This week on The Marketing AI Show, Paul takes the show on the road—to San Francisco for Jasper’s GenAI Conference—while Mike is here in Cleveland. The big news is Bard, Bing, and a $6 Billion valuation. Suddenly, it’s ChatGPT against the world.
Google responds to ChatGPT with its conversational AI tool, Bard.
Google just announced an experimental conversational AI tool named Bard. Bard uses Google’s LaMDA language model to provide natural language answers to search queries. Think of it like ChatGPT, but backed by all the knowledge and information that Google’s search engine has cataloged over the last couple of decades.
The announcement of Bard—a response to OpenAI and ChatGPT—prompted some critics to say the rollout was rushed, while others said they moved too slowly after ChatGPT took center stage in December and January.
If you missed it, the demo didn’t quite go as planned.
OpenAI gives Bing a new lease on life.
Microsoft’s Bing is getting more attention now than its previous 14 years combined. The latest version of the search engine is powered by OpenAI, complete with ChatGPT-like conversational capabilities. Bing can now respond to searches and queries in natural language, like ChatGPT, and use up-to-date information, like Google’s Bard release.
Kevin Roose, technology writer at the New York Times, took the new capabilities for a test drive and was impressed.
Will Bing and OpenAI make Edge, Microsoft’s browser, interesting for customers?
Cohere answers the call for ChatGPT for the enterprise.
Major AI startup, Cohere, is in talks to raise money at a $6 billion valuation and bring ChatGPT-like capabilities to businesses. Established in 2019 by former researchers at Alphabet/Google, Cohere is a big player in the world of AI. The foundational language AI technology allows businesses to incorporate large language models into their work.
The group is now in talks to raise hundreds of millions at a $6 billion valuation, reports Reuters, as the AI arms race heats up. Cohere is no stranger to the VC world, having already raised $170 million from venture capital funds and AI leaders like Geoff Hinton and Fei-Fei Li.
The appeal is the company’s focus on building for the enterprise, with an emphasis on real-world applications for their technology.
Listen to this week’s episode on your favorite podcast player, and be sure to explore the links below for more thoughts and perspectives on these important topics.
00:04:05 — Google announces Bard
00:20:28 — Microsoft’s Bing and OpenAI
00:27:05 — Cohere, fundraising, and $6B valuation
Links referenced in the show
Watch the Video
Read the Interview Transcription
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: I just feel like three, six months from now, we’re not talking about search engines here. I think we’re talking. Much bigger shifts in business communications, creativity, and the search stuff is just gonna have been like a footnote basically.
[00:00:15] Paul Roetzer: Welcome to the Marketing AI Show, the podcast that helps your business grow smarter by making artificial intelligence approachable and actionable. You’ll hear from top authors, entrepreneurs, researchers, and executives as they share case studies, strategies, and technologies that have the power to transform your business and your career.
[00:00:35] Paul Roetzer: My name is Paul Roetzer. I’m the founder of Marketing AI Institute, and I’m your host.
[00:00:43] Paul Roetzer: All right. Welcome to episode 34 of the Marketing AI Show. I’m your host, Paul Roetzer, along with my co-host Mike Kaput. What is happening with you this week, Mike? We’re, what is it? It’s Monday the 13th. We’re recording this. So, um, I am in San Francisco, actually, I’m on location. First time in San Fran in four years.
[00:01:03] Paul Roetzer: You’re back. You were traveling recently, weren’t you? Yes,
[00:01:06] Mike Kaput: DC Yeah,
[00:01:07] Paul Roetzer: I was in. Okay. Yeah, it’s, so, it’s funny, Mike and I see you like two, three times a week in the office. Sometimes like things are so crazy catch up. So I’m actually in San Francisco for the Gen AI Summit Conference. I’m not sure what the official name is, but Jasper’s event that’s happening actually on Valentines Day, February 14th.
[00:01:27] Paul Roetzer: So I’m, uh, anxious to see all, all the things happen in the world. Generative AI and language models tomorrow, so it should be, should have some stuff to report next week on the podcast, hopefully. That’s awesome. Yeah. So.
[00:01:52] Paul Roetzer: Huntington Convention, uh, center and yeah, Huntington Convention Center right across from the Rock and Roll Hall of Fame. If you haven’t been there. Um, brown Stadium, rock and Roll of Fame Science Center. It’s an awesome location in downtown Cleveland. So the conference brings together hundreds of professionals.
[00:02:06] Paul Roetzer: Uh, hopefully, I think we’re talking about three to 400 this year is the goal for the conference. So the focus this year is gonna be on experiencing the AI technologies, being able to engage with other forward thinking marketers and business. And really just exploring all the use cases and potential for artificial intelligence in your business and in your career.
