If you’ve been online at all the last week, you probably have heard enormous chatter about ChatGPT, the chatbot that might kill Google. It’s a really powerful tool. And the work its creators have done is incredibly impressive. But is it really a Google killer? And what should your business do about it right now?
This episode of Thinks Out Loud looks at ChatGPT and where it might be headed. We also take a tiny dive into GPT-3, the large language model ChatGPT is based on, to understand what it can do — and what it can’t. We ask what Google is doing with large language models of its own. And we explore how you might use tools like this now, and in the future, to benefit your business.
Want to learn more? Here are the show notes for you.
Will ChatGPT Kill Google? — Headlines and Show Notes
Show Notes and Links
- OpenAI – Wikipedia
- GPT-3 – Wikipedia
- DALL-E – Wikipedia
- OpenAI Impact Analysis: Microsoft, Google And Nvidia (NASDAQ:MSFT) | Seeking Alpha
- ‘Google is done’: World’s most powerful AI chatbot offers human-like alternative to search engines
- Google AI
- Google Faces a Serious Threat From ChatGPT – The Washington Post
- The limitations of scaling up AI language models | VentureBeat
- The Power and the Pitfalls of Large Language Models: A fireside chat with Ricardo Baeza-Yates. – Information Matters
- Large language models have a reasoning problem – TechTalks
- LaMDA: our breakthrough conversation technology
- Generative AI: autocomplete for everything
- AI Won’t Steal Your Job: Smart People Who Put AI to Work Will (Thinks Out Loud Episode 208)
- The Future Digital Trends You Must Think About Today (Thinks Out Loud Episode 342)
- Why Amazon Is Going to Launch a Search Engine (Thinks Out Loud Episode 345)
- Revisiting Day One at Amazon and Developing a Digital Culture (Thinks Out Loud) – Tim Peter & Associates
- Revisiting "Are You Destined to Lose to the Frightful Five?"
- Why Digital Transformation is More Than Just E-commerce (Thinks Out Loud 331) – Tim Peter & Associates
- 2022 Holiday Shopping Trends & Insights Report | Adobe for Business
- The Future is Already Here (Thinks Out Loud Episode 370)
- The Most Important Trends in Marketing 2023 (Thinks Out Loud Episode 369)
- The End of Google? (Thinks Out Loud Episode 372)
- What We Learned from Big Tech’s Earnings Q1 2023 (Thinks Out Loud Episode 373)
- Big Trends: The Early Innings of AI in Marketing (Thinks Out Loud Episode 374)
- Big Trends: Does Marketing Have a Future? (Thinks Out Loud Episode 375)
- How to Escape Big Tech’s Web (Thinks Out Loud Episode 395)
We have some free downloads for you to help you navigate the current situation, which you can find right here:
- A Modern Content Marketing Checklist. Want to ensure that each piece of content works for your business? Download our latest checklist to help put your content marketing to work for you.
- A Brief Introduction to Thinks Out Loud. As a bonus, we’ve also included this PDF document that highlights some of our core episodes to help you dig into what the show is about. We think it will help you capture the show’s essence while you’re working your way through the 300-plus episodes published so far. Download it here.
- Digital & E-commerce Maturity Matrix. As a bonus, here’s a PDF that can help you assess your company’s digital maturity. You can use this to better understand where your company excels and where its opportunities lie. And, of course, we’re here to help if you need it. The Digital & E-commerce Maturity Matrix rates your company’s effectiveness — Ad Hoc, Aware, Striving, Driving — in 6 key areas in digital today, including:
- Customer Focus
Subscribe to Thinks Out Loud
Contact information for the podcast: firstname.lastname@example.org
Past Insights from Tim Peter Thinks
Technical Details for Thinks Out Loud
Recorded using a Heil PR-40 Dynamic Studio Recording Mic and a Focusrite Scarlett 4i4 (3rd Gen) USB Audio Interface into Logic Pro X for the Mac.
Running time: 25m 27s
You can subscribe to Thinks Out Loud in iTunes, the Google Play Store, via our dedicated podcast RSS feed (or sign up for our free newsletter). You can also download/listen to the podcast here on Thinks using the player at the top of this page.
Transcript: Will ChatGPT Kill Google? (Thinks Out Loud Episode 367)
Well, hello again, everyone, and welcome back to Thinks Out Loud, your source for all the digital expertise your business needs. My name is Tim Peter. This is episode 367 of the Big Show, and thank you so much for tuning in. I very, very much appreciate it. It’s always so good that you’re here. I’ve said this many times on the show, but I wouldn’t do this show without you, so thanks so much for tuning in.
Now this episode is not the episode I thought it was going to be. I intended to talk today about the holiday shopping season and what it tells us about the state of digital, what lessons we can learn from how and where people are spending their money online. But something got in the way, something that I simply can’t ignore right at the moment.
