Crossing the Chasm
When a new technology emerges, it has to move beyond early adopters
In this edition:
My thoughts
Notable headlines
Learning, Tools, and Experiments
Using the MISTAKES Framework
AI Mistakes
My Thoughts
As the Generative AI revolution continues, a trend is beginning to form. Software companies are finding ways to integrate generative AI into their solutions regardless of industry.
It turns out that having built an AI model that can somewhat intelligently talk about everything can be integrated somewhere along the process of doing everything.
The biggest application of this technology is at a step in the process where MISTAKES can be made safely. This is what has (in large part) driven me to create the MISTAKES framework. AI is incredibly capable of helping us to understand how to start and how to make progress…. but it’s not really good at solving the problem completely, permanently, or even perhaps sustainably.
If you leverage a framework that allows for MISTAKES, you can obtain all the benefits of that intelligence and avoid the problems. No matter what technology can do for us, it should never put a human in a position to no longer ever understand the problem being solved.
The reality is that we’ve already been here. We know how this story ends. That book was written by Nicholas Carr in 2014. I can wholeheartedly recommend this book, especially today, because we already understand what happens when humans give up their thinking to a machine.
In the introduction to The Glass Cage: Automation and Us, Nicholas Carr shared an "Alert for Operators” published by the FAA on January 4, 2013. I won’t rehash the notice, but the essence of the notice was as follows:
The FAA learned through investigation, research, and incidents (some of which were fatal) that pilots had become too dependent on autopilot systems that reduced (or in many cases eliminated) the need for a pilot to control the aircraft directly.
What made the situation worse were the studies that showed that aircraft fuel efficiency was improved when autopilot was used over long distances. In other words, in a time when the price of fuel was having a major impact on operational costs, airline directives actually forced pilots to engage autopilot to save money.
I hope we don’t have to learn this lesson again, but that is where we are headed. So, my efforts to prevent this is to help you engage your brain and use this technology as a powerful assistant rather than simply handing over the controls.
Notable Headlines
DALL-E 3 Now Available for all Paid subscribers of ChatGPT
NVIDIA introduces TensorRT for Windows
An Open answer to GPT-4V: fuyu-8b
NVIDIA and Foxconn team up to build AI Factories
Learning, Tools, and Experiments
I’ve been playing with tools that accelerate my creator work. The two I have been using this month so far (besides ChatGPT and Perplexity) are Typset and Gamma.app.
Last but not least, I saw an interesting video for a new tool that connects with Figma called Builder.io. I haven’t even had a chance to play with this one, but it looks exciting.
As for ChatGPT, If you haven’t discovered the power of Custom Instructions, boy, are you in for a surprise, but let’s move into the framework to continue this discussion.
Using the MISTAKES Framework
For those of you who caught the most recent newsletter, you’ve had a chance to learn about the MISTAKES Framework. If you missed it, please go back and take a look. I’ll be referencing it here, but it has a more complete explainer. This is definitely my crowning achievement in 2023, and this section will become a regular part of the newsletter. The framework has an incredible amount of value, and I intend to help everyone understand how it can make a difference.
It turns out that AI makes the MISTAKES Framework even more valuable because it provides the needed structure for the building, evaluating, and using AI systems to create value.
One critical advantage of using a framework is the power it gives us to ask questions and get responses from Generative AI systems.
Eventually, I believe that advanced prompting skills will go away, and you will simply interact with these systems naturally and ask questions, but for now, it takes a bit more effort.
Here is a conversation I had with ChatGPT about using the Custom Instructions capabilities of ChatGPT. I gave it a very simple prompt:
explain how to use custom instructions in ChatGPT Plus
Before I entered this prompt, I filled out the custom instructions.
What would you like ChatGPT to know about you to provide better responses?
Our goal is to help people use technology to solve business problems.
We will use the MISTAKES Framework to answer queries; it is , a comprehensive working approach to innovation and problem-solving that allows us to work step-by-step.
1. Map: Identify the problem area, user needs, and context.
2. Investigate: Research and gather as much data as possible to fully understand the situation.
3. Sketch: Develop initial ideas, solutions, or prototypes.
4. Test: Validate the effectiveness of these prototypes.
5. Adapt: Refine and enhance your solution based on the feedback.
6. Kickstart: Implement or launch the final product or solution.
7. Evaluate: Measure the results, assess the effectiveness of the solution, and reflect on what could be improved.
8. Sustain: Maintain the solution, iterate on it based on continual feedback and assessment, and ensure it continues to provide value over time.
How would you like ChatGPT to respond?
Responses should be less than 500 words unless I ask for more.
Leep the language as simply as possible, and keep to an 8th-grade level in course content generation.
Responses should remain neutral.
The goal is always to simplify concepts and make things easy for others to understand.
For answers, we will use the structure of:
1. Explain how to use the framework step
2. What action do we take using that framework step
3. What is an example of using the framework item with the query.
4. What can I use AI to help me with for this step:
For any answer you give, also provide a suggested prompt that can be used to accelerate learning and success with the relevant MISTAKES Framework step.
As you can see from the length of this prompt, I got a lengthy response. And this is the amazing thing that happens when you apply a framework to generative AI. The answer is incredibly useful. It provides not only the way to start using it but also provides additional steps to take in adapting, using, extending, and sustaining the use of Custom Instructions. Your answer will vary because that’s how generative AI works. Still, I think you can see, without even trying it yourself, that there is tremendous value to be unlocked, not only by using Custom Instructions but by using powerful frameworks like MISTAKES!
