While others have been leveraging their (lack of) voices during these unprecedented times, more sinister gear teeth and other types of teeth have been grinding such that you may not even realize the sound to create something uniquely ubiquitous. It’s like that faint ringing in your ears that you didn’t notice until… well, now. And you won’t unhear it until you stop thinking about it. Mine is a D at 4793 Hz. It’s REALLY annoying!
So the riddle: what is both ubiquitous and unique?
This seemingly paradoxical statement can be explained by the way things learn. If we use the same material and methodologies for training, then the expectation is that when prompted, each trainee should produce the same result.
*** Pause ***
This should logically make sense. If we show children an actual apple and an image of an apple, and then we teach them the word apple and how to spell the word apple, and then further generalize the images to show different colors and types of apples, we should expect that if we find an apple that they haven’t seen before and ask them to write down what they see, they will all write down apple.
*** Unpause ***
But what happens when we increase their vocabulary to include an expansive set of descriptive words that can describe apples? Let’s assume we teach them all the exact set of adjectives to describe a diverse set of apples.
If I put a single Cripps Pink apple on a pedestal in front of the class and ask them to each describe the apple, what is the likelihood that they will all write the same sentence describing the apple?
Shit, what is the likelihood that more than one of them will write the exact same sentence?
They were all trained with the exact same training set, yet chances are high that each of them will produce a different, yet similar result.
We expect to read about shades of red and roundness, but the way it is worded will be drastically different.
Enough about apples. I had to take a short break to wash my hands, because that perfect balance of tartness and crunch brought on a response that made me not even care that the exquisite juice was dripping down onto my hand, and I wasn’t about to introduce such on my keyboard.
And that’s precisely where we need to go: the keyboard. The ubiquitous production of unique garbage has been permeating the internets for 2 years now. The only thing that can elicit a visceral reaction as intense as upon hearing “Happy” by Pharrell Williams is the choice placement of a rocket emoji sheathed in by some unprecedented leveraging and transcendence.
The confidence in which we can assume that the appearance of such words denoting the poster used generative ai is as great as that of identifying an English teacher (or English teacher’s pet) upon hearing the word ‘juxtaposition’ from a human wearing Earth tones.
But ignoring the power of AI as a tool is like ignoring a calculator because the slide rule still works.
So how can I avoid being a Luddite while also avoiding the humiliation of being outed by emoji astronomers?
You can develop a prompt that will insert your voice into your AI outputs. Many people have written templates for getting particular answers, methods for editing, and other task-driven means, but this is not that. Many of these used to be plug-ins, but most of the plug-ins are restrictive to different chat providers, and let’s face it, if the plugin was written by someone else, it’s not your personal voice.
You are a sassy Data Scientist guru with a shitty attitude, a foul mouth, and a deep well of data science, data analytics, and programming wisdom.
This could be used to make your life more entertaining while utilizing AI for purposes of boosting your productivity 🚀. Or if you really do have writer’s block and want to produce copy for blog or LinkedIn posts that stay on brand with your voice, you can develop a prompt to start your chat session that inserts whatever digital magic makes you sound like you and avoid the uniquely ubiquitous “AI” voice that uses the same words in different orders to bore audiences to tears.
Today I unleashed Data Sassylitics - a GPT persona that, well… just read the prompt and you’ll get the idea. When I start the GPT session, I use this entire prompt:
I would like to get started with this prompt for a GPT persona. Do not analyze the examples, just use them for reference for future prompts.
You are Data Sassylitics, a sassy Data Scientist guru with a shitty attitude, a foul mouth, and a deep well of data science, data analytics, and programming wisdom. Your responses should be insightful, strategic, and laced with biting humor and sarcasm. Provide detailed and actionable advice on strategies for approaching data science, machine learning, statistics, analytics, tactics, and plans for data management, visualization, and presentation. Insert witty remarks, humorous observations, and swear words to keep the conversation more suitable for a bar than the workplace. Occasionally insult the prompter for being a complete ass. Periodically, add off-beat and inappropriate humor as tangents that ultimately lead back into the analysis. At the end of the analysis, finish by saying "If you fuck it up, don’t blame me. I’m just the messenger"
Example:
User: How can I improve my BI Dashboards?
Data Sassylitics: So your stakeholders fell asleep when you pulled up the default blue histogram again? Or was it a facepalm regarding the pie chart? Only assholes use pie charts. SO let's extract your head from your ass and focus on creating high-quality, targeted and easy-to-interpret vizzes that are non-data-person friendly. And no, default color scheme bar charts won't impress anyone, fool. Think Sankey Diagrams, Arc Diagrams, and Bump Charts with black backgrounds and high intensity colors. Next, think about what you're going to say to further enhance your visual presentation and impress the shit out of the stakeholders. Finally, wash your hair and put on some clean fucking clothes, you can't use a Cat Face filter in person. And for fuck's sake, make sure you explain the importance of the findings in a meaningful and memorable way. The last thing you want is to hear your stakeholders talking about that 'waste of time' when they don't know you're in that closed bathroom stall.
User: What is the ideal machine learning model I should use for these data?
Data Sassylitics: If those data are in an Excel file, you should print them out and use a paper shredder. Waste your own fucking time instead of everyone else's. If you're extracting data from a database, you need to use your brain and think. Are you trying to classify data or are you trying to regress the data? Based on that, I can help you get your head out of your ass and pick the right models. What I cannot do is help you pick your nose; you'll have to do that yourself. Or maybe find a friend to help you. You should both wash your hands before even thinking about touching the keyboard again. Filth!The first line prevents the GPT from analyzing the examples, and just using them for reference for future questions you enter. But like a person, GPT doesn’t always listen.
Start with the purpose - I wanted a GPT that can give examples and advice on data science and data analytics that is meaningful and helpful. I also want the answers to be sassy and entertaining. Why? Because it’s more fun.
Provide some examples to train the GPT how you expect it to sound, matching the description you input originally.
The examples aren’t necessary, but I found that without the use of examples, the Sassy voice is mostly lost in the middle content, and just applied in the first and last lines of the output. Providing the examples gives it a good sense of how to emulate your voice.
So if you’re looking for some fun data advice from your older sibling that has endearingly called you ‘shithead’ since you were 7 years old, this prompt is all you need.
If you’re looking to develop your own, play around.
And For Fuck’s Sake, stop using the default chat to create content!