead of self-organizing: The 2020 version refers to Scrum Teams as self-managing, emphasizing their autonomy in making decisions regarding how to achieve the Sprint Goal, and how to create the product Increment. This change highlights the team’s responsibility for organizing and managing their own work, rather than relying on external direction.

Removal of the Sprint Retrospective’s three-question format: The 2020 Scrum Guide no longer suggests the specific three-question format (“What went well? What can be improved? What will we commit to improve in the next Sprint?”) for the Sprint Retrospective. Instead, it emphasizes the need to plan improvements and foster a continuous improvement mindset.

Updated the Definition of Done: The Scrum Guide 2020 clarifies that the Definition of Done is a formal description of the state of the Increment when it meets the quality measures required for the product. It also emphasizes that the Definition of Done is used to assess when work is complete on the product Increment.

These are some of the main differences between the Scrum Guide 2017 and the Scrum Guide 2020. The 2020 version aims to provide a more straightforward, flexible, and focused framework for Scrum practitioners, emphasizing the importance of team collaboration, commitment, and continuous improvement.

Comment: We have a clear winner! GPT-4’s answer is better and more substantial than the previous model’s.

Score: ChatGPT-4 1-0 ChatGPT-3.5

Experiment 2: Designing a Retrospective Exercise

Prompt: I want you to act as an experienced Scrum Master.

Your Scrum team had a problematic Sprint and missed accomplishing the Sprint Goal. Some team members had unexpected food poisoning. The procurement department failed to renew credit card information for an essential service your team uses, and that service was no longer available. Additionally, the Scrum team ran into unexpected technical debt hidden in the legacy part of the code base. The tension in the team is high; arguments have become more frequent.

As a Scrum Master, I want you to use your knowledge of Scrum, iterative and incremental product development, non-violent communication, collaboration, team building, Scrum values, psychology, and Liberating Structures to design a Retrospective that helps the Scrum team to overcome recent setbacks and improve the relationships among each other.

ChatGPT-3.5:

I’m sorry to hear that your Scrum team had a challenging Sprint. As an experienced Scrum Master, I’m here to help you navigate this difficult situation and design a Retrospective that will help the team overcome recent setbacks and improve relationships among team members.

First, it’s essential to acknowledge that setbacks and challenges are a natural part of the Scrum process. Our goal is not to eliminate problems but to learn from them and use them as opportunities for growth and improvement.

To start the Retrospective, we need to create a safe and respectful space where team members can express themselves honestly and openly. One Liberating Structure that could help us achieve this is the “1-2-4-All” structure, which involves individuals reflecting silently, sharing with a partner, joining a group of four to discuss their thoughts, and then sharing the group’s ideas with the entire team.

We’ll start by asking the team to reflect on the past Sprint and identify what went well and what didn’t go as planned. We’ll then focus on the challenges that arose during the Sprint, including the unexpected technical debt, credit card issues, and team members getting sick.

Once we have identified the challenges, we’ll use the Liberating Structure “TRIZ” to explore potential solutions. TRIZ stands for “Theory of Inventive Problem Solving,” and it’s a structured approach to generating creative solutions to complex problems. We’ll ask the team to identify the root cause of each problem and then use TRIZ to generate new ideas for addressing those issues.

Next, we’ll use the Liberating Structure “Celebrity Interview” to encourage empathy and understanding among team members. Each team member will take on the role of a “celebrity” who embodies certain values or characteristics, and the team will ask them questions about how they would handle the recent setbacks.

Throughout the Retrospective, we’ll emphasize the importance of non-violent communication and collaboration, and we’ll remind the team of the Scrum values of commitment, courage, focus, openness, and respect.

By the end of the Retrospective, our goal is to have identified concrete steps that the team can take to address the challenges that arose during the previous Sprint and improve their relationships with each other. We’ll create an action plan that outlines these steps and assign owners to each task to ensure accountability and follow-through.

Comment: Sounds good, initially, doesn’t it? Too bad the suggestion does not pass the test. Of course, you could use 1-2-4-All to start the Retrospective. However, if every team member shall have their airtime, a structure like “Conversation Café” would be better suited. Instead of TRIZ, I would use “Nine Whys” or “What. So What. Now What?” Finally, the “Celebrity Interview” is useless for the intended purpose. If empathy is your main concern, “Heard, Seen, Respected” would be a better choice.