[00:02:25] Paul Roetzer: Uh, the event will have talks from experts on generative AI ethics and legal human impact of AI is a big one for me this year. I’m really looking at building out the agenda to focus on some of those areas, including. Like workforce and talent development and reskilling and upskilling, things like that.
[00:02:41] Paul Roetzer: So we’re, we’re gonna go beyond just the use cases and the technologies and case studies and really get into some of the bigger issues around AI that, you know, you and I are focused on and talk a lot about. So, just a great way to get prepared for the next phase of your AI journey. Clear vision on a near term strategy to implement it in your business.
[00:03:00] Paul Roetzer: You can learn more at Macon ai. That’s MN ai. And the lowest rates actually end on Friday, the 17th. If you’re listening to this right after it comes out, uh, Friday the 17th, the rates go up at the end of the day. So I think it’s at 6 99 right now, which is the cheapest it’s gonna be. So you wanna come to Cleveland and be with us July 26th, Toth, definitely check that out.
[00:03:24] Paul Roetzer: I’m gonna turn over to Mike for our. Three topics. Again, if you’re new to the show, we’ve get been getting hundreds of new listeners every week, which has been awesome. Um, and, and the way we do this weekly is Mike and I just kind of pick three topics to riff about, and then we try and focus on just big things that’s happening.
[00:03:41] Paul Roetzer: The challenge we’re facing is it seems like we can pick three every day, so narrowing it down to three every week is . Kind of complicated, but we’ve got three, uh, good ones for today. So Mike, I’ll turn it over. Sounds great,
[00:03:54] Mike Kaput: Paul. Thanks, and I’m glad you had safe travels to San Fran. Um, yeah, definitely a, uh, hacked week of AI news.
[00:04:04] Mike Kaput: Um, first up, Google made some headlines, both, uh, in a good and a bad way, um, by announcing an experimental conversational AI tool that they’re calling Bard. And Bard uses Google’s Lambda language model to provide natural language answers. To search queries. So think of it like chat, G P T, but backed by all the knowledge and information that Google’s search engine has cataloged over the last couple decades.
[00:04:33] Mike Kaput: Now, this is still being tested by a group of what Google calls trusted testers who can’t yet. Use this as of, uh, today. Uh, but it should be rolling out in some fashion soon. Um, clearly this announcement is a response to OpenAI and Chat C B T and like I alluded to, there’s been a little controversy around it.
[00:04:54] Mike Kaput: Um, Google has been criticized both for a rushed rollout of the product, but as well as not moving fast enough to innovate in response to open AI and other competitors. So Paul, what did you make of the announcement when you
[00:05:10] Paul Roetzer: first heard of it? First, I, I laugh because if I’m not mistaken, when you and I recorded our episode last Monday, We said Google’s probably gonna make the news this week.
[00:05:23] Paul Roetzer: We thought it was gonna be on February 8th, and then they rushed the, the announcement, I think it was within an hour or two of us recording our episode. . Yeah. And so I just thought it was hilarious that it happened. And then Microsoft came out the next day with their demo of being an Edge, which I know we’re gonna talk about.
[00:05:37] Paul Roetzer: Then Google on Wednesday the eighth, I think it was, actually did the formal demo, which did not go well. So I, my initial thought was it’s the first time in my life, I guess, in Google’s life where they appear, uh, uncharacteristically vulnerable and unprepared. So from a PR perspective and people who don’t know my background, I actually was a public relations major coming outta the journalism school at Ohio University, and I spent a good portion of my early career working in pr.
[00:06:12] Paul Roetzer: As a former PR guy, it’s a disaster. Like it’s just, uh, setting their tech aside, which I still believe is probably the most advanced in the world. They just look like they have no clue what they’re doing. Hmm. And so that to me is like, again, I’m trying to separate the, the tech and the experience from the actual perception.
[00:06:32] Paul Roetzer: The reception is not good at all, and. I know that they got like the, um, the day of the demo, they lost like a hundred billion in market cap because the demo didn’t go well and Bard actually made an error that ended up being cause of a misplaced comma , where it was like learning the information from. But um, you know, I thought that just overall it looked bad and then I thought, It also is just representative of how little the market comprehends this stuff.
[00:07:03] Paul Roetzer: Hmm. Because they showed a demo of a large language model interface, which is a flawed architecture. Like large language models aren’t meant to be perfect. They’re making predictions about words and probabilities and things and like lose a hundred billion in market cap cause it got something wrong. It’s going to get something wrong all the time.
[00:07:24] Paul Roetzer: Like it’s just the nature of. Language models. So I don’t know. It’s kinda like just this crazy microcosm of where we’re at right now where Google and Microsoft are racing against each other. People are releasing things that aren’t ready for primetime. The market overreacts to everything that happens because they don’t understand the technology really at the core of all this, they don’t realize all the other stuff that Google’s got going on.
