In this episode of The Artificial Intelligence Show, Paul and Mike explore the latest in AI, starting with NVIDIA's impressive earnings, and the viral buzz around AI startup Groq. Our rapid fire section is packed with everything from the ethical considerations of AI companies licensing data, to AI mistakes in customer service as seen with Air Canada, and the critical efforts to combat AI deepfakes to protect elections.
Listen or watch below—and see below for show notes and the transcript.
Listen Now
Watch the Video
Timestamps
00:03:20 — NVIDIA’s blockbuster earnings report
00:21:18 — AI startup, Groq, goes viral
00:31:20 — The true impact AI is going to have on your work
00:42:35 — AI companies are licensing data from Reddit
00:45:42 — AI Agents in the wild
00:49:23 — Google Gemini Image Generation Issues
00:57:30 — Air Canada forced to give refund to passenger due to AI chatbot mistake
01:00:43 — Efforts to Fight AI Deepfake and Protect Elections
01:02:40 — Google Introduces Gemini Business
Summary
NVIDIA Blockbuster Earnings Report
NVIDIA, the dominant leader in AI chips with an estimated 80% market share, just had another blockbuster earnings report…
The company’s stock jumped 16% last Thursday when the earnings were released—and its market cap now sits at almost $2 trillion. In 2024 alone, the stock has risen 63%!
This is being powered by the need for GPUs (graphics processing units), the chips that power AI applications.
As every company becomes an AI company, NVIDIA expects demand for those chips to skyrocket through the coming years.
AI startup goes viral as it guns for Nvidia’s business
The startup Groq—G-R-O-Q, not G-R-O-K, the name of Elon Musk’s AI company—is an AI tool that serves responses to chat queries insanely fast, thanks to customized chips and a novel software design.
The company started going viral after users compared its speed in benchmark tests and Groq crushed OpenAI’s flagship product. (with some estimates saying it’s 13 times faster than ChatGPT.)
The system is built with custom-made AI chips called Language Processing Units (LPUs) that allow the company’s tool to serve results almost instantaneously.
Groq describes LPUs as “a new type of end-to-end processing unit system that provides the fastest inference for computationally intensive applications with a sequential component to them, such as AI large applications (LLMs).”
This is a big deal. The ability to serve up answers almost instantaneously from LLM-based tools could open up a whole range of possible AI use cases in businesses.
Groq’s system could also threaten Nvidia’s business if companies start demanding LPUs instead of GPUs thanks to the speed gains. (Groq also says it can deliver performance for way cheaper than Nvidia.)
The True Impact of AI on Knowledge Work
What is the true impact AI is going to have on your work? Thanks to AI advancements in the past few weeks, it’s a question that business leaders need to start answering—fast.
With Google’s release of Gemini 1.5, we now have a model that can handle 1 million tokens in its memory, or about 750,000 words. Not to mention, Google is also working on systems that will reach 10 million tokens, or 17,000 pages.
These systems are also getting much, much faster, as we saw with developments like Groq.
“Speed and memory are both vital to making AIs more usable and powerful in the real world. Imagine feeding AI hundreds of pages of instructions on how to do something, and then having it quickly do exactly that,” wrote AI expert Ethan Mollick this past week.
In one experiment, he fed Gemini 1,000s of pages of his work and the tool summarized, understood, and quoted from the work in less than a minute.
He said: “...the advent of massive context windows gives AI superhuman recall and new use cases” and noted this would have taken a team of human researchers days to do on their own.
As such, these types of emerging AI capabilities are going to profoundly change knowledge work, warns Mollick.
Today’s episode is brought to you by Marketing AI Institute’s AI for Writers Summit presented by Jasper, happening virtually on Wednesday, March 6 from 12pm - 5pm Eastern Time. To register, go to AIwritersummit.com
Links Referenced in the Show
- NVIDIA Earnings Call
- Groq AI Chip System goes viral
- AI Impact on Work
- Reddit is going public
- AI Agents
- Google Gemini Image Generation Issues
- Air Canada must honor refund policy invented by airline’s chatbot
- Efforts to Fight AI Deepfake and Protect Elections
- Google Gemini Business
Read the Transcription
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: Infrastructure is critical to everything. They have to build more data centers the supply chain that enables the building of those data centers and the management of those data centers is going to be critical.
[00:00:11] Welcome to the Artificial Intelligence Show, the podcast that helps your business grow smarter by making AI approachable and actionable. My name is Paul Roetzer. I'm the founder and CEO of Marketing AI Institute, and I'm your host. Each week, I'm joined by my co host, and Marketing AI Institute Chief Content Officer, Mike Kaput, as we break down all the AI news that matters and give you insights and perspectives that you can use to advance your company and your career.
[00:00:41] Join us as we accelerate AI literacy for all.
[00:00:48] Paul Roetzer: Welcome to episode 85 of the Artificial Intelligence Show. I'm your host, Paul Roetzer, along with my co host, Mike Kaput. Good morning, Mike.
[00:00:56] Mike Kaput: Good morning, Paul.
[00:00:58] Paul Roetzer: We have, [00:01:00] like, I you and saying before we get on this, it was like, not a crazy week, not like last week, where just like, Everything was happening at once as we preparing for the show.
[00:01:09] Paul Roetzer: It was like, wow, there's some like really sneaky, big topics this week that have some really big implications. So we've got some good stuff today. I'm actually like, I was going through kind of looking through the prepping for this one. These some topics I'm, I'm really excited to talk about and maybe.
[00:01:25] Paul Roetzer: I think there might be some, I'm trying to be real like cautious about too many of my own opinions and personal beliefs into things today, but there's definitely a few where we have to kind of get into some perspective on few big issues. So we're going to get into that. today's episode brought us by the Marketing Institute's AI for Writers Summit, which is presented by Jasper.
[00:01:49] Paul Roetzer: That is next week. It is March 6th, Wednesday, March 6th. It is a half day. virtual event. So it's to be from noon to 5 p. m. Eastern time, following [00:02:00] tremendous success of our inaugural AI for Writers Summit in March 2023, which drew more than 4, 000 writers, editors, and content marketers. excited to present the second edition the event featuring expanded topics and even more valuable insights.
[00:02:15] Paul Roetzer: You can check out the agenda, there's a free ticket option. So can check all that out at AI. Writer summit.com. ai writer summit.com. And as big thank you to our presenting sponsor, Jasper, we wanted to take a minute and give you a little more information about Jasper and what their platform can do. So Jasper is an AI co-pilot for enterprises that uses AI to generate on-brand content that reflects your brand style and voice. Jasper do everything from write blog posts for you, to repurpose or rewrite content. it sounds like you because it can securely store information about company and products that also cuts down on hallucinations, where the large language model just makes its stuff [00:03:00] Jasper can even make suggestions on how to improve content performance based on trends in your content data. If you're looking to augment your marketing and content efforts with AI, Jasper is well worth exploring, especially if you're in an enterprise. To learn more, go to jasper. ai. Alright, let's do it, Mike.
[00:03:20] NVIDIA’s blockbuster earnings report
[00:03:20] Mike Kaput: Main topics. Alright, alright,
[00:03:22] Mike Kaput: Paul. First up, NVIDIA, the dominant leader in AI chips. They have an estimated 80 percent market share. They just had another Blockbuster Earnings Report. Their company's stock jumped 16 percent last Thursday when earnings were released. And NVIDIA's market cap is now right around 2 trillion dollars.
[00:03:45] Mike Kaput: 2024 alone, the stock has risen 63%. Now, this is being powered by the need for GPUs, or Graphics Processing Units, which are the chips that power AI applications. [00:04:00] As every company kind of becomes an AI company, NVIDIA expects demand for these chips to skyrocket over the coming years. So first up, Paul, I wanted to get your thoughts this crazy successful earnings call and kind of what this might mean for knowledge workers and businesses who presumably can expect a huge boom in AI applications and power moving forward.