And that something is a tool that has gone just bonkers crazy over the last few days, all around the internet. Something called ChatGPT. You may have heard of it, maybe you haven’t. But ChatGPT is a chatbot built on the GPT-3 large language model from OpenAI. And it, along with other tools that have come along prior to this — like DALL-E are part of a field known as
And generative AI has enormous implications in business and in how we think about making our companies more digital. How we talk to our customers. How we connect with customers. So if you’re not familiar with generative AI, it’s
“…the field of AI that focuses on generating new data or content. Generative AI technologies such as generative adversarial networks (known as GANs) or Variational Auto Encoders (VAE) have made significant progress in recent years and are able to generate highly realistic images, texts, and other forms of data. However, there is still much work to be done in terms of improving the quality and diversity of the generated content, as well as making the generation process more efficient and scalable. In case you’re curious, generative AI can potentially be used to write marketing copy. In fact, there are already AI tools available that use natural language processing, NLP and generative algorithms to produce marketing copies such as product descriptions, headlines, and slogans.
These tools can help save time and effort by quickly generating multiple options for marketers to choose from. However, it is important to note that the quality of the generated content may vary, and it may require human review and editing to ensure that it is clear, compelling, and accurate.
Additionally, it’s important to consider ethical and legal issues related to the use of AI generated content in marketing, such as issues of transparency and accountability.”
If you think that generative AI isn’t that big a deal or isn’t going to be that big a deal, you might want to know that the last couple paragraphs I just read to you, starting with “…the field of AI that focuses on generating new data” through “…issues of transparency and accountability” were generated by ChatGPT.
I asked it a couple of questions about generative AI and how it can be used for creating marketing copy, and that’s what it gave back to me. That’s crazy. I mean, that was pretty solid information right there. It’s easy to read. It’s authoritative. It’s fairly comprehensive. Yes, the language is a bit stiff, and it used a fairly similar format for each answer — note the “howevers” in each paragraph. But it’s remarkable how far it’s come already. And the fact that we’re just a few months away from OpenAI introducing GPT-4, which is the next iteration, the next generation large language model that they will use behind tools like ChatGPT. That’s just crazy.
So, yes, generative AI does need to be more “…efficient and scalable,” and it still requires plenty of thought around “…transparency and accountability”. But we’re clearly, clearly in an age where these tools are going to play a role: demand your attention in terms of where they fit in your overall marketing, where they fit in your overall business, and where they fit in the marketplace.
Will ChatGPT Disrupt Google?
I have seen a number of articles over the last few days — and I will link to them in the show notes as I do — that have talked about things like ChatGPT as a potential replacement for Google. Not least of which is because it was funded, at least in part, by another member of the AGFAM, another Big Tech member, somebody who’d love to kick Google in the teeth: Microsoft.
What we also have to think about is that what it’s doing is very different than what Google does. It’s not finding the right answers that exist on the internet; it’s creating them on the fly. It is generating them from the data that it has consumed.
Google Has Its Own Large Language Models
I’m going to be really clear. I’m going to put a stake in the ground. I do think this is a risk to Google, but I don’t think it’s existential yet for a bunch of reasons. One, I think Google is very conscious of the risks here. They’re probably not overly worried. Google itself has a large language model known as LaMDA that’s very impressive in its own right. Google has been using machine learning and natural language processing as part of what they do for some time.
I mentioned in an episode a few weeks back that they have talked about how their ads business is the largest use case for machine learning within Google. They’ve been using these kinds of tools for a while.
Google Has a Clearer Path to Monetize Its Traffic
More to the point. One of the reasons I think they’re probably not too afraid is that they already have a way to monetize their answers.
We absolutely could see ads like, you know, if somebody put an ad alongside the text and I read you a minute ago, you know, “Want to learn more about Generative AI? Click here to see how Microsoft can help you apply these solutions to your business.” That would make a ton of sense. I absolutely could see that happening. But if the customer already got the answer they needed, the clickthrough rate on ads like that probably is going to be on the lower end.
At least that’s what I think. I plan to keep a close eye on this and I plan to update my point of view as I learn more. I don’t claim to be the expert on this yet, and I’m going to come back to that in a minute.
The Limitations of Large Language Models
A separate issue here is that large language models like GPT-3, like LaMDA, suffer from a number of limitations.
Training Data is Expensive
One of them is that they can run out of the text that has the right answers to learn from. Getting quality training data to train these large language models is very expensive in computing terms and in financial terms.
Training Data Can Make the AI Worse
Even more troubling is that their results can degrade as they ingest more data from less reputable sources or from other AI generated sources. You reach a point where you’re not training an artificial intelligence, you’re training an artificial stupidity.