[00:07:48] Paul Roetzer: And it’s just like sometimes you sit back and think, this is just crazy. It’s a bizarre timeline we’re living in at the moment. .
[00:07:55] Mike Kaput: So that kind of raises this question that it seems like everyone’s been asking. There’s obviously sky high expectations, there’s unrealistic expectations, but would you say that Google is behind on innovating in this space?
[00:08:10] Mike Kaput: I mean, we’ve talked a lot about them being essentially the, the creators. Much of the technology that is being used in these products, did they truly get caught flatfooted or is this just crazy expectations and maybe rushing out the door?
[00:08:25] Paul Roetzer: I don’t think they’re behind on the technology. I think that would be a, a misguided assumption if you don’t think that Google has more advanced tech than what we’re seeing.
[00:08:35] Paul Roetzer: They said as much that they’re basically connecting Bard to a light version of Lambda, which is their main language model. And because if they did it to the full language model, not only would it require massive amounts of computing power mm-hmm. , um, but it, it, the risks would be too. So they’re trying to kind of play the game, get something out into the market, but realistically, they weren’t ready to release a product at this point.
[00:09:02] Paul Roetzer: And again, it’s not that they don’t have more advanced technology, it’s that they just weren’t prepared to have to make the moves that they’re making right now because of what Open Microsoft did. And that’s becoming very apparent at the moment. So,
[00:09:20] Mike Kaput: This is obviously just the beginning of this technology, but we’ve already seen lots of tweets, thought leadership, think pieces on how these types of conversational interfaces and using large language models could impact overall the search experience.
[00:09:39] Mike Kaput: And I’m curious about your thoughts on overall search and what this could mean we’re headed towards. And also, do we have any thoughts on how. The impact on search could also impact seo, which obviously is marketers, is a big topic
[00:09:56] Paul Roetzer: us. You know, it’s interesting, I watched, uh, at least clips of the interview with Sat Satya Nadela, the CEO of Microsoft, and they asked about this question specifically, like the impact on publishers and links to sources.
[00:10:10] Paul Roetzer: And I think I saw a clip maybe where they even asked Sam Altman, the CEO of OpenAI about it. And I believe Sundar from, uh, Google actually addressed this as well. And right now it seems like they’re basically all taking the same position of, we don’t really know, like, that everyone’s moving so fast to get this out, that they’re almost on the fly trying to figure out how they’re going to still reward publishers who are creating all the content that’s training all the models.
[00:10:40] Paul Roetzer: So again, if you’re, if you’re new to this, the, the crux of the issue. If I go in and I ask Google a question, it can give me back a snippet. Like it can give me a summary today, but generally it’s still driving me to publisher links. I’m still going to a third party site to actually gather more information.
[00:11:02] Paul Roetzer: The premise of chat, g p t and what Bing is basically enabling is that I might not have to go to those sources anymore, that if it just writes me an answer and I’m happy with the answer, I don’t need to go on to other sources, other, you know, citations. Um, and so that’s creating a lot of uncertainty in marketing and search and in SEO is like, what is the value of published content as a brand, as a media company, as an individual.
[00:11:30] Paul Roetzer: Uh, The search engines are gonna stop sending traffic to our site, so, I’m not ready to say like the sky is falling in here. I think there’s a lot to be determined in the months ahead. I don’t even say years at this point. I think a lot of this is gonna start playing out in the next, you know, three to 12 months.
[00:11:48] Paul Roetzer: Um, but there is a lot of uncertainty and I think people are asking a lot of really good questions. Um, but I don’t think that Microsoft and Google actually have very good answers, right. Or at least answers that people want to hear. I did hear, um, was it Chamath? I think you and I both listened to the All In podcast.
[00:12:08] Paul Roetzer: Mm-hmm. , I think he had, he had proposed in this week’s episode, like Google just doubles what they’re paying publishers. Like basically Google just says, forget it. Like we’re just going to give, give away the money. Like it’s worth more to us to save. Than it is to, you know, pinch margins. And so he had a relatively, it was the best idea I’ve actually heard of.
[00:12:30] Paul Roetzer: Like, yeah, if Google like, don’t underestimate what Google’s capable of doing, they’re not going to just let their search business get cannibalized. So there’s some plays they could make that the average business person to market are probably isn’t thinking about because we don’t spend our days analyzing Google’s business model.
[00:12:47] Paul Roetzer: Mm-hmm. So I, the one other thing I wanted to like touch on here, there’s a few like tan tangential. Pieces of content or there were things related that I wanted to kinda get into for a moment. Um, and that is just the understanding of how language models work. Cause I think there’s still lots of misconceptions here.
