Borrowed Time: AI Chat With B2B Techfluencer Evan Kirstel

Borrowed Time: AI Chat With B2B Techfluencer Evan Kirstel

Evan Kirstel, a B2B tech influencer, recently interviewed Christopher Penn, Chief Data Scientist at, on his podcast. They discussed the rise of generative AI and its potential impact on marketing and other industries.

Penn has been involved with analytics and data science for over a decade. He got interested in AI around 2017 with advances in deep learning. The latest breakthroughs in transformer architectures like GPT-3 have enabled generative AI models that can write, summarize, translate and more.

There are many startups building products on top of models like GPT-3, but Penn believes most will fail unless they add unique value. He focuses on use cases and open source tools that give more control vs relying on third party services.

For marketers worried about losing their jobs, Penn says AI won’t replace jobs directly. However, people skilled at using AI will be far more productive than those who aren’t. Marketers need to skill up and integrate AI into their workflows to avoid displacement.

Penn sees the biggest near-term impact of AI in improving marketing operations. It can help with scheduling, email, status updates and other repetitive tasks. But he cautions against overusing it just to generate more content.

His advice for young professionals is to develop cross-disciplinary thinking, which AI still struggles with. Taking varied classes in literature, history, etc. builds the nuanced understanding of humanity that AI lacks. But everyone also needs to learn how to use AI tools.

Penn predicts quantum computing will eventually lead to machine consciousness, but not for at least 5-10 years with today’s AI architectures. He expects job losses but also new roles where humans are still preferred. Climate change is a larger concern than AI in the next decade.

Chatting with Christopher Penn @cspenn: Co-founder & #DataScience @TrustInsights, @mktgovercoffee

Can’t see anything? Watch it on YouTube here.

Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.

Evan: Hey, everyone. Super excited for this chat today with Rockstar, with legendary author, content creator, analyst, data scientist, Christopher Penn. Chris, how are you?

Christopher: You know, I can’t complain. It is, it’s 2023. Everything seems, you know, I just got through Fourth of July, I can still count to 10 on my hands. We’re all good.

Evan: It is good. And this is a special Boston area edition, where both in Boston know that we haven’t seen each other in five, six, seven years. So it takes, you know, social media to bring us back together again.

For those few folks who don’t know who you are, maybe share your journey into becoming a marketing data scientist, keynote speaker, author, etc. And we’re going to dive into AI, which I’m particularly excited about.

Christopher: Yeah, I mean, I started in it. So in the late 90s, and in the early 2000s, I was in it. And then in the 2000s, as many folks know, it became marketing technology.

And I sort of had my career pivoted to go with that, you know, running send mail on a Linux server became, you know, send the email newsletter, and so on and so forth.

And then in the early 2010s, I moved out of the financial services company, I was in into a PR agency. And one of the things that I got started there with was taking my experience in analytics and starting to move it into data science.

And of course, that logically gave way to artificial intelligence and machine learning mostly. So you have things like hell, how much traffic to the client’s website get? Well, let’s can we create a forecast model, you know, using at the time, a Rima and Arma and all those things. Classical algorithms. And over the last 10 years, AI has evolved. Shockingly, we first have had huge growth with deep learning with things like multi layer perceptrons and stuff.

And then really, the thing that has seems to have launched into the stratosphere, it was a 2017 paper called Attention is all you need, which is was the release of the transformer architecture, and alongside at the diffuser architecture.

So these two AI fundamental foundational technologies came out in 2017. And since then, they have been the underpinnings of everything you’re seeing with stuff like chat GPT and Dolly and stable diffusion, you know, AI created art and writing and poetry and all this stuff, all that’s predicated on those two architectures.

Evan: Absolutely. And it’s the most exciting time to be a content creator. And you must be every day must be like a kid in the candy store for you, in particular as such a practitioner. Before we jump into that, tell me about your content, you have a weekly newsletter, it’s highly praised.

You have AI for marketers, which are practical applications for AI in marketing, what else what did I miss? I there’s so much stuff that we create.

Christopher: So I’ve got two podcasts, one is marketing over coffee, which has been on the air since 2007.

  1. That’s a weekly show. And then the In Ear Insights podcast is the trust insights podcast also weekly, and then there’s a Thursday live stream called So What the marketing analytics insights live show, which is the company live stream. So there’s making content all the time, following a strategy that my friend and former CEO Todd Deferon talked about way back in 2008. The idea of content atomization, where you make a piece of content, you break it up into more content.