[00:04:28] Paul Roetzer: So I, you know, I had a couple thoughts. I was watching closely. I started investing in Nvidia myself, probably somewhere around 2014 or 15 when it was around 60 some dollars a share. and my personal feeling back then. So when we really started exploring AI deeply at a personal level, was that Wall Street just completely was missing the boat on AI and the companies that were positioned for once. Corporations started actually infusing AI and we started really [00:05:00] scaling up the use of AI that they were just completely missing that Microsoft, Amazon, NVIDIA, Tesla to a degree, that they were all AI companies and they were being valued based on what they currently did, not what they were going to do and what they were going to mean to society.
[00:05:17] Paul Roetzer: So, I've been huge investor in NVIDIA for long time, and so I watching kind of nervous energy leading to the earnings call last week because when you go and look, I have probably, I don't know, maybe six to 10 analysts, financial analysts that I follow pretty closely, mostly on Twitter, that I use as kind of a barometer for what I think might happen in the financial world.
[00:05:44] Paul Roetzer: And they split. So there was some analysts were like, yeah, there's no way they're going to hit these insane markers that they're, that, that wall street has put on them in terms of revenue and earnings, and, and growth. And then the other one's like, no, they're, they're [00:06:00] going to do They blow it away every, every quarter, they're going to do it again.
[00:06:02] Paul Roetzer: And so you realize like these people have no like they're, they have more access to information than obviously we do. And it's just educated guesswork. So there was some analysts who very confidently thought that, you know, you should take some, some money off the table, sell some of that Nvidia stock before the earnings, and then that night after the earnings, like it was up, I think 16 percent in after hours trading, like something crazy like that. So just give some context to, to Nvidia stock, if you to put 10, 000 into Nvidia back in 2014, that 10, 000 would be worth approximately 1. 7 million today. I unfortunately didn't have 10, invest back in 2014, so I didn't personally turn 10, 000 into 1. 7 million, but started investing and I kept investing over time. NVIDIA, after this earnings breakthrough, is now the fourth largest company in the world, surpassing Google and Amazon, who are six and [00:07:00] five now. And, and NVIDIA up at number So thing that. There was of things that came out of this that I are really relevant to all. Anyone who listens to this podcast, either because of your personal investing and you're trying to kind of out how to benefit from AI and which stocks are worth investing in, to what's actually happening in AI and, and what you can learn from the NVIDIA call and transcript you, you know, you read or listen to the transcript from Jensen.
[00:07:29] Paul Roetzer: is how they view the future. And the big thing to date is their growth has, has in many ways been driven by the need for their GPUs to train these models. So when we talk about inflection and Anthropic and OpenAI and Amazon and Google, they're all. Training these large language models, which requires a massive amount of compute, meta another one we've talked about.
[00:07:56] Paul Roetzer: So they need these GPUs to train. But the [00:08:00] real growth in future is what's called inference. It's once the training is done, and you and I Use the model. We, you know, we query ChatGPT or Anthropic or Gemini or whatever. That moment when it now goes and computes information to output something to us is called inference.
[00:08:17] Paul Roetzer: So what they're looking at is saying, Hey, in enterprises, we haven't even really hit the adoption curve yet. We've talked about this many times on show. All this growth in these tech stocks have happened without. So once enterprises actually start using these generative AI tools into their daily workflows, there's whole new level of growth that happens and a whole new level of need for these GPUs, for these processing units.
[00:08:46] Paul Roetzer: And so. The way Jensen has talked about it before is in these like four waves of AI. and so Gene Munster, who's one of the people follow, one of of the financial analysts, he's the managing partner Deepwater Asset [00:09:00] Management. So he's really active on Twitter. He shares great insights into these stocks and stuff.
[00:09:04] Paul Roetzer: And so he had tweeted last week about how Jensen talks about these sort of four waves. of AI. The first is sort of this training, this infrastructure build out where we're training these models. then you start to get into the inference level in like the second wave. And so he basically said we're at like 25 percent into the first wave.
[00:09:26] Paul Roetzer: The second wave where enterprises really start using these generative applications. And that will specifically take off with AI agents. Now talk more about AI agents. In a topic later on today, we've talked plenty about AI agents in recent episodes, what's these agents start getting integrated into the software we use?
[00:09:45] Paul Roetzer: Now we're talking about billions of agents being built and that like the average marketer, the average business person may be using dozens of different AI agents every day to assist in tasks. That is a whole nother level of computing power. And that's where inference [00:10:00] happens. That's where we're constantly calling on these models to take actions and help us deliver outputs.
[00:10:06] Paul Roetzer: So that's the second wave. We haven't hit that We're at the very, very beginning that wave, possibly. The third wave. industry. This is where like manufacturing and other industries really start figure out how to use it. And the fourth wave is sovereign AI, which is basically every government in the world infusing generative AI into every application.
[00:10:25] Paul Roetzer: So again, when we think about where this goes, we are going, we're, we're just entering growth phase of this technology. The AI is not widely adopted. We're in this really like really within this first wave, maybe starting to play into the second wave of some enterprise adoption, but early. And when Jensen, know, we talked about Sam potentially raising like Sam Altman, five to 7 trillion to build infrastructure, what they're all looking at, whether Google, OpenAI, Amazon.
[00:10:53] Paul Roetzer: They're basically looking out five 10 years from now into a world where these agents are everywhere. that [00:11:00] inference happening all the time for every task that's happening within knowledge work. And they're saying, we don't have enough data centers for this. And so there's a massive movement to build more capacity, to build more data centers, to build more chips. And so infrastructure is critical. So again, When we go back to like, what should you be investing in personally? I'm not giving stock advice. I'm not telling you to invest in video person.
[00:11:23] Paul Roetzer: I'm just setting the stage for where this is all going go. Infrastructure is critical to everything. They have to build more data centers the supply chain that enables the building of those data centers and the management of those data centers is going to be critical.
[00:11:38] Paul Roetzer: So Jensen said, he thinks. There needs be 1 trillion in infrastructure build out in the next five years, which is double current pace. So he doesn't buy in. was asked, actually asked about Sam's 5 trillion and kind of laughed. He's like, yeah, I don't know that we need that much because he feels like the AI is going accelerate.[00:12:00]
[00:12:00] Paul Roetzer: The efficiency of building and doing inference, like, you know, building these models and doing So we don't necessarily need as much that, but a trillion is a lot. And so that's kind of where we're looking. And when I look out the future, I'm thinking personally about investing, or thinking about like what's going to happen in business.
[00:12:17] Paul Roetzer: You have understand that they all believe AI agents are going to be key. Those agents are going to need a lot of computing power, which is going need a lot more chips and a lot more data centers. that's basically the next five to 10 years of business.
[00:12:32] Mike Kaput: Wow. There's some serious implications to what we're talking about here.
[00:12:37] Mike Kaput: And one of them is kind of what you highlighted about AI agents and Paul, you had shared some post earnings call thoughts on LinkedIn that were related to the impact of all this, of agents, of this infrastructure build out, of where we're going on agencies, value added resellers, VARs, [00:13:00] and knowledge workers, so can you share What caught your attention in some of the comments that Jensen, the CEO, had made in a post earnings call and kind of what we be thinking about prepare for all these coming changes.
[00:13:16] Paul Roetzer: Yeah, there was a tweet in particular from, it was John Fort, who's CNBC Overtime. And it's about 14 minute clip overall, but there was about 50 second clip that caught my attention. And so in this excerpt, Jensen says, the world's enterprise software platforms represent about a trillion dollars. These application oriented tools, and, or application oriented, tools oriented platforms and data oriented platforms all going be revolutionized with these AI agents.
[00:13:47] Paul Roetzer: that sit top of it. And the way to think about this is very simple. Whereas these platforms used to be tools that experts would learn to use, in future, these tools companies will also offer [00:14:00] AI agents that you can hire to help you use these tools or to help you reduce the barrier of using these tools.
[00:14:07] Paul Roetzer: So, Again, if you're a new listener to the show don't know my background and Mike's background, I owned and ran a marketing agency for 16 years. We were HubSpot's first value added reseller partner in 2007. So I built my agency on the back HubSpot and providing. Services to HubSpot customers that didn't want to use the platform themselves didn't have the internal staffing to use the platform.