[00:13:08] Paul Roetzer: So when we’re talking about Google and we’re talking about Bard and we’re gonna get into Microsoft and being chat pt, I think it’s a really, really important that people understand the fundamentals of language models and that they have inherent flaws. So these language models, which power barred and chat, G P T and G PT three, like.
[00:13:26] Paul Roetzer: What it’s it’s language model. They’re just making predictions about words, and I get questions about this every day on LinkedIn or in person. And so people are always like, well, why can’t it cite its sources? It doesn’t have sources like it learns from a corpus of knowledge, in this case, the internet, and then it predicts the words in a sentence.
[00:13:44] Paul Roetzer: It’s not drawing from any specific source. Like the, the analogy I’ve given is like if you were, or I wanted to write a story about what a large language model is like we wanted to write an article about it and let’s say we’re gonna spend a week and we’re gonna go read the hundred best articles about large language models.
[00:14:02] Paul Roetzer: And then we sit down and we write a 500 word blog post about what is a large language model and how do they work. There’s a decent chance we could then write that without a single. Like we, we went and consumed a bunch of knowledge. We synthesized it in our minds, and then we sat down and we wrote something.
[00:14:21] Paul Roetzer: So if you then said to me, well, Paul, where’s the citations? I would say, I don’t know. Like I, I just read a hundred sources on it, like I learned the topic so I could write about it. I can’t cite for you every single factor line within that, that story. That’s basically a simplified version of how a large language model.
[00:14:41] Paul Roetzer: It goes out, it consumes knowledge, the internet in this case, and then it synthesizes that and it writes an article or a response to you. So to ask it for citations, you then actually have to build a whole nother layer of artificial intelligence into it that then goes out and figures out what sources influenced that response.
[00:15:04] Paul Roetzer: So when Bing is actually like giving you a response and then it’s giving like some recommended reading. All it’s actually doing is connecting the language model to the internet. The language model gives you a response. Then an AI underneath it goes and finds related sources that seem to most heavily support the output of the language model.
[00:15:28] Paul Roetzer: So again, it’s like there’s this fundamental thing that needs to happen. Where we’re at with these language models is very baseline. There’s this great article from Rob Toes or TAs, I have to find out how to actually say his last name cause he writes amazing stuff. Um, But he wrote in Forbes about like, what’s gonna come next for language models.
[00:15:47] Paul Roetzer: And so I kind of, I put this on LinkedIn, I said basically, smarter, faster, more accurate, more specialized, and it will appear to be more humanlike. And so I’ll just take a moment and kinda outline the three things he says. And then I recommend people go read this. So the first he said is the models are gonna start to generate their own training data.
[00:16:05] Paul Roetzer: So as crazy as this sounds, These models can consume all of human written knowledge, and then they’re gonna run out of data to consume. So they can conceivably, if they haven’t already consume everything that exists on the internet. And as new information becomes available, they can consume all that too.
[00:16:22] Paul Roetzer: So what he’s proposing is that these language models will actually consume all the information. Start creating original content and then learn from their own content. So they’re gonna start creating content at scale to learn from new ideas and content. So that’s crazy, but it’s, it’s in the works. The second is that they can fact check themselves.
[00:16:42] Paul Roetzer: That one’s actually more simple. That’s the analogy of go look on the, like write something, go. Sources that confirm it and then come back. The other thing we just saw from, I think it was meta a couple days ago, released a paper where the AI is actually building tools into the language model. So when it gets to numbers, it actually has a calculator tool that it launches and runs.
[00:17:03] Paul Roetzer: When it gets to dates, it run, it runs a calendar tool. Um, so basically they’re building tools into the language models that can then verify the information. And the third and maybe the most intriguing to. Is this idea of what he calls massive sparse expert models. So if you’ve been paying attention to language model space, you hear parameters.
[00:17:23] Paul Roetzer: So the size of a language model is basically dependent upon how many parameters are in the model. So how big It’s, without getting overly technical, so GPT three has 5 billion parameters. If I, if I’m getting the numbers correctly, what happens right now if I run a search or a query into G P T three, it uses all 175 billion parameters every.
[00:17:48] Paul Roetzer: So think about the energy consumption required to get an output from it. Now, think about how the human brain works. When we need to make, like say we’re driving, we have to make a decision to stop or to, to go through a red light when, or a yellow light. Like, okay, yellow light, am I gonna accelerate or am I gonna stop?
[00:18:04] Paul Roetzer: You don’t pull from every neuron in your brain to decide whether or not to accelerate your car. There’s like some small piece of your brain that actually tells you instinctually whether you’re gonna go or not. So when the human brain does what it does, it’s actually only using very small percentages of the power.
