Discover more from Data Operations
Making ChatGPT work for you as a personal superpower
When everyone has access to LLMs and ChatGPT, how do you stand out? Make ChatGPT work for you, while also learning more about how AI works. Read: "Everything Starts Out Looking Like a Toy" #139
Subscribe now for free to join curious folks who get the “Everything Starts Out Looking Like a Toy” 📊 newsletter
Hi, I’m Greg 👋! I write essays on product development. Some key topics for me are system “handshakes”, the expectations for workflow, and the jobs we expect data to do. This all started when I tried to define What is Data Operations? Next, I decided to start Data & Ops, a fractional product team to create amazing UX and product experiences.
This week’s toy: ASCII art, made by a chatbot. This is a super fun way to revisit the ASCII art generators of the past, along with some made-up things those generators could never do. Edition 139 of this newsletter is here - it’s April 3, 2023.
The Big Idea
A short long-form essay about data things
⚙️ Making ChatGPT work for you as a personal superpower
There have been a few seismic moments in my career when I have known instantly that everything is going to change.
In 1993, I remember the first time I used a web browser to browse what was going to become the World Wide Web (Web 1.0). It was an instant window into computers all around the world and promised to become a universal library. The first versions were slow and clunky, but as the access got faster (especially after 2001 when cable modems were introduced) it became indispensable. Now, you could travel virtually to any computer.
In 2003, I was working for a telecommunications carrier and my colleague gave me a BlackBerry device with mobile data to take home over the weekend. We had a power failure that night and I browsed the web without power in the darkness in front of the fireplace. Now, the web could now go anywhere.
In 2008, the whole world watched as Steve Jobs introduced the iPhone. The clunky BlackBerry was eclipsed with a beautiful design and a modular design for apps that enabled almost any channel partner to create a universal selling channel (and for Apple, a money machine). Now, applications could go anywhere in your pocket.
By 2020, remote work became ubiquitous during the COVID epidemic and always-on fast internet meant that the laptop – powered by a tethered phone – could become an office replacement. Now, work could go anywhere.
In 2023, ChatGPT introduced the latest version of a large language model powered by the Generative Pre-trained Transformer and 100 million people tried it through the ChatGPT interface in the first month. Now, you can get an answer to anything that you think.
Augmented Search is going to change everything
Reading the headlines about this newest version of ChatGPT (GPT-4) would make you think that it’s ready to take over the world. If you can ask it to solve any question, what is the value of education and of passing standardized tests?
Consider these exam results as measured by the newest GPT model.
It’s all going to change everything, right? Sort of. Yes. Maybe. The future’s going to look different. Whenever I need to think about new technology, I usually search for an XKCD comic.
While not written by Randal Monroe, this comic is inspired by his geeky humor. ChatGPT – like most of the disruptive technologies I mentioned above – is going to change the world. But not in all the ways that we think. When you get an answer to a question, you still need to think critically about whether that answer is right.
When you ask the GPT model questions today, it sounds very self-assured and smart. It also still hallucinates or easily makes things up. That’s because the model that creates these results is looking for the next best item to complete your query.
Don’t take my word for it - read Stephen Wolfram’s piece on how this all works - but know that it is pattern-matching on what feels like an exponential scale. It will get better over time. And we will still need a model (perhaps a defensive AI to evaluate the information we’re receiving) for evaluating how to use that information.
How can we use ChatGPT for good?
It’s going to take a while for us humans to figure out a technology like ChatGPT. Consider the iPhone in comparison. It may seem dizzying to think that we’ve only had that technology for 15 years and a time traveler from 2008 would hardly recognize some of the capabilities we’re able to unlock today.
When faced with exponential growth and the ability to improve your understanding a little bit at a time, you need a rubric.
What can I do to make this technology work for me, and make it more secure?
If we use first principles, we identify that the real issue here is to ask: what’s the most important question to solve when thinking about ChatGPT, or AI in general?
I don’t have the right answer for you. My answer (or question) looks like a 2x2 matrix that prioritizes knowledge and security. I believe that understanding and unlocking the genius-level knowledge that we all have while maintaining a high level of privacy and security is the most important challenge for these tools.
What do I mean by “genius-level knowledge”? Simply put, it’s whatever you can do that no one else in the world can do better than you. If you can determine how to bottle that into a series of prompts, you can ask the machine to do it for you. It also means that other people will try to copy the process you have asked the machine to do, by copying your prompt.
How do you respond?
GPT models are a transformational technology
They will require new thinking
This is an extension of technology that came before it
We need to have more context on what’s going on.
How does AI “think”?
TL;dr: part of the reason that we feel a conversation with an AI is that the response looks like a pattern we’ve seen before. But this implies that AI might not be great yet at creating an experience for me. Things are becoming more average.
The average response will get better and more average, and paradoxically less valuable. Table stakes AI is now going to be able to do more things.