[00:14:31] Paul Roetzer: So we just became experts in HubSpot and we delivered services to companies that wanted to grow using their CRM system and our marketing and sales and customer service tools. So I looked at this quote and Mike worked with me there, starting back about 2011, Mike, I think. So we spent like 10 years that agency together.
[00:14:49] Paul Roetzer: So, as soon as I heard this, I was like, wow, like that's going to transform ecosystems. And we've known this, like we've talked about this on the show, to hear Jensen say was like a whole nother level [00:15:00] of like, okay, like this is actually happening now. They believe this to be true. And so I immediately started thinking about.
[00:15:07] Paul Roetzer: What's going happen to these agencies and VARs they change fast? Like, they need to really be thinking about what is the impact having AI agents that can do stuff almost no cost. I mean, like, realistically, the cost versus what would cost to traditionally hire humans do work is going to be next to zero.
[00:15:26] Paul Roetzer: And so all of the sudden I can go into HubSpot or Salesforce or Oracle or Intuit or whoever has built these VAR networks agency ecosystems. And I can hire an agency, or I can just turn on an AI agent, whether it's built natively within one of these platforms or a third party tool like multi on or adapt or, you know, hyper, right?
[00:15:47] Paul Roetzer: Which we'll aboutwhere else has built an agent basically just sits on your browser and can do task work for anything you do on your browser, which would include.
[00:15:59] Paul Roetzer: And so to me, the [00:16:00] implications are massive and, and I'm saying this specifically for agencies ours, but step back and no, this is like all knowledge workers, like anyone who works in these platforms, even on brand side, you're using tools that an agent is going to be trained use.
[00:16:14] Paul Roetzer: And the way it will happen is you will literally just turn the agent on. And it'll learn how do what you do by watching you once. And so that's what is going to be enabled is like, it's, so let's say I go into HubSpot and I'm going to do my 21 clicks to send an email.
[00:16:30] Paul Roetzer: learn from watching me click where click, and then it will learn how to do that.
[00:16:35] Paul Roetzer: And the next time I need to do that task. It will have learned and I will just click a button and it'll go do that task. it's like almost imagine going into your platform and say, I want to run a new report. And I turn on the agent and say, okay, learn, maybe it's just a learn button. And it now watches everything I do on my browser.
[00:16:54] Paul Roetzer: And it learns that task by watching. And then I finish training or whatever, and it's now done. [00:17:00] It now knows how to do task moving forward. So you're going to be able to do that. So if an agency a VAR, one thing you may do is offer services to train these models. Like that's part of what your job might become, one of the services you might offer.
[00:17:13] Paul Roetzer: So the, in the post, and we'll of put the same, what I would is like, follow me on LinkedIn is a great way to kind of see these kinds of things. And put the link, the post, link in there. you can go check out. But basically what I did is took our human to machine scale of like levels of automation of knowledge work, and I shared that, and then I shared.
[00:17:34] Paul Roetzer: A bunch of features within HubSpot platform. And I said, one way to kind of future proof your company or your career is go through the features of these platforms that you spent all this time in and try and take some educated guesses about which things the AI agents will do and which things will be left for the humans to do.
[00:17:52] Paul Roetzer: And you, if you're confident agents are going to do the vast majority of the work that those features enable. Don't [00:18:00] be building services around that anymore. And like really start moving to the features and capabilities these platforms enable. That going to be very minimal AI assistance. so I just, I think it's like really important thing that people start to truly look at the impact AI is going to have.
[00:18:17] Paul Roetzer: So we, what we call AI impact assessment, that you start to look out think about your own job description, the platforms that you use, the campaigns you run, workflows run, really start consider 12 months from now. What does this start look like? Because that the only way to get out ahead of this stuff.
[00:18:38] Mike Kaput: Wow, so it sounds like,
[00:18:41] Mike Kaput: If I'm hearing this right, so if I'm a business would historically hire an agency or a professional to do something, and I now can start thinking about, well, can an agent overlaid HubSpot, laying over HubSpot, simply help me do this instead? I mean, that's a [00:19:00] simple, but extremely disruptive idea.
[00:19:03] Paul Roetzer: Very much. And the way I think about it is, I don't realistically think changes much in, like, 2024. I think more people will start to experiment with these agents, but as we've on the podcast before,
[00:19:17] Paul Roetzer: You're to have to give up a pretty significant level of privacy. and you're going to to have a lot trust in companies that you allow to see everything do on your browser.
[00:19:30] Paul Roetzer: So I don't think adoption, one, I don't think the tech is quite there yet. So we're not at the stage where literally tomorrow, you're going to turn stuff on and going start doing this stuff. So we've got time. And is what I'm saying. you, you have time right now to plan this. If you're a company, you are a value added reseller or an agency that provides professional services, you have time to figure this out, but that time is going come really fast.
[00:19:51] Paul Roetzer: So I think 2024, there's going to be some breakthroughs in the ability to build these agents. Actually, NVIDIA, we'll talk next week about this, [00:20:00] but Jim Fan announced they just started whole new. Research arm within NVIDIA that obviously has probably more funding than any research firm in world to build agents.
[00:20:09] Paul Roetzer: Like everybody's going to these things. And so I think 2024 is going to see some breakthroughs and we're going start to see very early adoption, almost of like. Maybe where we are with AI today, where you're starting to find those use cases and like adoption is going to move. think 2025 might be where we are today with generative AI, with AI agents.
[00:20:31] Paul Roetzer: Like by this time next year, you're going to have some leading corporations that are starting to infuse AI agents at a more regular level, maybe through like Microsoft Copilot or something. But probably by 2026, 2027, tech is now there. We've solved all the privacy concerns. And now you're going to see just a takeoff point.
[00:20:51] Paul Roetzer: So I think that there is, again, there's time plan for this, but this may be way more [00:21:00] disruptive than generative AI on its own. And, and so as, as mind boggling as AI is, and that we're still trying to wrap our heads around this. When we have true AI agents, two, three, four years from now, it's like reinvents knowledge work.
[00:21:18] AI startup, Groq, goes viral
[00:21:18] Mike Kaput: another huge kind of development that's going to, I think, have some ripple effects in the world of AI is this AI startup that's going viral, and it's going viral in part because
[00:21:30] Mike Kaput: actually
[00:21:30] Mike Kaput: gunning for pieces of NVIDIA's business. This AI startup is called Groq. Now that's G R O Q, not G R O K, the name of Elon Musk's AI company.
[00:21:44] Mike Kaput: Now this Groq, Groq with a Q, the one we're talking about, Actually, apparently had this name far before Elon Musk's AI company. But when we're talking about Groq moving forward, it's this startup, not the one that must go. So Groq with Q is [00:22:00] an AI tool that serves responses to chat queries insanely fast.
[00:22:06] Mike Kaput: And it does that thanks to customized chips and a really novel software design. So it started going viral recently after users compared its speed in benchmark tests. And Groq just like crushed OpenAI's ChatGPT. And some estimates say it's even up to 13 times faster than ChatGPT. It's pretty insane. And what's notable about this is that the system is built with these custom made AI chips called Language Processing Units, LPUs, and these basically allow the company's tool, its chat interface.
[00:22:41] Mike Kaput: to serve results from popular language models like Mistral and Llama almost instantaneously. So Groq describes LPUs as, quote, a new type of end to end processing unit system that provides the fastest inference for computationally intensive applications with a [00:23:00] sequential component to them, such as AI large applications, LLMs.
[00:23:04] Mike Kaput: Now, this is a big deal because If we're able to serve up answers almost instantaneously from LLM based tools, this opens up tons of new use cases in AI for business. And it also just makes all sorts of use cases that previously were too slow to reach commercial production very commercially possible.
[00:23:27] Mike Kaput: Groq system could also threaten NVIDIA's business if companies start demanding LPUs instead of GPUs thanks to all these speed gains. Groq also claims it can deliver performance for way cheaper than NVIDIA. Now, while this is Groq's kind of moment in the spotlight, This company has been around since like 2016.