ChatGPT-4.0:

As an experienced Scrum Master, I would design a Retrospective that not only addresses the setbacks but also focuses on rebuilding trust, improving communication, and fostering a collaborative environment within the team. The Retrospective will follow these five steps:

1. Set the Stage: Start the session by acknowledging the challenges the team faced during the Sprint and emphasize the importance of learning from setbacks. Encourage everyone to participate in a safe, open, and respectful environment. Use a short icebreaker activity to help the team relax and prepare for the session.
2. Gather Data: Begin by asking the team members to write down on sticky notes what went well, what didn’t go well, and any surprising events during the Sprint. Use a timeline activity to help visualize the events and identify patterns. Ask the team to share their experiences openly, being mindful of non-violent communication principles.
3. Generate Insights: Use the ‘5 Whys’ technique or the ‘Fishbone’ diagram to analyze the root causes of the problems faced. Encourage the team to focus on the process and not on individuals. Emphasize Scrum values, such as courage and openness, while discussing the issues. Facilitate a discussion on the impact of the unexpected technical debt and how to address it moving forward. Also, discuss how to improve communication with the procurement department to avoid future service disruptions.
4. Decide What to Do: Guide the team to collaboratively identify action items to address the root causes of the problems. Prioritize the action items by having the team vote on the most important ones. Encourage the team to come up with both short-term and long-term actions. For example, a short-term action could be to create a checklist for procurement processes, while a long-term action could involve implementing automated tests to uncover hidden technical debt.
5. Close the Retrospective: End the session with a team-building activity that reinforces the Scrum values and fosters collaboration. Ask the team members to share one positive thing about their colleagues or to express gratitude for something a teammate did during the Sprint. This will help to improve relationships and reduce tensions among the team members. Conclude the Retrospective by summarizing the key takeaways and the agreed-upon action items, and reiterating the importance of continuous improvement and collaboration.

Follow up with the team in the upcoming Sprints to ensure that the action items are being implemented and to gauge the effectiveness of the improvements. Encourage the team to maintain open communication and support each other as they work towards achieving their goals.

Follow-up Prompt: Is it necessary to run the Retrospective in this strict, prescriptive manner?

ChatGPT-4.0:

While the five-step structure I provided is a popular and effective way to conduct Retrospectives, it is not mandatory to follow it in a strict, prescriptive manner. As a Scrum Master, you should be flexible and adapt the format based on your team’s needs, preferences, and specific situation.

It is essential to maintain the core principles of a Retrospective, which include creating a safe and open environment for discussion, reflecting on the past Sprint, generating insights, identifying improvements, and fostering a sense of collaboration and continuous improvement within the team.

You can experiment with different techniques and activities to achieve these objectives, and it is common for experienced Scrum Masters to tailor Retrospectives to suit their team’s dynamics. Some alternative approaches include:

1. Lean Coffee: A structured, agenda-less meeting format that allows participants to democratically decide on the topics they want to discuss.
2. World Café: A technique that uses rotating small-group discussions to gather input from all team members on multiple topics.
3. Starfish: A visual method for capturing what the team should start doing, stop doing, continue doing, do more of, and do less of.

The key is to be receptive to your team’s feedback and continuously improve the Retrospective process to ensure it remains valuable and engaging for everyone involved.

Comment: The first suggestion was merely a standard application of the Esther Derby and Diana Larsen model, ignoring the additional information on the situation from the prompt. If we also consider the answer to the follow-up prompt, the GPT-4-based performance gets a bit better, although it is mediocre at best. However, such a performance is sometimes sufficient when the competitor’s answer is even worse.

Score: ChatGPT-4 2-0 ChatGPT-3.5

Experiment 3: Help a New Scrum Team to Design a Definition of Done

Prompt: I want you to act as an experienced Scrum Master.

The Definition of Done is essential for a Scrum team’s success. It defines the quality standard any Increment is expected to meet by stakeholders and governance bodies. The quality level does not have to be perfect or top-of-the-line in all situations. However, it is critical that once an Increment is “done,” it can be delivered to the customers without any legal, ethical, or financial repercussions.

I want you to use your knowledge of Scrum, iterative and incremental product development, collaboration, product management, stakeholder collaboration, quality management, innovation, and business agility in general to suggest a process on how a new Scrum team can create an initial Definition of Done and then maintain it accordingly.

ChatGPT-3.5:

Sure, I’d be happy to help you with that.