But we really pivoted I pivoted, starting in about 2016 2017, to being video first, because video is the richest channel that has the most information. And then from video, you can take, you know, snippets and make Instagram reels or whatever. But you can take the audio and now you’ve got a podcast, you can take the audio and put it through pieces of transcription software. Now you’ve got text. And now with generative AI, you can take that text and have it summarized into blog posts, into emails into social media posts. There’s so many different ways to take one piece of video content and just break it up into all these pieces that you it really is the best way to generate a lot of content. And then with AI in your pocket is you know, that’s a force multiplier that allows you to really dramatically accelerate the amount of the total number of pieces of content you can create, and then publish where, wherever your audience wants you to be.

Evan: That’s a wonderful tactic and best practice. So when degenerative AI could have come onto your radar, and how have you managed to consume all of this frenzy of news and updates and analysis and startups and applications that are just coming out of the woodwork? I mean, it’s really hard to analyze what’s going on as it’s not just about barred or chat GPT or open AI, there is a thousand flowers blooming at the moment.

Christopher: There are 1000 flowers blooming and 990 of them will wither and die in six months.

Evan: Thanks for that optimistic note. I was kidding.

Christopher: It’s like any, any startup, there’s a tremendous number of companies now that are essentially just a UI on someone else’s technology, right? There’s not a whole lot of compelling value proposition above and beyond that those companies that survive will find some way to add value above and beyond what you can do. But you know, my my journey with generative AI started around 2020. When opening, I released a model called GPT two back then, I think it was GPT two.

Evan: Wow, you were you were early on as a marketeer.

Christopher: Because it caught my eye that this thing could generate text. And then you had a illithra AI is GPT j 6b model, the 6 billion parameter model. And that was sort of the beginning of the open source, large language model revolution. And these two things have kind of marched in parallel. And you start seeing more and more growth from you know, these models, you’re seeing very large models coming out of big tech companies, you know, Facebook has released its llama model, Google has palm two, of course, open AI has the GPT for 3.5 and for family anthropic has clawed to now. But then you also in parallel have this huge ecosystem of open source projects. So Facebook open source, the llama model to the community and that has created 1000s 1000s of new models, you know, derivatives and things that people have created forks, there’s wizard LM.

The way I think about the technologies and how to keep them organized, how to catatune through the clutter is twofold one. I look for use cases. Like what, what do we want to do with this stuff? Like a real simple example, we’re, we’re talking you and I right now in on a live stream. If either one of us says something notable, we’d want to make know that so we want to maybe to get this transcribed suit. So then okay, well, what AI technologies is operating in the transcription space whisper which is open AI is open source product is by the way, a phenomenal product. It’s one that I’m writing it down. Another tip from Chris Ben. It’s free. It’s open source that runs on your laptop, which is as long as you’ve got a good enough laptop or any any laptop you can play good video games on you can use this technology. And then it does the transcription for free.

So yeah, you have all these services like you know, you know, whatever dollars for how many minutes like now I’m just going to run it locally on my machine and you know you and you can just do crazy stuff with that. So transcription makes a logical outcome from our conversation. And then summarization makes a logical outcome. So I’d want to look at AI model that had can handle what 45 minutes and hours worth of conversation and boil that down and maybe post but also have the original text. So now I need to think about okay, well, how what kinds of technologies can handle that much text? A good example, that would be anthropics, Claude to model which got released yesterday. This can handle 100,000 tokens at a time which if you’re not familiar with, you know, AI terminology, that’s about 65 70,000 words.

Evan: Wow. So yeah, breaking new ground here supply chain and marketing are not two things typically get taught in the same sentence. Fascinating. You know, talking about you do a lot of educating people like me or your your clients and beyond but what would you give as advice to young people who are looking at marketing and wondering how this career is going to be affected by gen AI and just generally how do they get ahead of this wave beyond you know, obviously consuming all of your content and others? What advice would you have for them in their 20s perhaps?

Christopher: So there’s there’s a couple of things that machines really don’t do well and won’t do well for a while. One of which is they are very, they’re still not very good cross disciplinary thinkers that even with the largest language models, they still don’t exhibit truly emergent cross disciplinary thinking. So it’s very difficult for them to come up with things that humans do through our intentionally flawed memory mechanisms, right? When you’re trying to think of what to create, creating is inherently tied to memory and our memories are flawed in that we only tend to remember things that are high emotional valence, right? We don’t I don’t remember what I had for lunch two weeks ago on Thursday. It was not something that made a huge emotional impact on me. Do I remember what I had for for dinner at my wedding? Sure do. Because it was a very emotionally key thing.

So our memories are tied to emotion machines don’t have that machines have essentially perfect memory. But part of perfect memory means that no one memory is more important than other memories. And so when it creates is not creating in the same way that humans do it is our our dependence on emotion that creates memory loss. And that memory loss is what allows true creativity to kind of fill in the gap. Machines will get there. There’s early work and doing this, but it’s still not quite the same.