[00:23:47] Mike Kaput: It's founder and CEO Jonathan Ross. He helped invent the Tensor Processing Unit, the TPU at Google, which is that company's custom AI chip. So this is definitely a [00:24:00] company and a technology and an approach to serving up LLM results that is definitely worth watching. So Paul, I wanted to kind of get your opinion on what's
[00:24:09] Mike Kaput: on
[00:24:09] Mike Kaput: with Groq.
[00:24:10] Mike Kaput: Like have they really cracked the code here on Some better way to run AI applications.
[00:24:17] Paul Roetzer: Yeah,
[00:24:18] Paul Roetzer: talk, about Matt Schumer in a little bit and his AI agent work, but Matt Schumer, our friend at Hyperedit is actually the guy who set this all off. So he tweeted on February 18th, Let's
[00:24:30] Paul Roetzer: he said,
[00:24:31] Paul Roetzer: wild tech, you have to try.
[00:24:32] Paul Roetzer: And then he put the groq. com. They're serving Mixtral, which is open source model at nearly 500 tokens per second, answers are pretty much instantaneous, opens up new use cases and completely changes the UX possibilities of existing ones. So I went and tried it that night. I actually saw the tweet from Matt and I was like, oh my God, like it is insanely fast.
[00:24:54] Paul Roetzer: So you can, again, you can go try it for yourself. Just go to the homepage. You have to put any information in, just put a search in and a prompt in, and it, you're good. [00:25:00] Instantly just generate something. It's wild. So I think the main thing here is you have to remember like why NVIDIA built GPUs. It wasn't originally for AI training and inference.
[00:25:11] Paul Roetzer: They built GPUs for video games, graphics processing units. Like, so that. That was like the whole model. So these GPUs, while they had, while they learned around 2011 that they were, very, very powerful at training models, that wasn't what they were originally for. So there's some complexity in the engineering.
[00:25:32] Paul Roetzer: There's some complexity in how they do what they do. That these LPUs just sort of cut out because now they're basically just building something specific for, and it's funny to talk about Groq as a startup. It's a company, seven years old, like eight years old, they were starting in 2016. So, but it sort of like emerged last week.
[00:25:48] Paul Roetzer: I'd never heard of it. And we pay pretty close attention to this space. It wasn't a company I was aware of. so I dunno, my, my thought is, uh. I mean, doesn't Nvidia just buy them? [00:26:00] Like if they're really a threat, like does it somebody or just replicate what they're doing? Like, I don't know. So I think it's amazing innovation.
[00:26:09] Paul Roetzer: just buy
[00:26:10] Paul Roetzer: I think it's awesome that companies are doing this kind of thing, but I don't really worry it's going to like directly impact Nvidia's market share and dominance, and even there was a wall street journal article where they said.
[00:26:24] Paul Roetzer: CEO,
[00:26:28] Paul Roetzer: Jonathan Ross, said the company's on track to deploy 42,000 chips this year, and, 1 million in 2025, but they're exploring increasing those totals to 220,000 this year and 1.5 million next year.
[00:26:44] Paul Roetzer: So, for context,
[00:26:46] Paul Roetzer: in 2025
[00:26:48] Paul Roetzer: last year, financial Times reported that Nvidia was tripling its production of its H 100, which is, its. most modern GPU in 2024. The goal is produced, produce [00:27:00] 2 million in 2024. so again, LPUs from Groq right now on track for 42, 000 chips this year. GPUs from NVIDIA, 2 million this year.
[00:27:15] Paul Roetzer: So it's not like. All of a sudden they're going to show up and just like take market share. But I think it's very much like, you know, ChatGPT or Perplexity, like we talk about, where all of a sudden it's just like this amazing phase of innovation being fueled by AI, where no business seems safe, like Perplexity out of nowhere, all of a sudden just starts getting all this love as a.
[00:27:38] Paul Roetzer: As alternative to Google. You and I both like love perplexity now and use it all the time. And so you see perplexity and open AI come up against Google and force them to take action. You have Groq all of a sudden is like, people are like, Hey, wait, is this threatening NVIDIA?
[00:27:51] Paul Roetzer: And
[00:27:52] Paul Roetzer: so I think like the main takeaway here for me is. People have probably heard us talk about this idea of the future of all businesses, AI or obsolete. [00:28:00] And so I wrote this post last year that, that basically have three options with your company. AI native, AI emergent, and obsolete. So AI native is you build a smarter version from the ground up. That sounds like, kind of like what we're doing here with Groq.
[00:28:12] Paul Roetzer: know, G R O K or Q Groq, not the other one. perplexity, same deal. Like you come at the big players with a smarter, more affordable.Faster approach, Andthat's the AI native model. The AI emergent is got
[00:28:27] Paul Roetzer: figure out how to be innovative again and how to push product to market like Google's having to do now, like they're being pushed hard to function in a way they're not comfortable with.
[00:28:37] Paul Roetzer: So this is the innovators dilemma, Clayton Christensen, like it's really hard when you're a big established enterprise to move fast and break things and be willing to put products out that have maybe a higher risk. Then you're, you know, our liabilities than you're used to doing. So I think that the main takeaway for me is every business in every industry faces the opportunity to disrupt and has the risk of being [00:29:00] disrupted.
[00:29:00] Paul Roetzer: And you pretty much have to live that way moving forward. I don't care what company you're in and what industry it is. You have to assume someone is going to build a smarter version of your company. And it's way better to be the one that does that yourself. Be the AI emerging company that figures this out and says, what's a smarter version of our business.
[00:29:19] Mike Kaput: Yeah, kind of like we were talking about in the previous topic with how VARs and agencies need to start planning. You almost have to assume Okay, even if Groq is not making all these chips tomorrow or if NVIDIA is not buying them tomorrow, this technology exists. So you almost have to like plan. What does that actually mean when several years from now, it's everywhere, right?
[00:29:41] Mike Kaput: Same as AI agents and you really have to challenge yourself, it sounds like, to think using a blank piece of paper instead of your
[00:29:49] Mike Kaput: preconceived
[00:29:50] Paul Roetzer: Yeah, and I know you've read the Elon Musk biography with Walter Isaacson. I'm almost done with it now, and Regardless of what you think about [00:30:00] Elon, one, that book makes it very clear why he does what he does.
[00:30:04] Paul Roetzer: Like, his tweets and everything else makes way more sense once you read that book, but to all of the challenges of Elon and his personality and the way he approaches things, he is like the Prototype of first principles thinking like everything he does. you're never going to see an Elon run company face an innovators dilemma.
[00:30:26] Paul Roetzer: I don't think like they are at their core, disrupting themselves every single day at a rate that's almost impossible to keep up with and. You know, I think there's probably some balance between the way they do that and the way he runs companies and maybe how the average business leader would run a company.
[00:30:44] Paul Roetzer: But I think there's something you can take from his algorithm for how he analyzes and builds companies, putting all this other stuff aside. and I think that's, Again, if you haven't read that book, I would suggest it because I [00:31:00] think you'll understand what becoming an AI emergent company really means, when you, when you understand how he builds businesses.
[00:31:07] Paul Roetzer: And again, I'm not endorsing
[00:31:09] Paul Roetzer: the way he
[00:31:10] Paul Roetzer: goes about doing it, but I think there's principles of it that you can take away and apply in maybe a more human centered way, let's say.
[00:31:20] The true impact AI is going to have on your work
[00:31:20] Mike Kaput: This actually leads really nicely into our third topic, because we're talking a little bit more about. Some new material and new thoughts around the true impact that AI is going to have on your work.
[00:31:33] Mike Kaput: And thanks to just some of the velocity of the AI advancements we've seen, especially in the past few weeks, this is a question that increasingly business leaders need to start answering fast.
[00:31:44] Mike Kaput: for instance, you know, Google's released a Gemini 1. 5. We now have a model that can handle 1 million tokens in its memory, which is about 70, 750, 000 words.
[00:31:57] Mike Kaput: Google's working on systems that will reach 10 million [00:32:00] tokens that they've at least publicly talked about. That's 17, 000 pages of content. These systems are getting much, much faster, like we saw with Groq, and AI expert Ethan Mollick this past week really summed this up nicely. He said, Speed and memory are both vital to making AIs more usable and powerful in the real world.