[00:18:22] Paul Roetzer: It’s how it does everything so efficiently. And the belief is, and Google DeepMind’s team has proven this is possible, that you can actually have specialized segments of the model that fire based on what the prompt is, based on what the needed output is. So rather than having to continually build these much, much larger models through say, like a trillion parameters, there’s been this rumor like P PT four is gonna be a trillion parameters.
[00:18:45] Paul Roetzer: It might not need to. If they find a way to segment those parameters to be specialized for what is being asked of it, and that’s a major potential breakthrough, like that could really unlock a lot of potential with language models. So I, I just, again, I wanted to say that the other thing I’ll end with here is, um, Greg Brockman, the c e o of, or the, the president and co-founder of OpenAI tweeted on Sunday, the Sunday morning.
[00:19:14] Paul Roetzer: Most amazing fact about AI is that even though it’s starting to feel impressive, so again, experience. A year from now, we’ll look back fondly on the AI that exists today as quaint and antiquated equal cause for excitement and deliberative caution, important to get the tech and its deployment. Right? So the basic takeaway I want to again leave with here is, What we’re seeing from Google, from Bard, the overreaction to getting a single thing wrong in a demo.
[00:19:45] Paul Roetzer: It is so minuscule as to where language models are going and where AI is going, and that’s why I just kind of like laugh when this overreaction occurs over a single thing. Or like people jump in and be like, oh, I don. You know, Bing or I don’t like Bard and it’s just never gonna work or replace humans or, you know, have a meaningful impact on it.
[00:20:05] Paul Roetzer: And I just think it’s, it’s silly to like think. So thanks for letting me just like, jump on a soapbox for 10 minutes. , talk about language.
[00:20:16] Mike Kaput: No, it’s, it’s such an important topic because as we’re gonna just dive into now, I mean, Google’s not the only one, um, in this space. I mean, part of what we’ve been discussing is Microsoft.
[00:20:28] Mike Kaput: Microsoft, literally the same, in the same week, launched a version of its Bing search engine powered by OpenAI. So by, uh, Greg’s company that you just, you quoted mm-hmm. , um mm-hmm. . And so Bing will now have chat G P T, like conversational capabilities, so it can now respond to searches and queries in natural language like chat, J p T, and also use up-to-date information like Google’s Bard product.
[00:20:54] Mike Kaput: So, Kevin Russ, a big tech writer at the New York Times. He took the new capabilities for a test drive and reported being extremely impressed. And we now are hearing that Microsoft is also going to incorporate the technology into its edge browser. And I don’t know the exact numbers, but I think you saw crazy rises in the app store for the browser, uh, rises in the usage of Bing as a result.
[00:21:21] Mike Kaput: So, Paul, what do you kind of, on the heels of Google, how do you view this announcement?
[00:21:28] Paul Roetzer: So if we go through the same lens as Google, they’re, the Microsoft PR team is winning. They, they, they won last week. Uh, Microsoft certainly came out of last week looking way better. Google. Um, so kudos to them on their positioning.
[00:21:42] Paul Roetzer: I think anyone who’s watched interviews be Satya, it is almost hilarious how much he is reveling in this moment. , he is very much enjoying the fact that they are forcing Google’s hand here and he’s almost taunting them. And I think we talked about this a little bit last week. Um, I think he’s won. He said something.
[00:22:03] Paul Roetzer: They had to dance and like were the ones that made him do it, or something to that effect of like, he, he’s, he’s not shy about the fact that they’re happy that they’re forcing Google’s hand here. Uh, okay. So that being said, I am, I’m a fan of Microsoft. Um, we don’t use Microsoft generally speaking. I mean, uh, we, we use primarily Google products as an institute.
[00:22:28] Paul Roetzer: My agency previously, I am on the wait list for Bing the, the new version of Bing. And I will say that when I went to join the wait list, rather than just asking for my email, I had to log into a Microsoft account I didn’t know I had, so it made me register for Microsoft, which then. Made me realize how much I hate having to navigate Microsoft’s accounts because when my children wanted to play Minecraft, I had to create Microsoft’s accounts.
[00:22:58] Paul Roetzer: Then to actually manage anything in Microsoft, do you have to navigate these crazy web of like, which account am I in? Which, what preference pages am I in? Can I do it in the app? Oh no, I gotta go to the Microsoft site. And it reminded me why m. Has been in the position they’re in all this time, there is a friction filled process to work with Microsoft products, in my opinion.
[00:23:23] Paul Roetzer: Mm. Whether it’s been Skype or Minecraft or Microsoft products. Again, I didn’t even know I had an account. And then I went, I was like, well, I don’t even know where this account goes to. And then it was actually verifying on an old email address and I have no idea how to even change that. And I was like, oh my God.