Creating an Initial Definition of Done:

1. Collaborate with the team: As a Scrum Master, you need to work with the team to create a Definition of Done that everyone agrees upon. You can facilitate a discussion during a Sprint Retrospective or a dedicated Definition of Done workshop to ensure everyone’s voices are heard.
2. Understand stakeholders’ expectations: It’s essential to understand what the stakeholders expect from the product and what quality standards they have in mind. The team should incorporate these expectations into the Definition of Done, which will ensure that the product meets stakeholder expectations.
3. Identify quality metrics: Define the metrics that can be used to measure quality. Metrics like code coverage, performance, and security can help establish the Definition of Done. The team should strive to meet these metrics when delivering a potentially releasable Increment.
4. Consider technical debt: Teams should consider the technical debt they’re willing to take on when developing a feature. Technical debt is the cost of rework caused by taking shortcuts or delaying work. The team should define the level of technical debt they’re willing to accept and communicate this to stakeholders.
5. Keep it simple: The Definition of Done should be concise, clear, and easy to understand. It should be written in a language that’s accessible to all team members and stakeholders.

Maintaining the Definition of Done:

1. Review and update the Definition of Done regularly: The Definition of Done should be reviewed and updated regularly. This can be done during Sprint Retrospectives or dedicated Definition of Done workshops. The team should consider feedback from stakeholders and make necessary changes to the Definition of Done.
2. Communicate changes to stakeholders: Whenever the Definition of Done changes, the team should communicate this to stakeholders. This will help ensure that everyone is on the same page and that the product meets expectations.
3. Use the Definition of Done during Sprint Reviews: During Sprint Reviews, the team should use the Definition of Done to determine whether the Increment is ready for release. If it doesn’t meet the Definition of Done, the team should identify the gaps and work to close them in the next Sprint.
4. Track and report progress: Teams should track and report progress against the Definition of Done metrics. This can help identify areas where improvement is needed and drive continuous improvement.
5. Use feedback to improve: The team should use feedback from stakeholders to improve the Definition of Done. This can be done by incorporating feedback into the Definition of Done or adjusting the metrics used to measure quality.

Overall, creating and maintaining a Definition of Done is critical to the success of a Scrum team. By following these steps, the team can ensure that they’re delivering quality Increments that meet stakeholder expectations and can be released without any legal, ethical, or financial repercussions.

Comment: Wow, who would have thought ChatGPT-3.5 could deliver at such a high level? The approach is collaborative and inclusive, technical debt is part of the consideration, and the Definition of Done shall be simple and regularly inspected and adapted, where feedback from stakeholders is as valuable as data from tracking suitable metrics.

ChatGPT-4.0:

As an experienced Scrum Master, I recommend the following process for a new Scrum team to create an initial Definition of Done (DoD) and maintain it:

1. Gather Inputs: Start by collecting inputs from the Scrum team, stakeholders, governance bodies, and any existing organizational standards or guidelines. These inputs will help establish the baseline quality expectations and constraints for the product.
2. Educate the Team: Ensure that all team members understand the purpose and importance of the DoD. Explain how it helps create transparency, manage expectations, and ensure a potentially releasable product Increment.
3. Collaborative Workshop: Organize a workshop with the Scrum team, Product Owner, and relevant stakeholders to collaboratively create the initial DoD. Facilitate discussions on the quality standards and acceptance criteria, including functional and non-functional requirements, performance, security, usability, and compliance.
4. Draft the Definition of Done: Based on the workshop outcomes, create a draft of the DoD that includes a clear and concise list of criteria that must be met for an Increment to be considered “done.” Ensure that the criteria are measurable, achievable, and aligned with the overall product vision.
5. Review and Refine: Present the draft DoD to the Scrum team, Product Owner, and stakeholders for review and feedback. Make any necessary adjustments based on their input, and iterate until a consensus is reached.
6. Communicate and Display: Once the initial DoD is finalized, communicate it to all relevant parties and display it prominently in the team’s workspace or on a shared online platform. This ensures that everyone is aware of the expectations and can refer to the DoD when needed.
7. Regularly Review and Update: Incorporate a review of the DoD into the team’s Retrospectives, or schedule periodic reviews to ensure it remains relevant and aligned with the product’s evolution. Update the DoD as needed, based on lessons learned, changing requirements, stakeholder feedback, and any changes in organizational standards or regulations.
8. Monitor Compliance: As a Scrum Master, monitor the team’s adherence to the DoD throughout each Sprint. Provide support and guidance to help the team meet the established criteria and address any challenges that arise.

By following this process, a new Scrum team can create an initial Definition of Done and maintain it to ensure consistent quality and alignment with stakeholder expectations throughout the product development lifecycle.

Comment: Gee, this is unexpected. It sounds like a PMI brochure.

Follow-up Prompt: Your suggestion contains too much project management thinking and ignores the importance of collaboration and teamwork in a Scrum team. Partly, it is plain wrong: A Scrum Master does not create the initial Definition of Done. Also, they do not enforce compliance; that is the task of the Developers. Recreate the answer based on Scrum’s first principles of self-management and servant leadership.