So if you are a young professional, or you’re maybe you’re in school right now, you need to be looking at having as many cross disciplinary experiences as possible. Like, take that 19th century French literature class take that, you know, intro to Islam class, take all these things that will give you a better and more nuanced understanding of humanity, because humanity is what the machines are calibrating towards and there, it’s very difficult for them to do that. Because we are such weird creatures.

The second thing is you’re this is an acknowledgement, everyone has to get skilled up on the use of these AI tools, you have to know it because the number here’s the part that folks don’t want to talk about the jobs that are going to be affected the most are the entry level jobs, right? If you have someone whose job is just writing press releases, say at a PR agency, well, guess what, the machines can do that in almost entirely now. So you don’t need humans to do that anymore. We do need humans to edit it to QA it to like, hey, you made up a quote from a CEO doesn’t exist. This is probably not something we should do here. But there will be far fewer jobs available at the entry level because machines will be doing so many more of them. So if you are one again, if you’re one of those people who are skilled with AI, and your peers are not, you have an advantage, you will be one of the people who, you know, a hiring manager will say, Well, you know, why should I hire you versus having a machine do it, you could say because I am good at working the machines. And I can dramatically increase your productivity and your results, whereas all the other people who are competing for the same job, they can’t do that. And this is every field.

What happened this past week was just absolutely stunning. Open AI opened up code interpreter, which is part of the GPU for the system in the paid version of chat GPT code interpreter is the dumbest name for the most brilliant product ever. It is a junior data scientist is what it really is. You can take for example, you could export, say your personal finance, maybe export your bank account data, right in a CSV file, and you insert it into code interpreter, be sure to turn off logging so that you’re not handing open a higher financial data. But then you say, run a financial analysis on this data. And tell me, you know, where am I wasting my money every month, and it will go through and it will write code, Python code to do that, and then show you the analysis it does, right?

So if you are a person who is skilled with these tools, you can fill a ton of different entry level roles, think about, you know, bookkeeping, you with the help of code interpreter could be a good novice, you know, no, no years experience bookkeeper in the workforce with this tool, even though you never went to school for it, because the machine is good enough at that junior level task. So everyone needs to understand this stuff. But people who are junior in their careers, most, they need to understand the most of all, because they will be far fewer positions available.

Evan: Great point. I hope folks are listening and taking note. You’ve given us so much wisdom and insight, but any anecdotal stories or case studies of maybe your clients or otherwise who’ve been using generative AI really effectively in their marketing campaigns or in content. I mean, we’re all using it to some degree, but where is it having the most impact? Would you say the most impact is having right now is in marketing operations, right?

Chris: It is in being able to help people do their jobs faster. Everything from, you know, building schedules and calendars, replying to emails, creating commodity content, like here’s our status update.

One of the places we use it, every, every reporting period of one of our clients is we take in hundreds and hundreds of pieces of feedback from one of our clients, website survey systems, where as you know, simple satisfaction surveys, and we say, Okay, summarize this content into the top five categories in the top five, top five positive and top five negative categories of feedback for the customer’s website. And so instead of having to read literally 22,000 pieces of feedback every month for this client, they can look at the top five issues, positive and negative and the percentages, and the machine is summarizing all the stuff so well and so easily that allows them to make decisions very, very quickly.

So op as much as people like, Oh, yeah, generate infinite amounts of content, like, yes, you can. But that’s kind of like taking, you know, a Porsche 911 to the grocery store, like, yeah, it does the job, but it’s kind of overkill. But in operations in streamlining and giving you access to stuff is where it really shines. The other place that it shines and is so underused is in professional development. People don’t think of these tools as professional development and training tools. And they are exactly that. I’ll give you an example. In large language models, there’s these two concepts called parameters and weights, the number of parameters that a model has in the model weights. And if you read the technical explanations about it, it’s like, okay, this is, you know, here’s how these things are calculated. And here’s the mathematics. And for a lot of people, that explanation just goes into slides right off their brain, right?

Go into a tool like chat GPT, for example, and say, explain within the context of large language models, parameters and weights in terms of pizza. And it will say, if a large language model is a pizza, the parameters, the variety of the ingredients of toppings, and the weights are how many of each topping there is like, Oh, I get it now. Now I can speak intelligently about this. If you’ve been in a meeting or at a conference or any place where you don’t want to ask a question out loud, because like, Oh, God, everyone’s gonna think I’m dumb, right? You know, my team should think I will think I should know this already. You just pull up your phone, you type the question into chat GPT and say, explain this to me and then five minutes they’re like, now I know what you’re talking about. And I can participate in this meeting again. It is such an underrated tool for helping people get up to speed very quickly.

Evan: Oh, such great advice. As we wrap up here, let’s take a look a bit longer out maybe five years, give us maybe an upside and a downside scenario, best case worst case on how we might be living in five years with generative AI in our daily lives. What do you think? Couple predictions.

