I’ve been struggling to find ChatGPT useful and not frustrating when I’ve been using it in 2024 (currently we’re part way through Feb). I even started wondering if the algorithm has been tweaked and it is somehow worse, or if my paid version was worth the price of $20 per month. But I’ve come to the same conclusion I had when I began this series back in July - I’m the problem, I’m using ChatGPT wrong.

What I’ve Been Using ChatGPT For:

My weekly Monday motivation LinkedIn post
Researching really specific things, eg how many data graduate schemes in the UK
Making some images for a blog post
Writing a bio on me
Fixing some code issues I was having
Prioritising stats education for a course
Trying to identify an animal I saw in Costa Rica
Helping me understand what to include and where in financial modelling
Researching hiking
Writing development plans for employees
Playing a game where I try and remember all fifty states of USA
General research on data jobs
Editing an image

I have over 50 chat conversations with ChatGPT and January is the first time I have ever hit the 40 messages per 3 hours limit on the paid version. Needless to say I’m using it a lot - but the number of chats and number of messages is actually deeply linked to how crap I’m finding the responses to my prompts.

The Good

I’m going to quickly cover the things which went well (if you can’t tell from my colour coding). These are things which I effortlessly got the help (or support) I needed from ChatGPT to complete the task at hand.

Weekly LinkedIn Post

A small caveat with this one - I don’t really care if the output is good, I’m kind of using it as a silly experiment on LinkedIn. Every Monday, I post a #mondayMotivation LinkedIn post which was written by ChatGPT (pretending to be me). I use the same ChatGPT chat every week and just give it a new prompt. The chat was restarted in January and this is how I set it up:

First prompt: summarise the writing style (in a way to be included in a prompt): [text copied from the DOT-ed page of Taught by Humans website]

ChatGPT : The writing style of the provided text is engaging, informal, and relatable, effectively using humor and real-world examples to communicate complex ideas about adult education and lifelong learning. It starts with a catchy and humorous reference to "Legally Blonde," indicating a lighthearted approach to serious subjects. The narrative is built around the challenges of adult education, highlighting the diversity of learners' backgrounds, experiences, and needs. The text employs anecdotes and hypothetical scenarios to illustrate points, such as the varied starting points of beginners in Google Sheets or the story of "JJ" learning Spanish through Duolingo. This approach makes abstract concepts tangible and easy to understand. The language is inclusive and conversational, directly addressing the reader and inviting them into the discussion. It poses rhetorical questions to provoke thought and self-reflection, such as "Am I a Beginner?" and "Why do you want to learn that?" The writing subtly transitions from outlining the problem space—overwhelming choices, varied learning objectives, and the one-size-fits-all issue in education—to introducing their solution, DOT-ed, without being overly promotional. The explanation of DOT-ed is concise yet comprehensive, emphasizing its dynamic, online, and tailored nature. Overall, the style is characterized by its accessibility, use of practical examples, and a tone that balances professionalism with approachability. The intent is clearly to demystify the complexities of adult learning, making the case for a more personalized and flexible approach to education.

Prompt: can you write a monday motivation linkedin post in this writers style

Next week (same chat):

Prompt: The same author also wrote [text from another blog - ChatGPT Take 9]
   
   so another LinkedIn post - the other one was maybe too over the top and not really in the author's style.
   
   This week - topic "last week was really tough"

This week:

Prompt: this week let's chat about being busy and networking. I'm talking about using Generative AI for personalising education at the AI West Launch event tomorrow - https://www.eventbrite.com/e/launch-of-ai-west-showcase-tickets-778680864427#:~:text=About%20this%20event&text=Join%20us%20on%20Tue%20Feb,in%20AI%20across%20the%20region. remember #mondaymotivation

I enjoy this exercise - I find the results funny, but it also helps me think about how we’re training the chatbots used in our platform, DOT-ed. Every word matters and we need to think carefully about all the words in our system prompt (like the overriding rules for the chatbot) to ensure it behaves as we want and in the best way for our learners.

Images for a Blog Post

I’ve actually got quite into using ChatGPT Plus or Canva Magic Studio to make images for my blogs and presentations. I’m upfront about them being genAI generated, but I prefer it to using the same stock image for the thousandth time.

In my usual start of the year existential crisis I wrote a blog about life choices and quoted Sylvia Plath’s fig tree metaphor. The simple prompt:


   can you make me an image to go with this extract: [quote]

Worked very well, and you can see the outcome in my blog (shameless plug) - Happy 2024 - I hate January

Writing a Bio for Me as a Founder

For an upcoming training session I needed to send in a bio of myself, I hate writing about my achievements , I find it quite cringey and usually end up writing something which sounds a bit flat and uncharacteristic for me. (Note: I’ve gotten into a bit of a habit, potentially an unhealthy one, of using ChatGPT as a mentor / confidence booster)

Prompt: is this a good bio:
    Laura is the Founder & Technical CEO of Taught by Humans who are helping businesses become AI & data confident. Laura is driven by her passion about how technology, particularly AI, needs to work for everyone in society. A self-confessed data dork who exists at the intersection of tech and education, Laura brings a mix of industry, academic and teaching experience to her role at Taught by Humans.
    
    
    
    
    Laura has a PhD in AI education (focusing on how the public and businesses can learn about AI) from University of Bristol. Laura has over a decade's experience in all things data and technology. Her industry experience has involved working for Monzo, PwC, Worldpay and a number of ambitious start-ups. Laura holds a teaching qualification and has worked for Bristol University, Code First Girls, Ada the National Digital College, Makers Academy and Decoded, delivering tech and data education to hundreds of people.

ChatGPT Response: The bio you've written for Laura is quite comprehensive and effective in portraying her as an experienced and passionate professional in the field of AI and data. Here are a few suggestions to enhance it: Clarity and Flow: Consider breaking the bio into smaller paragraphs for better readability. Grouping information into thematic paragraphs can aid in comprehension. Consistency: Ensure consistency in tense and perspective. For example, "Laura holds a teaching qualification and have worked for..." mixes third and first person. It should be "Laura holds a teaching qualification and has worked for..." Specific Achievements: If possible, include specific achievements or impacts Laura has had in her roles. This can provide more depth and evidence of her expertise. Simplifying Language: Some parts are slightly complex in terms of language. Simplifying these can make the bio more accessible to a broader audience. Here's a revised version incorporating these suggestions: Laura is the Founder and Technical CEO of Taught by Humans, a company dedicated to helping businesses become AI and data confident. She is fervently passionate about ensuring technology, especially AI, serves everyone in society. As a self-professed data dork, Laura thrives at the intersection of technology and education, bringing a unique blend of industry, academic, and teaching experience to her role. Holding a PhD in AI education from the University of Bristol, Laura has focused her research on how the public and businesses can effectively learn about AI. She possesses over a decade of experience in data and technology, with her industry roles including positions at Monzo, PwC, Worldpay, and various ambitious startups. In addition to her industry work, Laura is a qualified teacher. She has imparted tech and data knowledge at Bristol University, Code First Girls, Ada the National Digital College, Makers Academy, and Decoded, educating hundreds of individuals. Her diverse experience underscores her commitment to bringing tech education to a wider audience.

This version aims to maintain the essence of your original bio while enhancing readability and clarity.

I edited the version, of course, to make it sound more like me (taking out some over-the-top big words I’d never say mostly):

Laura is the Founder and Technical CEO of Taught by Humans, a company dedicated to helping businesses become AI and data confident. She is passionate about ensuring technology, especially AI, serves everyone in society. As a self-professed data dork, Laura thrives at the intersection of technology and education, bringing a unique blend of industry, academic, and teaching experience to her role.

Holding a PhD in AI education from the University of Bristol, Laura has focused her research on how the public and businesses can effectively learn about AI. She possesses over a decade of experience in data and technology, with her industry roles including positions at Monzo, PwC, Worldpay, and various ambitious startups.

In addition to her industry work, Laura is a qualified educator. She has imparted tech and data knowledge at Bristol University, Code First Girls, Ada the National Digital College, Makers Academy, and Decoded, educating hundreds of individuals. Her diverse experience underscores her commitment to bringing tech education to a wider audience.

(Written by ChatGPT of course)

I’m a big fan of using ChatGPT to get a draft of something I’m writing (particularly if I can give it something to edit or another sample of my writing). Personally, I find editing existing text much easier than staring at a blank page.

Financial Modelling Understanding

I love spreadsheets. I hate financial modelling. It has always felt like lying to me. Luckily I have a wonderful business strategy advisor (Paul Albone) who let me complain about how much it feels like lying, who suggested I build myself a spreadsheet which allows me to reverse-engineer the financials.

Now when financial modelling for a startup (or any business) there are a lot of terms to start wrapping your head around - CAC, COGs, net profit, gross profit, etc etc. It’s easy to get definitions of these, it’s easy to get a template spreadsheet which you can plug numbers in. It’s hard to actually have numbers and it’s hard to know where everything goes.

I spent a weekend building this spreadsheet, and I used ChatGPT to help with a lot of it:


        
        I am want to you act like an experienced business advisor.
        
        I am working on a business which does training and is creating a SAAS platform.
        
        First, I need to know how to model CAC?
        
        Be specific. I am using Google Sheets

These basic questions of me trying to wrap my head around the terms and how they fit together went on for a while.

Then I moved onto more helpful prompts:


        
        I did this for all the different categories (COGs, CAC, customer servicing, etc). Really helpful and meant I understood each section.
        