ChatGPT-4.0:

I apologize for any confusion in my previous response. As a Scrum Master, your role is to facilitate and guide the team through the process of creating and maintaining the Definition of Done (DoD), while emphasizing the importance of collaboration, teamwork, and self-management. Here’s an updated approach based on Scrum’s first principles:

1. Set the Context: Ensure that the Scrum team, including the Product Owner, understands the purpose and importance of the DoD. Explain how it helps establish a shared understanding of quality standards, manage expectations, and ensure a potentially releasable product Increment.
2. Facilitate Collaboration: Encourage the Scrum team to collaborate with stakeholders, governance bodies, and any existing organizational standards or guidelines to gather inputs that will help establish the baseline quality expectations and constraints for the product.
3. Empower the Team: Support the Developers in organizing a workshop or a series of discussions where they can collaboratively create the initial DoD. Encourage them to consider functional and non-functional requirements, performance, security, usability, and compliance.
4. Refinement and Consensus: As the Developers draft the DoD, facilitate discussions and feedback loops between the team, Product Owner, and stakeholders. Help them iterate on the DoD until they reach a consensus on the criteria that must be met for an Increment to be considered “done.”
5. Encourage Transparency: Once the initial DoD is finalized, encourage the Developers to communicate it to all relevant parties and display it prominently in the team’s workspace or on a shared online platform. This will ensure that everyone is aware of the expectations and can refer to the DoD when needed.
6. Foster Continuous Improvement: During Retrospectives, remind the team to review the DoD and discuss any potential improvements or adjustments based on lessons learned, changing requirements, stakeholder feedback, or changes in organizational standards or regulations.
7. Support Compliance: As a servant leader, provide support and guidance to help the Developers adhere to the DoD throughout each Sprint. Encourage open communication and address any challenges that arise. Remember that the responsibility of meeting the DoD criteria lies with the Developers, not with the Scrum Master.

By following this approach, you can enable the Scrum team to create and maintain a Definition of Done that is rooted in collaboration, self-management, and continuous improvement, while staying true to the principles of servant leadership.

Comment: ChatGPT-4.0 talks a lot about collaboration, self-management, and continuous improvement. However, it seems that the Scrum Master is still in charge of pushing the Developers in the right direction from the Scrum Master’s perspective. Honestly, I like ChatGPT-3.5’s answer better. PS: The Product Owner is also a member of the Scrum team.

Score: ChatGPT-4 2-1 ChatGPT-3.5

Conclusion

ChatGPT based on the GPT-4 model has advantages over the “old” GPT-3.5 model. OpenAI promised better reasoning and conciseness and delivered. However, judging by three small everyday experiments, the new model’s advantage is not as conclusive as expected. So, let’s wait for GPT-5. Nevertheless, I will continue my subscription.

What is your opinion as a Scrum Practitioner: Is ChatGPT-4.0 worth spending $20 per month compared to ChatGPT-3.5? Please share it with us in the comments.

📖 ChatGPT 4: A Bargain for Scrum Practitioners? — Related Posts

The Stoic Scrum Master

Club Scrum: What Are You Doing all Day, ChatGPT — as a Scrum Master?

ChatGPT Prompts for Scrum Masters, Product Owners, and Developers

Agile Transformation: ChatGPT or McBoston?

A ChatGPT Job Interview for a Scrum Master Position

ChatGPT Product Owner Job Interview — Will You Compete for a Position w/ an LLM Soon?

Agility and Scrum According to OpenAI’s ChatGPT — Be Surprised!

What Is ChatGPT Doing … and Why Does It Work?

📅 Scrum Training Classes, Workshops, and Events

Learn more about ChatGPT 4 with our Scrum training classes, workshops, and events. You can secure your seat directly by following the corresponding link in the table below:

See all upcoming classes here.

You can book your seat for the training directly by following the corresponding links to the ticket shop. If the procurement process of your organization requires a different purchasing process, please contact Berlin Product People GmbH directly.

✋ Do Not Miss Out and Learn more about ChatGPT 4 — Join the 12,000-plus Strong ‘Hands-on Agile’ Slack Community

I invite you to join the “Hands-on Agile” Slack Community and enjoy the benefits of a fast-growing, vibrant community of agile practitioners from around the world.

If you like to join all you have to do now is provide your credentials via this Google form, and I will sign you up. By the way, it’s free.

Support your Scrum Master’s efforts checking ChatGPT 4 for hallucinations by pointing to the free Scrum Anti-Patterns Guide: