Permanence and Transience – Experiments in GenAI

No, GenAI will not write a song as good as Bob Dylan's worst

Permanence and transience are words we often used to describe things in the physical and virtual worlds. They’re also handy words to describe creative works like painting or music. You can also apply them to how we think of work.

I was thinking a lot about those words in the past year as I’ve been working with OpenAI’s GPT platform. Building a bunch of virtual representations of colleagues at Content Evolution. (Here’s a post from my colleague Kyle Shannon, CEO/founder of Storyvine, and Content Evolutions’ chief generative officer, announcing the project we are working on.) These CE GPTs are intended to give prospective customers, or potential CE members, an ability to engage with the concepts and POVs our members have developed and applied to strategic business, and in some cases state governmental challenges and opportunities. 

What I wrote/published 10 years ago for my employer says more about them than me

Finding: Chris was right

Old work artifacts aren’t as important as what you learned from making those artifacts. I used to be so proud of my package of “clips” I’d update and trot out to prospective employers. That was until I started doing corporate PR and communications for tech companies in late ‘80s and early ‘90s. I worked with one VP, “Chris,” who covered a lot of roles at the company.  We were out having a drink after I’d been there a few months and I mentioned how I thought the clips were important to my hiring.  He quickly disabused me of that notion saying that he assumed I knew how to write and seeing stories published two years before wasn’t particularly relevant. They just showed I was employed by the publication at the time I noted on my resume.  What was important to him was how I responded to questions about what I’d learned during those years.

“Training” AI platforms on my old clips, presentations and research reports only helps somebody understand me in a strict chronological sense. But an exchange with my GPT is more likely to give you a more accurate representation of me and how I can help a company than my resume on LinkedIn. 

Don’t be too concerned with your “voice”

Finding: Your “voice” is hard to program (so far)

Am I concerned that the GPT doesn’t always capture my “voice”? Not particularly. To be overly concerned about that is to miss what I see is the real value of a GPT digital twin in the business context: it’s not meant to be a replacement for me. Think of it as a stand-in for me through which you can an idea about how I can assist you or your business by asking questions about your business opportunities or challenges.*

Also, it can be difficult to really define your “voice.” In this context I’m talking about print/text /audio/video artifacts that we use to train GPTs and what they convey about our personality, word choice, syntax etc.  In other words, our own view of our “voice” is likely pretty subjective.  What I’ve been trying to do with one of my GPTs is micro-tuning to get at what I think my “voice” is at this stage of my life.

Even if you have a rich and long list of publications, appearances, interviews etc. – you’re still going to be surprised

This is a sort of corollary to the previous “findings” I’ve noted: if you have published a great deal of material and it’s not sitting behind a paywall or in your company’s IP portfolio, you do want to use the GPT to provide an easy-to-use interface to allow a stranger to access your work. I may not want to read everything by philosopher, Luciano Floridi, who has written extensively about the ethical implications of technologies but, damn, this bot a student built gives me a great way to find out more about his work. And who knows, maybe I do start reading everything he’s written.

That ability to discover more about Floridi and his work also applies Floridi himself. The professor notes in the story that one of the things that surprised him is that the bot has literally reminded Floridi about things he’d written years ago.

Making a GPT for yourself or your business is easy – making one that delivers value (not just time-wasting ephemera) is hard

it’s easy to make a GPT for yourself or your business because the actual mechanics of a build are trivial. It’s what information you use to train your GPT and how you focus on prompt-crafting focused less on hyper-specificity of your “bio” and more evocative prompts that enable the GPT to contextualize a response to a question about current issues that reads (or sounds) like you addressing it in real-time, not how you would have answered it at some point in the past. It’s a trickier proposition than you would think even if you’ve been published in many publications, websites, books etc.

We at Content Evolution are continually reminded of this lesson and it’s forced us to create an approach to delivering business value with the GPT-builder tech that does more than deliver an automagical, conversational-ized version of a resume or a company mission statement. 

So I’m going to keep mucking about with this stuff for a bit longer. There are more than a few ponies in there somewhere. 

(*)Note: I did not train my GPT on any of my work done for my previous employers but I was quoted in industry and trade press interviews and spoke at events that were recorded.) _

(Want to chat about how I can help you conceptualize and enable your GPT (or GPTs)? Ping me at: m2@straydogconsulting.me )