[00:23:39] Paul Roetzer: Like, and in that moment, and again, this is just one person’s, uh, subjective opinion. I don’t think at the end of the day, Microsoft makes a meaningful impact on market share here because the friction to move off of the way I or other people do things is so great. Like for me to test being out and play around with it.
[00:24:04] Paul Roetzer: No problem. I’ll do it. I’ll check it out. But chances are either Google’s gonna have a better version in three months, or go here built into Google or Neva uh u.com, like take your pick. And so I think it’s a really interesting moment for Microsoft. I’m happy that they’re having this, this success. I wouldn’t bet against open AI and their relationship.
[00:24:28] Paul Roetzer: But Microsoft has a lot of legacy perceptions about how challenging it is to work. Their platform, at least for me. Mm. That I cannot see myself at the this moment doing what, like Kevin in the New York Times said, like, bings down my default browser, right? Um, or edge, my new default browser bings the default search.
[00:24:48] Paul Roetzer: I, I don’t see that happening, um, in any significant way. I, I could see maybe market share shifts a few percentage points, and that’s not an insignificant amount of money, but at the end, I, I think. I think Satya has a larger strategy at play here than just search. So again, I’m not saying I’m, I’m keeping these comments specific to like search in particular.
[00:25:14] Paul Roetzer: Mm-hmm. . Cause that’s what we’re talking about. But I think there’s some maybe a Trojan horse here that, that they’re doing this for a reason. And winning search isn’t actually the reason they’re doing this, but it’s a really interesting way to force Google’s hand in the process. Um, so yeah, I don’t know.
[00:25:36] Paul Roetzer: Again, I’m just kinda like ripping here, but that was my, I I’m happy. It looks impressive. People seem really impressed by it. I haven’t had a chance to test it yet, um, because I’m on the wait list through whatever account. I finally got logged in through, and I, at the end of the day, uh, I, I don’t think. I don’t think we’re gonna see a meaningful shift in, in metrics, um, from a search perspective.
[00:26:02] Mike Kaput: Yeah, and like we’ve discussed on previous podcasts, search is really, you know, it’s easy to get distracted by kind of the big headlines and and shiny objects here. I mean, search is just one piece of the puzzle. I mean, Microsoft, for better or for worse, and I’m sure their user experience is not great on some of the other productivity apps they have, but they’re embedded in so much of corporate America.
[00:26:24] Mike Kaput: Yeah. And they’re rolling out these types of capabilities. In other tools as well, like teams
[00:26:29] Paul Roetzer: and office. Right. And those people aren’t switching. Like that’s the thing is if you’re gonna have enterprise users, I’m sorry, like Microsoft people aren’t switching to Google for a better search experience and Google people aren’t switching to Microsoft for a better search experience.
[00:26:40] Paul Roetzer: It’s just like, I’m not saying it’s trivial cause it’s majority of Google’s business ad revenue, revenue from it. But I, I just feel like three, six months from now, we’re not talking about search engines here. I think we’re talking. Much bigger shifts in business communications, creativity, and the search stuff is just gonna have been like a footnote basically.
[00:27:04] Mike Kaput: Yeah, that makes sense. And it’s so crazy to me as well, with kind of our third topic here, just we’re so busy talking sometimes about the big tech giants that, I mean, there’s so many other major players in this space, and one of ’em. That isn’t as well talked about, but is extremely well known. Um, in AI Circles is a company called Cohere, and they’re a major AI startup that is in talks to raise money, uh, an undisclosed amount right now at a 6 billion valuation and bring chat G p t like capabilities to businesses.
[00:27:39] Mike Kaput: So here’s been around for several years, established in 2019 by former researchers at Alphabet, Google’s parent company. And they build foundational language, AI technology that allows businesses to incorporate some of these large language models into their work and. CO here has already raised 170 million, uh, from big venture capital funds, but also notably, they are supported by some of the big luminaries in the AI space, including Jeff Hinton, who is one of the kind of godfathers of the field.
[00:28:14] Mike Kaput: And Fafe leaves also a major, major player. Um, what I thought was really notable about this announcement, not. The valuation itself, which is significant. But the company also has a stated focus for building for the Enter enterprise with an emphasis on actually real world applications for these type of conversational and language technologies.
[00:28:38] Mike Kaput: So CO here is less talked about than some of the other heavyweights in this space, but they’re worth paying attention to. Can you kind of walk us through why
[00:28:48] Paul Roetzer: that. Yeah, I think you highlighted a, a good portion of it. I started paying attention to them right away because Jeff Hinton name was attached to, as was as you said in the Series A.