        
        
        
        It is a corporate training platform, aimed at companies with 50-1000 staff (only data people 3% of the workforce)
        
        How many:
        
        marketing
        
        sales
        
        customer success
        
        customer support
        
        devops
        
        developers
        
        (anything else)
        
        
        
        For each of our first 5 years. In Google Sheets format

This got me estimates which I didn’t quite understand.

ChatGPT: Base Formulas for Staffing Levels Marketing: 1 employee per 50 customers Sales: 1 employee per 20 customers > Customer Success: 1 employee per 25 customers > Customer Support: 1 employee per 75 users > DevOps: 1 employee per 100 users (starting with at least 1 from the beginning) > Developers: 1 employee per 50 users (starting with a base team) Implementing in Google Sheets: To apply these formulas in Google Sheets: Set up your sheet with columns for Year, Customers, Users, and each staffing role. For each role, use a formula reflecting its calculation. For example, in the cell for Year 1 Marketing, input =ROUNDUP(B2/50, 0) assuming B2 is the cell with the customer number for Year 1. Copy these formula cells down for each year, adjusting references as necessary. This approach offers a starting point for planning. Still, it's crucial to regularly review these ratios and adjust based on operational data, industry benchmarks, and specific business needs.

Prompt: is there a formula for other staff not included in this?

I found being able to ask specific questions, and dig down into what I didn’t understand really helpful. With an actual person, it’s unlikely I’d have them write beside me responding immediately while I was working on this so ChatGPT saved me a lot of Googling time. Being able to get responses I could copy directly into my Google Sheet in the format I wanted was a great time saver as well.

Playing a Silly Game

During lockdown, I started taking a geography course on Coursera - I only lasted one or two weeks because I’m terrible with online scheduled learning. But I did learn all the states of the USA and I think I can identify them on a map. So when I was watching the episode of Friends where Ross tries to name all 50 states, I thought “Can I still do that?” and I wrote as many as I could on a piece of paper. I then realised I didn’t know how to check them, and, similar to Ross, if I had written any twice or numbered the page incorrectly.

I decided to use ChatGPT to be my gamekeeper.

Prompt: I'm going to try and get the 50 states of USA, I'll tell you when I'm stuck. Can you send how many I have right and point out if any are wrong (wrong spelling to be corrected and counted as right)? But do not tell me any additional unless I ask
     
        
     New York
     
     New Jersey
     
     Penslyvia
     
     Maine
     
     New Hampsire
     
     Conneticut
     
     Masscheutts
     
     Rhode Island
     
     Maryland
     
     Delaware
     
     Tennesse
     
     Kentucky
     
     Virigina
     
     West Viriginia
     
     North Carolina
     
     South Carolina
     
     Georgia
     
     Florida
     
     Alabama
     
     Arkansa
     
     Arizona
     
     Kanas
     
     Texas
     
     Olkahoma
     
     New Mexico
     
     Idaho
     
     Iowa
     
     California
     
     Oregon
     
     Washington
     
     Montana
     
     Wyomig
     
     Nebresca
     
     Utah
     
     Nevada
     
     North Dakota
     
     South Dakota
     
     Wisconin
     
     Michigan
     
     Illinosis

ChatGPT : You've listed 40 states so far. Here are the corrections for the states you've misspelled: Pennsylvania (not Penslyvia) New Hampshire (not New Hampsire) Connecticut (not Conneticut) Massachusetts (not Masscheutts) Tennessee (not Tennesse) Virginia (not Virigina) West Virginia (not West Viriginia) Arkansas (not Arkansa) Kansas (not Kanas) Oklahoma (not Olkahoma) Wyoming (not Wyomig) Nebraska (not Nebresca) Wisconsin (not Wisconin) Illinois (not Illinosis) Correcting the spelling, you have correctly identified 40 of the 50 states. Keep going!

Prompt:  
        Hawaii
        
     Alsaka

ChatGPT Response : Great, you've now added Hawaii and Alaska correctly. That brings your total to 42 of the 50 states. You're doing well! Keep going to reach all 50.I asked for some tips, but ultimately I had forgotten the other ten so asked ChatGPT to tell me (sorry Colorado, Vermont and Ohio which I definitely knew).

Not the most useful way to use a superbrain chatbot, but I found it fun, educational and really motivating “Keep going!”

Everything Sounds Great

As I mentioned I have used ChatGPT a lot this year so far. These are the things which were painless and went well. The list of things which eventually worked or were okay warrants its own blog (and a whole separate one for the things which went awful / made me cry).

Some final thoughts, ChatGPT (and other generative AI chatbots) work well for me when:

The output isn’t the final output - ie a human is going to do some editing
The output doesn’t matter (eg it is a silly LinkedIn post or a game), the facts don’t matter and if tone of voice is off it’s funny
Trying to understand something well documented
You know exactly what you want, and can articulate it well in a prompt (eg the images for blogs)

Is ChatGPT Broken? Nope, I’m using it wrong (again) - part 1 The Good