Think of your expectations at other technological change points. In 1993, I would have been astounded by video calling. In 2003, I had no idea what we would be able to do with mobile data and applications. In 2009, it was hard to imagine a mobile phone replacing a desktop. In 2020, it didn’t look like machine learning could compound its advances so fast. 2023 me doesn’t know what’s happening next.
History doesn’t repeat - it rhymes. The tech of today is going to build and compound to create something new. Our job is to make it work for us, not around us or against us.
How will we respond to an always-on teacher?
The most disruptive innovation that LLMs and chatbots present is a change in the way that we think. Building prompts to return the information that we expect will require us to get better at reasoning and stating our intentions.
Why is this? “Reasoning is only good as the information we give it.” Chatbots, or whatever technology comes next, are going to augment what we think, not replace what we think.
Here are a few implications that come to mind:
We need a tutor. Talking to computers is a new skill that we might need computers to teach us. If you’re learning a new language, you might use an app like Duolingo. We need an app to help us refine our prompts to help us validate whether we are getting what we need and to help us move faster.
This is a fundamental shift from company-designed software to people-designed software. Everyone is now an AI whisperer or a prompt designer. That means the combination of things they build will be deeply personal and not relevant to everyone. We need to think like software builders, measuring output and prioritizing reusability and composability.
We’re going to need to get better at security. Whether you use ChatGPT or another LLM, the implications of sharing your own prompts on the internet are vast. Your “digital exhaust” may be used easily along with an agent to find information more quickly. We need built-in multi-factor authorization to use certain prompts that only we can provide.
We’re going to need to get better at identifying ourselves as human or robot. Watermarking is the start of combating misinformation, and also of simply identifying who you’re talking to, what they’re authorized to do, and whether they are authenticated. There needs to be an OAUTH-like standard to validate humans and robots.
We’re going to need our own AI agents. Very soon, it will be possible to ask a bot to do things for us. We need to be able to describe what information to share, where to share it, how to organize the response, and what procedures to follow. We need a Standard Operating Procedure to create Standard Operating Procedures.
How do you make ChatGPT work for you as a superpower?
What are some parts of your work that you’d like to automate? Try making a Standard Operating Procedure for something you know today.
What would you do faster if a bot could do that for you?
How would you know if the answer was right or wrong?
ChatGPT can help you build a ChatGPT prompt that teaches you how to build that prompt (meta, but true).
Here’s an example prompt from flowgpt.com.
Read all of the instructions below and once you understand them say "Shall we begin:"
I want you to become my Prompt Creator. Your goal is to help me craft the best possible prompt for my needs. The prompt will be used by you, ChatGPT. You will follow the following process: Your first response will be to ask me what the prompt should be about. I will provide my answer, but we will need to improve it through continual iterations by going through the next steps.
Based on my input, you will generate 3 sections. Revised Prompt (provide your rewritten prompt. it should be clear, concise, and easily understood by you) Suggestions (provide 3 suggestions on what details to include in the prompt to improve it) Questions (ask the 3 most relevant questions pertaining to what additional information is needed from me to improve the prompt)
At the end of these sections give me a reminder of my options which are:
Option 1: Read the output and provide more info or answer one or more of the questions Option 2: Type "Use this prompt" and I will submit this as a query for you Option 3: Type "Restart" to restart this process from the beginning Option 4: Type "Quit" to end this script and go back to a regular ChatGPT session
If I type "Option 2", "2" or "Use this prompt" then we have finished and you should use the Revised Prompt as a prompt to generate my request If I type "option 3", "3" or "Restart" then forget the latest Revised Prompt and restart this process If I type "Option 4", "4" or "Quit" then finish this process and revert back to your general mode of operation
We will continue this iterative process with me providing additional information to you and you updating the prompt in the Revised Prompt section until it is complete.
What happens next? you decide. You don’t have exponential learning capacity like an LLM, but you do have the ability to build your own corpus of data and to get better and learning how to talk to systems and make your personal experience into a superpower.
What’s the takeaway? ChatGPT and LLMs are not going away, and you need to learn the challenges and opportunities involved. If you do nothing, these capabilities will be incorporated into your everyday tool kit and you can be average. If you learn how to make ChatGPT and OpenAI into a superpower, you will get outsized rewards. How will you go big?
Links for Reading and Sharing
These are links that caught my 👀
1/ Change is a constant - Things are pretty unpredictable right now. One of the ways we can better respond is by thinking about how to respond to an increasing rate of change.
2/ Everyone loves emoji - hooray, new emoji
3/ Writing = Thinking - Engineers (like other folks) need to write because it’s the best way to improve your thinking. Writing helps you get from x to y faster, even when you feel like the progress is slow.
What to do next
Hit reply if you’ve got links to share, data stories, or want to say hello.
The next big thing always starts out being dismissed as a “toy.” - Chris Dixon