[00:29:01] Paul Roetzer: So they raised, um, they did 40 millions Series A in 2001, and then they did 5 million series b2. So Jeff Hinton, I, we’ve talked about Hinton on the sh the show before, but basically he was one of kind of the godfathers of modern ai or the deep learning movement and coined the phrase deep learning. And then him and a team of PhD students, um, in 2011, 2012, uh, proved the ability of neural nets and deep learning to do advanced image recognition with the ImageNet competition.
[00:29:40] Paul Roetzer: He then formed a company, sold it to Google for 44 million. Went on to kind of work in Google Brain team there. So Hinton is a legend and so when I first heard of Cohere years ago and Hinton and Fafe Lee being associated with it, I was like, well, they’re not gonna be associated with anything that’s not super legit.
[00:29:59] Paul Roetzer: And I didn’t really know the team behind cohere or anything like that, but I just knew Hinton’s involved. It’s gotta be something we’re paying attention to. So I started following along with coherent and, and kind of making notes and subscribe to their content and just been keeping track of ’em. But then when you realize who Aiden Gomez is, so the c e o and one of the co-founders, and you start doing some back research on him, he was one of, I think there was nine authors on Google, um, in 2017 of the attention is all You need paper where they introduced the transformer, which is the architecture that underlies language models today.
[00:30:33] Paul Roetzer: So what, what powers, G P T and Chad, g p t, and all these things spared Lambda. This breakthrough called Transformers is what actually enabled this to occur. And Aiden Gomez, I believe was the lead author on that paper. And Hinton was one of his advisors, I think is the way it worked out. So Gomez and I, I believe like seven of the nine people from that team actually all left Google in recent years to do their own thing.
[00:30:59] Paul Roetzer: So Gomez went, took a lot of the learnings from what they were doing there and built Co here to build a language model from the ground up basically. So then you. Kind of connecting the dots. It’s like, okay, so Transformers, I was just re-listening On the flight to San Francisco, I was re-listening to the Lex Friedman interview with Andres Kapai.
[00:31:18] Paul Roetzer: Mm-hmm. Who was the head of AI at Tesla. And just announced last week, he’s going back to open ai. Yeah. In that interview, Friedman said like, what is the most, uh, monumental thing basically that has happened in AI in recent years, in your opinion? And he said the creation of the transformer. Hmm. So, I mean, it is a extremely important moment in the development of AI and the advancement of what we’re seeing in generative AI today is the creation of this transformer architecture.
[00:31:47] Paul Roetzer: And so to know that Aiden Gomez was actually at the forefront of that and now building this company, those things alone make it worth following. And then for me, in, in kind of paying attention to what they’ve been. They’ve done some very interesting things from a marketing and branding perspective to try and educate people on how language models work and what they are.
[00:32:08] Paul Roetzer: Mm-hmm. and the different ways to use them in marketing use cases like content summarization is one of the big ones. Um, so I don’t know. I’ve just been following them. I’ve played, they, you can get access to their playground, so just like open a AI has a playground that you can get access to and test their language models.
[00:32:23] Paul Roetzer: You can do the same thing with cohere. So I’ve been doing that I think probably six months ago or so, I got access to that. It’s not as user friendly, like I can definitely see it’s built more for developers. Um, the results certainly weren’t as impressive as like playing around with chat G P T or g p T three in open AI playground.
[00:32:41] Paul Roetzer: But the company and where they’re going to me is very fascinating and I believe they just announced, um, I know one, one, interesting. They’re agnostic and cloud, so they’re not like open AI where they’re gonna do a deal only with Microsoft. So I believe they’re already embedded with AWS SageMaker. I think you can actually use Coherent in Amazon SageMaker, and then I think they also have a deal with Google Cloud, and that article said it was rumored that Google’s actually gonna be one of the people investing in the next round.
[00:33:10] Paul Roetzer: Mm-hmm. So, yeah, long story. I would pay attention to what CO here is doing. I think it’s a really smart company. Obviously they have a density of talent there that seems to be getting better and better, and I think they’re going to be a, A name. People are gonna hear over and over again in the months and years ahead.
[00:33:29] Paul Roetzer: Yeah,
[00:33:29] Mike Kaput: that’s a, just a crazy amount of talent and pedigree from a company that most people probably outside this space probably haven’t necessarily heard of or been following. Yeah. Um, how do they kind of fit into the AI competitive landscape, so to speak, in your opinion?
[00:33:46] Paul Roetzer: Well, they’re, they’re, so you right now, you have the companies that are building on top of language models like.
[00:33:55] Paul Roetzer: Again, I’m just gonna generalize here, but let’s say like writer Jasper Hyper, right? Word, tune. Um, for the most part, they build on top of the APIs from people building the language models like OpenAI Co. Here would be a builder of language models. So they’re building their own, they’re not building on top of someone else’s api.