[00:32:20] Mike Kaput: Imagine feeding AI hundreds of pages of instructions on how to do something, and then having it quickly do exactly that. In one experiment, he fed GemIIni recently, this past week, thousands of pages of his work, and the tool summarized, understood, quoted from the work in literally less than a minute. He went on to say the advent of massive context windows gives AI superhuman recall and new use cases.
[00:32:45] Mike Kaput: He noted that this single use case would have taken human researchers days to do on their own. Really, the point here is we're hitting almost this inflection point where these tools are really going to have this insane, profound [00:33:00] effect on knowledge work as we know it. And Paul, this is kind of a opportunity and warning that you've been talking about since day one of Marketing AI Institute that business leaders must understand this technology and what it's capable of as quickly as possible.
[00:33:18] Mike Kaput: Now
[00:33:18] Mike Kaput: on LinkedIn this past week, you actually gave a really cool example of how practically business leaders can start doing that. Could you walk us through
[00:33:26] Paul Roetzer: that? Yeah, so first a quick note on the Gemini 1. 5, you know, we talked about that, I think it was last week or the week before when they announced it, we're going to talk about Some negative stuff with Google in a minute, but I want to say like the people I follow who have access to 1.
[00:33:43] Paul Roetzer: 5 are blown away by it. so the early feedback, like Ethan Mollick, who has access to it as a researcher is just, they say it's mind blowing basically of what it's. What it's capable of. And it does already seem to be as powerful as their ultra model.
[00:33:59] Paul Roetzer: as [00:34:00] a
[00:34:00] Paul Roetzer: so just keep an eye on Gemini 1. 5. I know their whole naming conventions are super confusing, honestly.
[00:34:06] Paul Roetzer: Like we need like a checklist ourself to just like reference back. What, what 1. 5 is versus. Pro and Ultra and we're, I don't know, it's like wild, they're brand new stuff. But anyway, um, yeah, so the LinkedIn post you're referencing on the true impact AI will have on your work. It was one of those, like, I just got up Saturday morning and I had like 10 other things I needed to do, but it sort of like took over my mind of, it was kind of on the heels of the HubSpot platform example, the AI impact assessment where I was going through and saying, look at the features, like.
[00:34:38] Paul Roetzer: And what I realized is you and I, Mike, spend a ton of time, especially in our talks and the workshops we run, trying to make AI as approachable and actionable as possible. And so we, we constantly try and develop frameworks that make it really simple for people to understand the impact on themselves. And so in this case, I was trying to show like.
[00:34:58] Paul Roetzer: Listen, if you just [00:35:00] take the thing you do all the time and just find ways to infuse AI, you're going to see success with it. And so I happen to be like, I was trying to think like, what can I show as an example? And then I realized like, okay, I'm, I'm knee deep in planning our marketing AI conference agenda for September.
[00:35:17] Paul Roetzer: And once we get through writer's summit next week. I got to finalize getting all these speakers recruited and getting the agenda locked in. And so rather than going into Asana, which is where we keep all of our tasks and me actually like building the official task list, I was like, let me just do this really fast as a demonstration.
[00:35:33] Paul Roetzer: And so I went into ChatGPT and said, you're an event planner responsible for creating the event agenda, researching speakers and selecting speakers. Build a task list of all the activities involved. And then it like, it gave me a list pretty quickly. Not as fast as if I had used an LPU, but it came out fast.
[00:35:50] Paul Roetzer: and it gave me phase, task, and details. And so I said, can you give me this in a CSV format for download? So it did. And I downloaded the CSV. And so [00:36:00] then I took that CSV and I added a couple of columns, like estimated time to do the tasks, all human, estimated time if I used AI. Estimated time saved with AI and then like sample AI tools I would use and kind of how I would use them.
[00:36:14] Paul Roetzer: So again, you can go to LinkedIn and you can download this thing yourself and see this. I think it's really helpful because I actually went to the notes column. I was like, Hey, listen, I would use ChatGPT to do this. I would use Perplexity for this. And so in total.
[00:36:27] Paul Roetzer: There
[00:36:27] Paul Roetzer: was 16 tasks related to the speaker, know, identification, recruitment, and processing.
[00:36:33] Paul Roetzer: About a hundred or 220 hours of work manually. If I did this all the traditional way, it would take me about 220 hours of work. So I went through and found out of the 16 tasks, nine of them where I could infuse AI in some way, at some level, Um, those nine tasks would save me 42 hours or about 19 percent of the time.
[00:36:55] Paul Roetzer: So this isn't even like.
[00:36:57] Paul Roetzer: This
[00:36:58] Paul Roetzer: isn't an advanced use case where there's [00:37:00] tons of it. There's a whole bunch of tasks here where you just got to do it. Like, there was an example, like writing personal thank yous at the end. I said, this isn't an AI thing. This has to be personal and it has to be human based.
[00:37:11] Paul Roetzer: Like this has to be you. So there was a bunch of stuff in here that really you can't shortcut with AI. And even in that environment. It was still saving almost 20 percent of the time. And so my point with this post, which is a pretty quick post, like I didn't go into a bunch of details, it was kind of like real quick thoughts, and then like, here's a sample worksheet you can use.
[00:37:31] Paul Roetzer: My whole point was, if you want to find ways to apply AI, think about the projects, the tasks, the campaigns that you're already doing. And just do something like I did, where you go through and say, okay, which tools could I use at a task level? To help me here. And my guess is you're going to unlock 20 to 40%.
[00:37:50] Paul Roetzer: Like there's plenty of other things, Mike, you and I do where I could have easily seen a 40 percent savings in time with AI. Again, this was, it's hard to take the [00:38:00] time out. For example, when I'm recruiting speakers for the main stage. I've, I watch year round for people and I keep a list, a sandbox of potential speakers.
[00:38:09] Paul Roetzer: When I go to actually recruit them, I will sit there and watch an hour video of their presentation to see what kind of presenter they are, to get their point of view on things. You can't shortchange that. I can't ask AI to go watch it for me and tell me if I should You know, bring this person in. That agenda is the entire MAICON experience.
[00:38:25] Paul Roetzer: Like if I bring in a crappy speaker, you don't have a great experience and you're not coming back.
[00:38:31] Paul Roetzer: I
[00:38:31] Paul Roetzer: turn that over to AI. It might help me identify some people I don't know. It could help me assess maybe things they've written. But at the end of the day, I got to watch the video myself because it's on me, whether or not they're a good speaker when I bring them in.
[00:38:44] Paul Roetzer: So I just, yeah, I think like, one of the ways we teach in our Applied AI workshops is like job description, Mike, where we'll have them go through like, what do you do? What are the 20 responsibilities? And I think this is just kind of a continuation. It's like another way to look at it. Take the campaigns you run, the projects you're doing, [00:39:00] and just find ways to infuse AI within these.
[00:39:03] Mike Kaput: sort
[00:39:04] Mike Kaput: of complimentary to this, I liked a lot when Ethan Mollick was talking about this subject. He had these four questions that he recommended that companies ask themselves in the face of all this rapidly advancing AI. But also I think these are really good questions to ask as an individual in
[00:39:21] Mike Kaput: career
[00:39:21] Mike Kaput: as well.
[00:39:22] Mike Kaput: And very quickly they are, what useful thing do you do that is no longer valuable? What impossible thing can you do now? What can you move to a wider market or democratize? What can you move upmarket
[00:39:35] Mike Kaput: personalize? So I think.
[00:39:37] Mike Kaput: You do, in addition to reviewing the really in the weeds practical tasks, campaigns, responsibilities you have every day, it also requires the minds that shift that simply some of the stuff you do you simply do not need to do the way you've always done it and there may be better uses of your time than what historically you have been [00:40:00] tasked with in your role or in your company or in your market.
[00:40:03] Mike Kaput: So I'm curious, Paul, like Did you, what were your thoughts kind of on these questions? Anything you'd add to this?
[00:40:10] Paul Roetzer: It kinda reminded me of that post I wrote the day that I saw DALL-E 2. So when DALL-E 2 came out in 2022, wrote something, where I basically was trying to grapple myself with the impact I was gonna have on knowledge, work and creatives.