It keeps learning from worse and worse data as you expand the net — no pun intended in this specific case. One of the things that GPT-3 does, one of the things that GPT-3 does is it does not pull data directly from the internet. It mostly comes from books and things like that. And it’s only up to date through 2021. Because when you turn it loose on the internet, the quality of data it pulls in gets, on average, worse.
Large Language Models Are Often Wrong
And that leads to problems where as we’re seeing currently some non-trivial percentage of the answers that ChatGPT provides are flat out wrong.
Estimates range somewhere between the low single digits up to maybe 10%. But that’s a tough problem to engineer away. Where can the people who create these tools consistently get training data that contains the right answers? Even worse, how do the engineers know which answers are correct, especially at scale? For that matter, given that AI is itself a black box, we don’t know how it gets to the answer it gets. How do the engineers know it will consistently produce the same answer to questions that are worded slightly differently?
I’ve seen this myself in a couple of tests that I’ve run with ChatGPT, where you ask the question one way, you get one answer, you ask it a different way, you get a slightly different answer — and sometimes contradict contradictory answers.
It’s something that they’re working through. But it’s difficult to simply deal with from an engineering perspective. Not impossible; but difficult.
Large Language Models Like GPT-3 Are Subject to Bias
The other challenge that you have when you open this up to data from anywhere, is that there’s a known problem that these models tend to exhibit biases around race, religion, gender, particularly if you turn them loose on the internet at large.
As we all know, there’s some really nasty stuff on the internet. There’s some pure garbage out there online. Large language models are very much a “garbage in/garbage out” kind of situation.
By the way, I didn’t say that. The folks at OpenAI who built the GPT-3 language model did in a recent paper. They wrote a paper and made that freely available to talk about the bunch of limitations here. And that doesn’t lead to a system that brands are going to be super comfortable putting their ads on.
So they have some ways to go before they’re an immediate threat to Google.
The Upside of ChatGPT
They are also very, very cool. And there are lots of practical applications to come from their use. I used ChatGPT earlier this week to write a difficult email that I’d been struggling with. I had to give some bad news to somebody and I had ChatGPT write the first draft. I edited it. I cleaned it up. I put it more in my voice. But it gave me something to work from that was really quite good.
And so we might see tools like ChatGPT automating away specific tasks that are maybe a little more difficult, a little more challenging, or not worth huge amounts of energy mentally for people to do.
The Risk to Google
I also love that we’re talking about ways that Google could be challenged by it. I have talked many times before on the show about why Google’s strength is brittle and how customers can shift in a minute to something else. There’s no lock-in to Google for most searchers. And that’s where they make all their money, right?
If I wanted to switch from an iPhone to an Android phone, that has a cost associated with it for me. I have to buy a new phone. I have to learn how to use the new OS. I have to pay for some apps that I had before. So it’s not free for me to switch. I have to learn how to use it if it’s a little different.
If I want to switch from my Mac to a Windows PC, same thing. There’s a cost. If I want to switch from the car I drive to a different car, there’s a cost. My insurance rates may change. There’s all kinds of things associated with it that lock me into the existing platform.
But, if I want to use Bing to conduct a search or Ask to conduct a search or ChatGPT, all I have to do is type a different URL. The switching cost to me is zero. And I can’t think of another company in history that has been as large or as powerful or as dominant in the marketplace as Google, where all of its customers and all of its revenue theoretically could go away tomorrow. Instantly.
People could just switch without doing anything all that complicated or all that expensive.
How Can Google Compete?
Better User Experience? Increase Lock-In?
Google has at least a couple of different ways it could respond. One is to create better experiences, better results. To learn from this, to say, “Maybe we need fewer ads. We need to create an experience that people love,” to continue to attract use among their customers if there’s a potentially better alternative out there.
The other of course, is to try to find ways to lock consumers in, make it harder for them to switch.
I would hope it’s the former. I’m also not an idiot. The second is a very real possibility. And a hybrid somewhere between those two is probably the most likely thing that we’ll see.
But I do think that it’s something that they’re probably sitting up over the last week and saying, “wow, we need to pay attention to this.” I suspect they have been all along. I doubt OpenAI — which has been open to artificial intelligence from researchers for several years — doesn’t have members of Google’s AI teams working with it.
What Can You Do?
Now, I don’t like to talk about these things solely in the abstract. I like to talk about them in terms of what you can do. And I’m going to come back to two things I’ve talked about in the past, but I think they’re really important.
Core and Explore
One, remember the principle of core and explore. ChatGPT is very cool. It’s “ooh, shiny.” Don’t get distracted by the “ooh, shiny.” You still want to spend most of your time, most of your resources, most of your budget on the things that work today.
Core. The Washington Post, in talking about the threat that ChatGPT presents to Google, noted that it had reached 1 million users in five days. For comparison, it took Instagram two and a half months to reach that same level of success. That’s extraordinary. But a million users in internet terms is… not that many. Google gets almost 4 million search queries every minute.