[00:34:13] Paul Roetzer: Now I, again, I say generally because these AI writing tool companies are. Trying to get into the building their own language model game, because they don’t want it to be dependent upon any one API to do what they’re doing. Um, but yeah, I would say they’re, they’re up there with like open ai anthro, I think we talked about last week that just took a big investment from Google’s another player there.
[00:34:34] Paul Roetzer: Um, and then you have like stability, AI, and hugging face that are kinda opensourcing these models so people can fine tune ’em and build on top of them. Uh, but yeah, there’s, I mean, there’s probably. I don’t know, eight to 10 major noteworthy companies right now in this space at different stages. And I think they’re gonna become very common names in the business world in the coming, you know, year or two.
[00:34:58] Paul Roetzer: Yeah.
[00:34:58] Mike Kaput: So speaking of that, and kind of to wrap this topic up in our episode up, I wanted to end on getting your take about co. Here’s focus on building for the enterprise. And I mentioned this multiple times in interviews. The CEO is pretty adamant that, hey, like. A lot of the things we’re seeing out there in the usage of language models is great, but we’re laser focused on real world use cases and building for business
[00:35:22] Paul Roetzer: applications.
[00:35:25] Paul Roetzer: I think we’ve talked about this before, if not I’ve, I’ve said it in talks, but I believe that the near future for enterprises is going to be custom training models on their data. So the more. Blog posts, webinars, eBooks, white papers, videos, podcasts. If you think about the archive of all the content, you have to take a general model like G P T three and then train it on your corpus of knowledge specific to your organization.
[00:35:58] Paul Roetzer: That’s really valuable. Connected to your CRM database. Now you have customer data infused into there. So the more proprietary data sources you can layer into these language models, multimodal two. So audio, video, images, text. If you think about your organization and all of that content you have and the ability to have that also inform the outputs of these language models, I think.
[00:36:24] Paul Roetzer: Critical. Uh, and it’s where most of them are gonna try and go. So it seems like CO here is just going there first. They’re not gonna get distracted by trying to build like cute little chat interfaces that don’t make much money. And, you know, they’re fun and they’re useful, but they’re flawed. I think they’re just gonna say, we’re just gonna bypass that whole thing and let’s just go right to the value creation at the, at the top level.
[00:36:49] Paul Roetzer: It’s an interesting play. I, I would. I would invest if I had the opportunity. I’ll say that . I, I think they’re, I think they’re a major player and they’re doing very smart things right now. Very cool. Well,
[00:37:03] Mike Kaput: Paul, as always, thank you for your time and insights. Um, enjoy the rest of San Francisco .
[00:37:09] Paul Roetzer: I will do one, one more note before we go a bonus for everyone.
[00:37:13] Paul Roetzer: We’re not doing a rapid fire session today, , however. We can link to this. Google did for the first time state they will not penalize AI content just because it was generated by ai. Yes. So that happens since last week’s episode, it’s been one of the big question marks around SEO and language models and content creation.
[00:37:32] Paul Roetzer: So they said that it’s still all about, is it helpful? Is it valuable? And it won’t matter whether or not AI wrote it. So just. Throw it out there as our one rapid player item for the week. . .
[00:37:44] Mike Kaput: Yes, there’s, and that certainly, uh, is going in a direction that maybe not everyone in marketing predicted, right?
[00:37:51] Mike Kaput: Because there’s a lot of commentary that, oh, Google will penalize this stuff. So not exactly true.
[00:37:57] Paul Roetzer: It’s always nice when we have actual confirmation from the source and we all don’t have to just kinda sit around making educated guesses. , right? Yes. We can put this one to rest. It will not penalize. Good AI generated content, let’s say.
[00:38:13] Mike Kaput: Perfect. Which I’m sure we’ll see more of incoming days, .
[00:38:17] Paul Roetzer: All right. Well, next week should be interesting. I’ll be back from the Gen AI conference. I’ll have all the new information from Jasper and, uh, everything that’s gonna be talked about here. And I’m sure it’s gonna be another wild week in ai everywhere else.
[00:38:30] Paul Roetzer: So, uh, thank you to everyone for listening. Uh, definitely subscribe and leave a rating and we would love to not only see the ratings, but re reach out to Mike and I, we’d love to hear from the listeners and, uh, get connected. So thanks everyone. We’ll talk to you next week.
[00:38:44] Paul Roetzer: Thanks for listening to the Marketing AI Show. If you like what you heard, you can subscribe on your favorite podcast app, and if you’re ready to continue your learning, head over to marketing ai institute.com. Be sure to subscribe to our weekly newsletter, check out our free monthly webinars, and explore dozens of online courses and professional certifications.
[00:39:06] Paul Roetzer: Until next time, stay curious and explore ai.