[00:40:26] Paul Roetzer: So my wife, again, if people don't listen to the show before, my wife is an artist. my daughter at 12 wants to be an artist. My son wants to be a video game developer at 10. I'm a writer by trade, came outta journalism school. And so when I saw Dolly. I realized everything was changing, that AI was going to be creative.
[00:40:44] Paul Roetzer: It was going to be able to think and understand and reason all the things we'd thought was going to happen were happening. And so the post I wrote was trying to understand this and trying to help other creatives understand it. And so what I wrote at the time was, you have to come to grips with what will be lost, [00:41:00] what will be gained and when.
[00:41:02] Paul Roetzer: And the, what will be lost is like, what is the AI going to do that I thought? I was uniquely capable of doing. What is the thing that
[00:41:09] Paul Roetzer: kind
[00:41:09] Paul Roetzer: of defines me as a person, as a professional, that at some point I have to accept AI is now able to do, and that's a hard, that's the hard part, the positive perspective is, but what am I now able to do?
[00:41:22] Paul Roetzer: I couldn't do before. How do I become a better artist? How do I, you know, enhance my creativity? How do we become more innovative? And then when is it going to happen to me based on this trajectory of AI capabilities? And so. It fits really well with this AI agent conversation, because I think there's going to be a lot of things in the next couple of years that as knowledge workers and creative professionals, we have to come to grips with AI is probably going to be better than us, or at least at human level, um, expert human level in a lot of things that we currently think only humans are capable of doing.
[00:41:56] Paul Roetzer: And that's, again, it's not like [00:42:00] You and I aren't here cheering on AI agents saying, Oh, I can't wait till they just take all, all the knowledge work from us. We're realists of like, I don't see how this isn't the outcome, but we have time now to think about this. And so the, and when part is, AI agents are coming, when might be 12 months from now for some of you, it might be three to five years from now for others.
[00:42:23] Paul Roetzer: It's just a matter of time though. And so I think, like, kind of like Ethan Malik is saying, you got to kind of look at what are we going to be good at still? What is the thing we're going to be capable of? And let's start preparing for that moment.
[00:42:35] Reddit goes public
[00:42:35] Mike Kaput: Alright, let's dive into our rapid fire topics this week.
[00:42:40] Mike Kaput: First up, the social media site Reddit is going public. And its IPO filing reveals that it actually has some data sharing arrangements with companies. specifically AI companies, where it is licensing Reddit data to train their models. Reddit says that [00:43:00] these deals are worth more than 200 million in total.
[00:43:04] Mike Kaput: They did not reveal, however, which AI companies are licensing data from its platform. Bloomberg has reported that a single, quote, large unnamed AI company had entered into a 60 million per year. Licensing deal. Now, all we can do is speculate here, but it is also worth noting that OpenAI's CEO, Sam Ullman, has an 8.
[00:43:25] Mike Kaput: 7 percent stake in Reddit, which makes him the third largest shareholder of the company.
[00:43:31] Mike Kaput: So
[00:43:32] Mike Kaput: why is this such a big deal that AI companies are apparently licensing data
[00:43:37] Mike Kaput: from Reddit?
[00:43:38] Paul Roetzer: we've kind of known this. This is, I thought it was,
[00:43:41] Paul Roetzer: Isn't it Google that had the Licensing deal with them?
[00:43:47] Mike Kaput: Yea I am not sure if Bloomberg revealed it or I think they had said it was like an unnamed.
[00:43:53] Paul Roetzer: Okay, I thought I saw another source that said it was
[00:43:55] Paul Roetzer: Google, which I
[00:43:56] Paul Roetzer: thought was odd given Sam's stake in opening, [00:44:00] Reddit, we'll, we'll We'll check that and put it in the show notes if we have it, but it is, it is one of them. I thought I saw that it was actually Google. um, regardless, it was interesting to see Sam's massive stake in this.
[00:44:15] Paul Roetzer: That is going to be worth a lot of money. Um, we've talked about this before. The future of these models is going to be licensed and synthetic data. Like they There's these ongoing debates whether or not they were legally allowed to take all the copyrighted material they had. So we've known that all these proprietary data sets are going to be worth a lot of money to properly train these models.
[00:44:39] Paul Roetzer: And so GPT 5 and beyond, Gemini, Ultra 2 and beyond, like, they're all going to use synthetic data so that the AI generates, and they're going to use licensed
[00:44:49] Paul Roetzer: It's going
[00:44:49] Paul Roetzer: to be probably the primary training sets. It's why Elon Musk turned off access to Twitter's feed, so that only XAI, his AI [00:45:00] company, and Groq with a K, has access to Twitter because they think that that's what will make Groq unique and different than everybody else, is it's the most real time data that represents everything happening in the world, and so when you combine that with Tesla and its worldview.
[00:45:15] Paul Roetzer: You now have like this amazing potential model. So it's not surprising at all that Google or whoever has this major licensing deal with them. they're all gonna probably do it. That's how they're gonna make money in the future. And so I think, you know, this impacts media companies. It impacts online sites that have this proprietary data impacts your company.
[00:45:36] Paul Roetzer: If you have a bunch of proprietary data, licensing of data is gonna be huge moving forward.
[00:45:42] AI Agents in the wild
[00:45:42] Mike Kaput: All right,
[00:45:43] Mike Kaput: so related to. Some of the things we've already been talking about, we just got even more proof of AI agents starting to come out into the wild. You know, our friend Matt Schumer, who we talked about before at the AI Company Hyper Right, announced that their Agent Studio product would [00:46:00] be shipping this week.
[00:46:01] Mike Kaput: So Agent Studio allows you to simply show an AI how to do a repetitive task online in your browser, and then it will go do it for you every time after. Now, this is just one example of. The AI agents that we're seeing in the world. I mean, there are now multiple ways to train AI agents to do things for you.
[00:46:21] Mike Kaput: There was another really interesting example from a user on X named Paige Bailey. Paige recorded a screen capture of looking for an apartment on Zillow, and then Google Gemini was then able to generate Selenium code, which is a type of code that allows users to automate browser actions. And it was able to help the AI go ahead and perform the task.
[00:46:44] Mike Kaput: the result was an application that opened Chrome, navigated to Zillow, typed in Cupertino, California in the search bar, clicked on for rent, set the price range up to 3, 000 a month, and set the bedrooms to two or more, then clicked on the [00:47:00] apply button. So We've talked a lot about AI agents this year so far, including in our main topics.
[00:47:06] Mike Kaput: mean,
[00:47:06] Mike Kaput: Paul, is it time for business leaders to start experimenting with things like AgentStudio once it comes out this week?
[00:47:14] Paul Roetzer: Uh, if you have like research arm that's closely guarded on usage, probably like you can definitely experiment with the technology. I wouldn't, again, because you're going to give up visibility to everything you do on the browser.
[00:47:29] Paul Roetzer: this is probably like a limited. Roll out where you're, you're, you're kind of in testing mode. Um, Paige actually works for Google DeepMind and she, mind officially blown and she works with this technology. And I, like when I saw that tweet, my initial reaction was, is this an emergent capability? Like, did they not know it was able to do this, like to basically build an agent, which wasn't something we thought it did, or is this like.
[00:47:58] Paul Roetzer: A hidden [00:48:00] feature within Gemini that they're going to unleash. And I actually don't know the answer, but the fact that it can do it is kind of wild. and it shows you how prevalent these things are probably going to be when you got to assume ChatGPT is going to have these same capabilities. And now whether you're using the AI agent studio or you're using Adept or multi on or whatever it's going to be, there's going to be.
[00:48:25] Paul Roetzer: Explosion
[00:48:26] Paul Roetzer: of AI agent capabilities. And again, I think this year is going to be early. Testing is good. Start to see what they're able to do if you're willing to give up the privacy side.
[00:48:36] Paul Roetzer: um, but
[00:48:38] Paul Roetzer: it's. It just seems like, I don't think we've had our ChatGPT moment yet. And I think that's what I'm going to wait to see is do we have a ChatGPT like moment in 2024 where AI agents start to go mainstream where, you know, you have, like with ChatGPT a month after it had it, you had a hundred million users, like, are we going to have that kind of moment and is it [00:49:00] going to come from like a hyperwrite?