1. **Map**:
- Identify the problem area: You want to use custom instructions in ChatGPT Plus to achieve a certain goal.
- Action: Understand the scope and capabilities of ChatGPT Plus and the purpose of using custom instructions.
- Example: Determine whether custom instructions can help in providing more directed or specific responses from ChatGPT Plus.
- AI assistance: Use AI to research the capabilities of ChatGPT Plus and to understand how custom instructions work.
Suggested prompt: What is the main goal of using custom instructions in ChatGPT Plus, and how do they enhance the interaction?
2. **Investigate**:
- Gather data: Look into the documentation of ChatGPT Plus, tutorials, or community discussions to learn about custom instructions.
- Action: Research the syntax, usage, and examples of custom instructions in ChatGPT Plus.
- Example: Find examples where custom instructions have been used to guide the model's responses in a specific direction.
- AI assistance: Utilize AI to help find and collate resources, tutorials, or forum discussions regarding custom instructions.
Suggested prompt: Can you find examples where custom instructions significantly improved the quality or relevance of the responses from ChatGPT Plus?
3. **Sketch**:
- Develop initial ideas: Create a basic outline or draft of how you want to use custom instructions.
- Action: Draft a few custom instructions based on your requirements.
- Example: Write a custom instruction to ensure that ChatGPT Plus provides concise answers.
- AI assistance: Use AI to help draft different variations of custom instructions to see which works best.
Suggested prompt: What are some different ways you can phrase your custom instructions to achieve the desired output?
4. **Test**:
- Validate effectiveness: Test your custom instructions with ChatGPT Plus to see how well they work.
- Action: Run several tests using different custom instructions to evaluate their effectiveness.
- Example: Test how well a custom instruction helps in getting more detailed answers from ChatGPT Plus.
- AI assistance: Use AI to automate the testing process by creating a script that sends various queries with custom instructions to ChatGPT Plus and evaluates the results.
Suggested prompt: How did the custom instructions affect the responses, and were there any unexpected outcomes?
5. **Adapt**:
- Refine your solution: Based on the testing results, modify your custom instructions to improve their effectiveness.
- Action: Make necessary adjustments to your custom instructions.
- Example: Modify a custom instruction to make it more clear or precise based on the testing feedback.
- AI assistance: Use AI to suggest modifications based on the analysis of testing results.
Suggested prompt: What improvements can be made to the custom instructions to better guide ChatGPT Plus towards the desired responses?
6. **Kickstart**:
- Implement your solution: Apply the refined custom instructions in your interactions with ChatGPT Plus.
- Action: Start using the refined custom instructions in your daily use of ChatGPT Plus.
- Example: Use a custom instruction to get ChatGPT Plus to provide step-by-step explanations.
- AI assistance: AI can help monitor and log the performance of the custom instructions over time, helping to identify any areas for further refinement.
Suggested prompt: How have the refined custom instructions impacted the quality of interaction with ChatGPT Plus?
7. **Evaluate**:
- Measure the results: Assess the effectiveness of the custom instructions in achieving the desired output from ChatGPT Plus.
- Action: Analyze the responses from ChatGPT Plus to determine if the custom instructions are working as intended.
- Example: Assess whether a custom instruction aimed at getting more concise answers is effective.
- AI assistance: Use AI to automate the evaluation process by analyzing the responses against a set of predefined criteria.
Suggested prompt: Are there any patterns in how ChatGPT Plus responds to the custom instructions, and how can these insights be used for further refinement?
8. **Sustain**:
- Maintain your solution: Continue to use and if necessary, refine your custom instructions based on ongoing interactions with ChatGPT Plus.
- Action: Keep a record of effective custom instructions and continue to refine them as needed.
- Example: Over time, refine a custom instruction to better suit evolving requirements.
- AI assistance: AI can help by continuously monitoring the effectiveness of custom instructions and suggesting refinements based on new insights or changes in ChatGPT Plus behavior.
Suggested prompt: How can you ensure that your custom instructions continue to provide value in your interactions with ChatGPT Plus over time?
If you enjoyed this section on using the framework, you’ll really love what’s happening inside my new course. Not only do I share more in-depth knowledge on how to use these tools, but I also share the methods that accelerate using them to get the actual work done as well.
AI Mistakes
I’ve started to schedule a weekly live session on YouTube, and I’m having fun doing it, so I’ll continue to do this for the foreseeable future. Please go check out the channel, and if you’re reading this immediately after I publish it, you can join the live session or catch the replay here:
https://www.youtube.com/live/heyFqZOghu4
Business Pro Online Cohort Course
Who is this for?
Are you a product owner, IT professional, developer, individual contributor, or analyst?
Would you like to be someone who not only knows how to use technology to but actually knows how to get things done and create value?
I will help you:
Transform your perspective and approach to problem-solving
Unlock your potential with practical AI skills and knowledge
Extract real value from everyday opportunities
Take action to make a difference for yourself and others
How it Works
Join a community of like-minded people
Learn from Daniel directly
Take action and make MISTAKES daily
Check out the full course description at the link below and join the waitlist if interested.
Learn from MISTAKES with AI
If you have questions, have comments, or are just looking for some clarity, please don’t hesitate to reply to the newsletter, or feel free to grab a free consultation with me.