Christopher: I can’t even tell you five months, right? If you think about it, chat GPT was released in November of last year has not even been a full year. And this this circus train has just gone, you know, off the rails and into the sky. We’ve got, you know, models popping up everywhere. We’ve got thousands of new companies, we have all sorts of crazy emergent properties happening in the largest models. I have no clue. What I can say is this, the current architectures will not permit true consciousness, right? It will not permit machines to be self aware, this is computationally not possible with the current today’s architectures. The system that will allow that is quantum computing. Because quantum computing is essentially the way they work is massively parallel like our brains, right? Like the gray matter in here, we are our brains are essentially really slow, but extremely complex parallel processors. Quantum computing allows that but at a much faster pace, assuming we can stabilize them right now that about 1000 qubits or so, which is like 1000 brain cells. And you have like billions, if not trillions of brain cells in your head that create that that interlinking complexity creates emergent properties like consciousness. Once quantum computing finally gets up to speed and can start tackling things like language models, then you have the conditions for consciousness.

In terms of what’s likely to happen, we can count on two sets of properties that are going to be consistent, right? Everybody wants better, faster and cheaper. So if you’re wondering about the motivations of any company in its use of AI, it’s going to want those things, right? And people are generally greedy, stupid and horny. So we can expect that any product that serves those things, those those human impulses is going to do well, then people will use AI for those applications, whether you want them to or not. There will be substantial job losses, but there will also be substantial job creations. As people find services in places and things where machines don’t work well, there’s a new service, for example, in Japan, someone whose job it is to help you quit working at a company because of complex personal relationships, like that’s a service. It clearly is.

The big thing that in a five to 10 year time span, that is a much greater, more pressing problem that people need to be planning for now, in terms of supply chain and business continuity, is climate change, climate change is happening so fast. We’re in what is essentially a an accelerating feedback loop. As things get warmer, things that create conditions for increasing warmth get worse. There’s a whole bunch of methane trapped in the permafrost around the Arctic Circle, and methane, methane deposits essentially in the ocean. As the planet warms up, this gets released, which creates warming even faster. And there’s not a whole lot, you know, that to stop that particular train. As a result, things like these wildfires that we’ve been having wildfires are going to be getting worse, they’ll be around longer, they’ll be much bigger. And so even something as simple as you know, all those masks that we bought for for the pandemic, we’re going to want to have them on hand, because that’s how you block at least the particulates from from wildfires, but all the supply chain stuff we’ve been seeing rippling since the pandemic was going to continue, it’s going to get worse.

So companies need to be thinking about from a business continuity perspective, a, how can I build redundancy and safety in my supply chain? And B, how can I use technology to communicate faster with my my stakeholders, my customers and things so that they are informed faster, I can keep customers happier for longer, knowing that it’s going to be more and more challenging to provide physical goods and services.

Evan: Oh, such a great point. I was just talking to a big telecom customer of mine this morning, who is using drones and third party fire detection apps to actually detect fires before they spread and to deploy drones auto magically to, to put them out. I mean, so the use of AI and drones and 5g and IoT and all this tech is coming together for good. What’s been such a delight chatting with you? What are you what are you up to the rest of the summer? Personally, professionally? Any any travel ahead?

Christopher: Oh, tons of travel. So in a couple weeks, I’ll be in Cleveland for the marketing AI conference. So I’ll be keynoting that talking about large language models. And then in September, I’ve got a like, I’ve got one week or there’s four different events that week. So there’s content, jam, content marketing world, marketing analytics and data science conference and a private event. And then I’ve been doing a lot of private talks at companies just trying to help these companies get up to get their employees up to speed on generative AI as quickly as possible. So that’s been that’s been sort of the bulk of the speaking stuff is, you know, hour long workshop, hour long talks are six hour workshops internally at companies to say like, hey, let’s get your team up to speed. Let’s show you this stuff. But in a big version of a talk that’s customized for your industry. So you can see how you would apply this today, like your telecom company, client, for example, yeah, you would show a ton of examples. Like here’s how you would use voice the customer data from your call center to guide your marketing strategy, like how you would create marketing copy from the voice of the customer, because it resonates better when customers see the language that they would be using themselves rather than what a marketer came up with, which may or may not even be true.

Evan: Wonderful advice. Thank you so much for spending time with us the and thank you the audience here for watching. Reach out to Chris @cspenn on Twitter and beyond. Thanks so much.

Christopher: Thank you for having me.

Evan: Take care.

You might also enjoy:

Want to read more like this from Christopher Penn? Get updates here:

subscribe to my newsletter here

AI for Marketers Book
Take my Generative AI for Marketers course!

Analytics for Marketers Discussion Group
Join my Analytics for Marketers Slack Group!


Leave a Reply

Your email address will not be published. Required fields are marked *

Pin It on Pinterest

Share This