[00:49:01] Paul Roetzer: Kind of a true startup within the space, or is Google or open AI or somebody going to turn on an AI agent capability, Microsoft, where all of a sudden it just becomes everyone's like experimenting these things and seeing what's possible and setting up workflows for their email campaigns and all that stuff.
[00:49:20] Paul Roetzer: I don't know that that'll happen this year, but it's certainly a possibility.
[00:49:23] Google Gemini Image Generation Issues
[00:49:23] Mike Kaput: So, Google actually ran into a little trouble recently with Gemini. It recently disabled a feature in Gemini that generated images of people. And it did that because the results were becoming wildly inaccurate in certain cases, and sometimes offensive to certain people.
[00:49:44] Mike Kaput: the controversy centered around some users attempting to generate images of historical figures and that included Gemini ending up generating images. that were not exactly [00:50:00] historically accurate. They were presenting images like racially diverse Nazis and the U. S. Founding Fathers in a number of different races.
[00:50:09] Mike Kaput: Now, this was after people had been asking for kind of historical figures as they were in history, not kind of with Google adding its own spin on things. So Google apologized that the image generation wasn't working as well as it should. And it said it built GemIIni with guardrails to make sure that it doesn't generate, you know, violent or explicit images.
[00:50:32] Mike Kaput: And because the company's users come from around the world, it also built it to make sure that you get a range of different ethnicities or other characteristics when you type in an image generation prompt.
[00:50:44] Mike Kaput: what
[00:50:44] Mike Kaput: they say what happened here is these kind of guardrails that were intended for really positive outcomes, the tuning of them failed to account for cases where images definitely should not show a range of people.
[00:50:56] Mike Kaput: You should not be generating a very diverse [00:51:00] crowd of, say, Nazis. And over time, this model ended up becoming really cautious, much more so than Google intended, and even refused to answer certain prompts entirely. So this whole thing has people online. In certain camps claiming that the model was politically or culturally biased.
[00:51:21] Mike Kaput: It's become this major social flashpoint. Google is claiming that it's really just some of the engineering having some unintended consequences. So, Paul, this rubbed
[00:51:34] Mike Kaput: some
[00:51:35] Mike Kaput: people the wrong way. what's going on here? Like how seriously should we take the fact that this might be biased politically, socially, culturally, Google's explanation?
[00:51:46] Mike Kaput: Can you kind of unpack
[00:51:47] Paul Roetzer: this for us?
[00:51:48] Paul Roetzer: If you're new to how these models are trained and how they work, this may seem kind of shocking to you. Uh, this is widely [00:52:00] known to be a major challenge with these models. This is, this is how this works. Now, the problem is if you live in the Twitter echo chamber,
[00:52:09] Paul Roetzer: widely
[00:52:10] Paul Roetzer: political and it's increasingly by the minute becoming more political because Elon Musk and his allies are using this as basically, a flashpoint to try and like, go at Google and try and make the case that Grok won't have these problems, because Grok going to be truth seeking, and it's just like,
[00:52:29] Paul Roetzer: it's
[00:52:29] Paul Roetzer: maddening.
[00:52:30] Paul Roetzer: I mean, personally for me, I can't stand this stuff. Like, Everybody has to make everything political, even if it's not. So, Google admits, and you can, in the show notes, you can go read the blog post where they explain what happened. So they admit they screwed up, but the question becomes whether or not their culture will actually enable them to fix it.
[00:52:52] Paul Roetzer: So if you're in the Elon camp and his allies, They're basically going to be of the opinion of there's a bunch of people [00:53:00] within Google who did this on purpose, and they don't want to fix this. This is, unless they clear house at Google and put a new CEO in place, who like forces change to their culture.
[00:53:12] Paul Roetzer: This is basically what you're going to get. It was not an accident. It was a design choice made by humans who now need to be fired. That is thelike /EAC. Opinion of what is going on right now.They
[00:53:26] Paul Roetzer: don't bother addressing fact like everybody has had these issues, open AI, stability, everybody, anybody who releases a model has to deal with the same problems, which is these things are trained on data available on the internet.
[00:53:39] Paul Roetzer: And if you go search. For anything on Google or any search engine, it's going to largely come back with kind of biased data. So if you say, show me a pictures of a doctor or show me a picture of a lawyer, show me a picture of a CEO, you're going to generally get white males. Like it's just what the internet produces.
[00:53:59] Paul Roetzer: And so [00:54:00] these models are kind of trained on what will come back.
[00:54:05] Paul Roetzer: so what they do is they go in and they kind of try and fine tune these things so they're more diverse, which is a logical thing do.
[00:54:14] Paul Roetzer: Now in Google's case, though, what they're saying is like, we, we went overboard. Like that, that we're trying to show so much diversity that when you ask for a picture of the Pope, you should get A picture of the Pope.
[00:54:27] Paul Roetzer: You shouldn't get a diverse representation of what the Pope should be in the future. And that's basically what they're doing is they're trying to show this diversity. So unless you put out a model and just put it in the wild and let it do everything it's capable of, make everything, it can be racist issues, sexually explicit issues.
[00:54:45] Paul Roetzer: It can be all these things. That's the open source models. Let's just put this thing out there and let people get whatever they want out of the thing. Unless you put the guardrails in place. That's what's going to happen. So the people that are doing this could [00:55:00] have zero political motivations. They may just be trying to do what is ethical, moral, aligned with reasonable human values.
[00:55:08] Paul Roetzer: But everybody's going to turn into a political thing, either because they can gain points for their own AI system or because they actually believe that it's a political thing and there's some conspiracy going on here. This is why it drives me nuts. It's like, we can't just get to.
[00:55:24] Paul Roetzer: A solution, but the problem is
[00:55:28] Paul Roetzer: there isn't really one that the challenging thing is if you go and do like, I saw one about, like how to butcher a cow, like basically how to make steak, Gemini won't tell you how to do it.
[00:55:41] Paul Roetzer: Like it won't give you the images. It won't give you videos. It'll basically say you like, shouldn't cut up a cow in essence, but you could go do the exact same search in Google and get a bunch of YouTube videos, how to's, like all this stuff, you can go do it on YouTube and you can get a bunch of videos showing you how to do it.
[00:55:59] Paul Roetzer: So [00:56:00] it's, it's really weird.
[00:56:02] Paul Roetzer: Google as a company allows all this stuff in search and YouTube, but yet their own model doesn't seem to do it. So my feeling is because the way these models are trained and the way they work, this is a very challenging technical issue. And it's a challenging societal issue, but it becomes really important because the more people rely on these models for their information, they most, the more they go to like a Perplexity or a ChatGPT instead of a search or instead of YouTube directly, we're going to start having a generation learns what is true and what is not based on what these models output.
[00:56:40] Paul Roetzer: The problem is there's no quick fix. And I think what's going to happen, and we've seen like OpenAI, again, when they got lambasted, when they first came out and it was claimed it was too far to the left, so they pushed it to the right and then they couldn't get to the middle politically.
[00:56:53] Paul Roetzer: Sam said is, listen, we're going to make this thing where you can just make it whatever you want it to be.
[00:56:58] Paul Roetzer: So I think what's going to [00:57:00] happen is you're going to basically have like. Temperature settings on your models in the future, on your OpenAI, ChatGPT, or your Gemini, that's like, give me the explicit stuff. Like, I don't care. Just give me like, whatever this model is capable of, or like, no, let's tune this thing to be politically this way or that way, or like 13 and under or like.
[00:57:16] Paul Roetzer: Whatever. I think that's the only way to solve this is to let individuals choose what sort of experience they want. through settings. Otherwise, we're going to keep having these arguments all the time.
[00:57:30] Air Canada forced to give refund to passenger due to AI chatbot mistake
[00:57:30] Mike Kaput: in another example of AI gone a bit wrong, Air Canada, the airline, was forced to give a partial refund to a passenger thanks to a pretty serious AI mistake.