Online shopping revenues, the thing that I thought this episode was going to be about when I sat down to write this script the other day, are around 210 billion so far this year. So you know, ChatGPT is cool, but it’s not huge. We want to spend most of our time focusing on the core.
Explore. At the same time, if you focus only on where we are right now, that’s probably a mistake. Thatis a mistake. You know the famous line from hockey legend Wayne Gretzky, about skating to where the puck is going, not where it is? That 210 billion in online shopping revenue I just mentioned is the slowest growth we’ve seen in years. It’s clear that people also enjoy in-person experiences, both in terms of how they find the things they want to buy, and in terms of the kinds of things they want to spend their money on in the first place.
After three years of a pandemic, people don’t want to spend as much time in front of a screen. They want to go places. They want to see the world. They want to be with friends and family. They’re ready to explore.
That’s where the Explorer side of “Core and Explore” comes in. You have to allocate a small amount of your budget — 10%, 20% — to explore new possibilities, new products, new channels, new technologies. Who is using large language models to help you be more effective at what you do?
You want to test, you want to learn, you want to grow. And then put more of your resources to work in the areas that you find produce results. So it’s core; put most of your energy there. And also explore; save some energy, save some resources, save some budget to test and learn because that’s how you’re going to be ready regardless of what happens with GPT-3 — or anything else that comes down the path.
The second thing you want to do, and it’s highly related, is you want to keep learning. I don’t claim to be an expert on generative AI or large language models specifically, I have experience with natural language processing (NLP) and its use in marketing. I know a little bit about artificial intelligence more broadly. But I’m still very much learning about how we can apply these tools in useful ways.
Your job is to do the same. You’ve heard me say before that AI won’t take your job, but smart people who use AI will. Economist Noah Smith and the researcher roon had a great piece the other day about the effect of AI on people’s jobs. They called it “autocorrect for everything.” They said that “AI doesn’t take over jobs, it takes over tasks.” And my favorite line in their piece stated, “Dystopia is when robots take half your jobs. Utopia is when robots take half your job.” Notice the singular and plural difference.
I think the latter is going to be much more common. And the people who do that, who put that into practice are who you really should worry about.
If you’ve listened to this show before, you’ve heard me tell the joke about two guys walking through the woods when a bear comes charging towards them. The first guy immediately pulls out a pair of running shoes. And the second guy says, “What are you doing? You’ll never outrun that bear.” The first guy says, “I don’t have to outrun the bear. I just have to outrun you.”
If you are not learning, if you’re not growing, if you’re not reading about these things and playing with it and creating a ChatGPT account for yourself, and getting some experience with it, you are leaving your running shoes in your backpack. You’re the second guy in the joke. And it won’t be funny when the bear comes for you, your job, your business, or your industry.
One of the reasons I’m not bearish about Google is that I don’t think they’re likely to act like the second guy. They may trip, they might fall. But they’re running. They’re learning. And that’s the lesson that you should learn. We have to keep learning. We have to use core and explore. We have to keep finding out about how these tools can be used in our business because we know the machines are learning fast.
But humans who learn fast, people who learn fast are the ones who are going to win. This may not have been the episode I originally planned for, but I certainly intend that you are one of those people who win. You can do this. Keep learning. Use core and explore. Keep those top of mind and you’re going to be fine.
Show Closing and Credits
Now, looking at the clock on the wall, we are out of time for this. I want to remind you that you can find the show notes for today’s episode, as well as an archive of all past episodes by going to Tim peter.com/podcast. Again, that’s Tim peter.com/podcast. Just look for episode 367.
Subscribe to Thinks Out Loud
Don’t forget that you can click on the subscribe link in any of the episodes you find there to have
Think Out Loud delivered to your favorite podcatcher every single. You can also find Think Out Loud on Apple Podcasts, Google Podcasts, Stitcher Radio, Spotify, Overcast, anywhere fine podcasts are found.
Leave a Review for Thinks Out Loud
While you’re there, I would also very much appreciate it if you could provide a positive rating or review for the show. Those help new listeners find the podcast. It helps them understand what we’re all about. It helps get the word out about Thinks Out Loud. And it means so very much to me that you help build our community. I very, very much appreciate it.
Thinks Out Loud on Social Media
You can also find Thinks Out Loud on LinkedIn by going to linkedin.com/TimPeterAssociates. You can find me on Twitter using the handle @TCPeter. And as always, you can email me at email@example.com. Again, that’s firstname.lastname@example.org.
With all that said, I just want to say once again how much it means to me to have you tune into our little show here every single week. I do hope you have a great rest of the week.
I hope you have a wonderful weekend ahead. And I’ll look forward to speaking with here on Thnks Out Loud next time. Until then, please be well, be safe, and as always, take care everybody.