[00:57:40] Mike Kaput: So, a customer named Jake Moffat He went to Air Canada's website to book a flight, unfortunately, on the day his grandma died. He wasn't sure what policies the airline had around bereavement rates, which are these flexible rates. They are sometimes offered by airlines to people who are in need. So, [00:58:00] Jake did what many of us would do.
[00:58:02] Mike Kaput: He asked the company's chatbot. Unfortunately, the chatbot provided inaccurate information about the company's policies. It encouraged him to book a flight and then, due to the policy, request a refund after booking the flight. This turns out to go against the company's actual guidelines. Moffitt didn't know that.
[00:58:23] Mike Kaput: So, after he books his flight, his request for a refund is denied. And Air Canada, furthermore, refuses to admit fault and budge on the refund. The company actually said that because the chatbot response linked to a page with the actual policy, Moffitt should have known he couldn't request a refund. Moffat was unsatisfied with that, and challenged the ruling in Canada's version of Small Claims Court.
[00:58:51] Mike Kaput: And he mostly won, receiving a partial refund, but what's really interesting and wild here is Air Canada's argument in court [00:59:00] against
[00:59:00] Mike Kaput: According
[00:59:00] Mike Kaput: to Ars Technica, the company essentially argued that Moffat should never have trusted the chatbot, and that it's not liable for the chatbot's misleading information.
[00:59:11] Mike Kaput: Because the chatbot is, quote, a separate legal entity that is responsible for its own actions. Now, Paul, this is just one kind of crazy story about AI gone wrong, and I'm no lawyer, so maybe this legal argument from Air Canada is less than watertight here. But as a business leader, like, what should we be thinking about here?
[00:59:33] Mike Kaput: I mean, should we not be using chatbots on our website?
[00:59:37] Paul Roetzer: Yeah, so just for context, this happened in November 2022 before ChatGPT came out. So it wasn't like ChatGPT gone wrong per se, like current model. But I do think that the overall like lesson here is there are instances where generative AI is phenomenal and you should be racing to infuse it into your workflows.
[00:59:56] Paul Roetzer: And there's instances where humans have to remain in the loop. And if you [01:00:00] run the risk of liability, we saw one not too long ago with like a car dealership, I think, where they sold a car to somebody for like 200 instead of 60, 000. There's enough stories out there to know that. Relying on a chat bot to give factual information, isn't probably there yet.
[01:00:18] Paul Roetzer: We haven't eliminated the hallucinations and people can mess with these things. Like he didn't do it intentionally, but you can get them to give you things by just kind of working with them. So I think it's buyer beware, like just. Be cautious of the use cases you are applying AI to, and I don't know that we're quite there yet where generative AI replaces the need for a human to be in the loop on customer service related items.
[01:00:43] Efforts to Fight AI Deepfake and Protect Elections
[01:00:43] Mike Kaput: So this past week we also saw a couple new initiatives to combat the growing problem of deepfakes. So first up, some leading technology companies signed an accord to combat the deceptive use of AI in 2024 elections worldwide. This is [01:01:00] called the Tech Accord to Combat Deceptive Use of AI in 2024 Elections, and it's a set of commitments to deploy technology that counters AI generated content meant deceive voters. Has been signed by companies like Adobe, Amazon, Google, IBM, OpenAI, TikTok, and
[01:01:18] Mike Kaput: and more. And as part of the accord, these companies are pledging to work together on tools to detect and address deepfakes and then drive education around them. Now, at the same time, over 300 tech and AI experts have released an open source have
[01:01:34] Mike Kaput: released an open letter urging immediate government action to combat the rising threat of deepfakes So some of the notable people that signed this include the former US presidential candidate Andrew Yang Stuart Russell who's a big AI voice and expert in the space and Francis Haugen who is a prominent meta Facebook whistleblower this letter supports all the legislative efforts to target what they're [01:02:00] calling the deepfake supply chain.
[01:02:02] Mike Kaput: And it makes some recommendations on how to start combating the problem of deepfakes. So, Paul, we talked at length about how important it is to combat deepfakes this
[01:02:13] Mike Kaput: Do
[01:02:13] Mike Kaput: these new efforts give you any more confidence we're working towards a viable solution?
[01:02:18] Paul Roetzer: No, but it's better than no progress. I'm, I'm very happy to see the movement and to see, you know, major companies getting involved.
[01:02:30] Paul Roetzer: It doesn't mean my confidence level has changed that we're actually going to have a solution in near term. But again, it is certainly better than. Not having this information to report on.
[01:02:40] Google Introduces Gemini Business
[01:02:40] Mike Kaput: Alright, So in our last topic here in a long list of rapid fire this week
[01:02:46] Mike Kaput: uh, Google has introduced Gemini Business, a new plan for companies that lets them use its Gemini models.
[01:02:55] Mike Kaput: And it's also. Another name change, another rebranding has [01:03:00] renamed Duet AI for Google Workspace to Gemini for Google Workspace. So this basically means Gemini, what we've been talking about over the last couple episodes, this is now going to be built into all of Google's Workspace apps.
[01:03:13] Mike Kaput: And it means that more businesses will be able to access Gemini through this new plan.
[01:03:20] Mike Kaput: So the plan gives you access to Gemini for Workspace for as low as 20 per user per month if you make an annual commitment. And companies can also buy Gemini Enterprise, which helpfully is replacing Duet AI. For Workspace Enterprise, are you confused yet? For 30 bucks per month with an annual
[01:03:41] Mike Kaput: So Gemini Enterprise has the same capabilities as Gemini Business, but you can have more usage before you hit usage limits. It also has a few more capabilities related to AI Power Meetings, including a bunch of translation features, and soon it will help you take meeting [01:04:00] notes.
[01:04:00] Mike Kaput: So Paul, I don't think we have any more time to go into all Google's naming conventions here, but seems like long story short, we're getting kind of almost like a ChatGPT Plus plan here for Gemini.
[01:04:13] Paul Roetzer: I think, and I, I have Duet AI for Google Workspace, I think that means I have Gemini Business now, or maybe I have Gemini for Google Workspace. I honestly don't know.
[01:04:25] Paul Roetzer: I will report back to you all when I figure out what we actually have and if it does what it's supposed to do.
[01:04:35] Mike Kaput: Yeah, I will simply say that at the very least it is worth it.
[01:04:40] Mike Kaput: Fighting through the jargon and terminology because at least for a personal Gemini Advanced account, which is a paid account Gemini is pretty breathtaking in my opinion So at least on a positive note, it's worth fighting through all figuring out all this stuff So you can actually use some really cool AI technology.
[01:04:57] Mike Kaput: Agreed.
[01:04:59] Mike Kaput: [01:05:00] All right, Paul Thank you so much for breaking everything down for us this week I want to remind our audience if you haven't already to subscribe to our newsletter. It's called This Week in AI You can find it at MarketingAIInstitute. com forward slash newsletter. We break down all the topics we just discussed even more in depth and also include all the other topics we didn't get to and don't have time for in a single podcast episode.
[01:05:24] Mike Kaput: And trust me, every week there are tons of them. It's simply the quickest way in addition to this podcast. To stay up to date on artificial intelligence. So go check that out. Paul,
[01:05:35] Paul Roetzer: thanks again. Thanks everyone for joining us. We will be back next week. and again, AIwritersummit. com as well. Don't forget that.
[01:05:43] Paul Roetzer: It is next Wednesday, March 6th. If you're a writer, editor, or content marketer, please join us. Thanks, Mike.
[01:05:49] Thanks for listening to The AI Show. Visit MarketingAIInstitute. com to continue your AI learning journey. And join more than 60, 000 [01:06:00] professionals and business leaders who have subscribed to the weekly newsletter, downloaded the AI blueprints, attended virtual and in person events, taken our online AI courses, and engaged in the Slack community.
[01:06:12] Until next time, stay curious and explore AI.
Claire Prudhomme
Claire Prudhomme is the Marketing Manager of Media and Content at the Marketing AI Institute. With a background in content marketing, video production and a deep interest in AI public policy, Claire brings a broad skill set to her role. Claire combines her skills, passion for storytelling, and dedication to lifelong learning to drive the Marketing AI Institute's mission forward.