Latest / Sidecar Sync / Google AI Overviews, DeepL’s Translation Revolution, and Personalizing with rex’s AI Engine | 33

Transcript

Google AI Overviews, DeepL’s Translation Revolution, and Personalizing with rex’s AI Engine | 33 · Sidecar Sync

• 0:00
  Speaker 1: I mean, it's getting better.
• 0:01
  Speaker 1: Right now, AI is kind of like an idiot savant, you know.
• 0:03
  Speaker 1: It's basically incredibly brilliant in some ways, but unbelievably stupid in other ways, almost comically so.
• 0:09
  Speaker 1: But that's changing by the minute.
• 0:11
  Speaker 1: Welcome to Sidecar Sync, your weekly dose of innovation.
• 0:17
  Speaker 1: If you're looking for the latest news, insights and developments in the association world, especially those driven by artificial intelligence, you're in the right place.
• 0:26
  Speaker 1: We cut through the noise to bring you the most relevant updates, with a keen focus on how AI and other emerging technologies are shaping the future.
• 0:34
  Speaker 1: No fluff, just facts and informed discussions.
• 0:37
  Speaker 1: I'm Amit Nagarajan, chairman of Blue Cypress, and I'm your host.
• 0:42
  Speaker 1: Welcome back everyone to another episode of the Sidecar Sync.
• 0:47
  Speaker 1: As always, it is a pleasure to be here and I'm excited to have a great conversation with you guys about all the exciting things happening in the world of AI at the intersection of the association universe.
• 0:58
  Speaker 1: My name is Amik Nagaraj and I'm one of your hosts.
• 1:01
  Speaker 2: And my name is Mallory Mejiaz.
• 1:02
  Speaker 2: I'm one of your co-hosts and I run Sidecar Baraj and I'm one of your hosts.
• 1:05
  Speaker 1: And my name is Mallory Mejiaz.
• 1:06
  Speaker 1: I'm one of your co-hosts and I run Sidecar.
• 1:07
  Speaker 1: It's great to be back.
• 1:07
  Speaker 1: Before we jump into all the exciting things we have to share with you this week, we're going to take a moment for a quick word from our sponsor.
• 1:16
  Speaker 2: Today's sponsor is Sidecar's AI Learning Hub.
• 1:18
  Speaker 2: The AI Learning Hub is your go-to place to sharpen your AI skills and ensure you're keeping up with the latest in the AI space.
• 1:25
  Speaker 2: When you purchase access to the AI Learning Hub, you get a library of on-demand AI lessons that are regularly updated to reflect what's new and the latest in the AI space.
• 1:36
  Speaker 2: You also get access to live weekly office hours with AI experts and, finally, you get to join a community of fellow AI enthusiasts who are just as excited about learning about this emerging technology as you are.
• 1:48
  Speaker 2: You can purchase 12-month access to the AI Learning Hub for $399.
• 1:54
  Speaker 2: And if you want to get more information on that, you can go to sidecarglobalcom.
• 1:58
  Speaker 2: Slash hub.
• 1:59
  Speaker 2: Amit, it's been an exciting few weeks.
• 2:03
  Speaker 2: How are you?
• 2:05
  Speaker 1: I'm doing great.
• 2:06
  Speaker 1: I'm keeping up with all the craziness that's going on, but your world's changed a lot in the last couple weeks too, right.
• 2:12
  Speaker 2: It has.
• 2:12
  Speaker 2: Indeed, I have exciting news for all our listeners.
• 2:15
  Speaker 2: It's the first time you're hearing me join you from Atlanta, which is the place that I now live technically Decatur, which is a suburb of Atlanta.
• 2:22
  Speaker 2: I moved last week right before attending MMC plus T, which is ASAE's Marketing Membership, communications plus Technology Conference, and I will say, amit not my brightest move to attend a conference while I was moving states, but you know there wasn't an AI to help for that, unfortunately.
• 2:41
  Speaker 1: I don't know, but you told me, I think, wasn't it true that your fiance did the move while you were at MMCT?
• 2:47
  Speaker 2: Okay, I helped pack up the U-Haul prior to MMC plus T, but yes, I flew out the morning that he drove to Atlanta, so he would be so happy to know that I'm telling you all the truth.
• 2:59
  Speaker 2: I didn't have to help on the back end and I'm very thankful for it.
• 3:05
  Speaker 1: I think that was a great move.
• 3:06
  Speaker 1: I bet ChatGPT helped you plan the whole thing out.
• 3:08
  Speaker 2: Absolutely, and I did get to meet a lot of Sidecar fans and new followers at MMCPlusT, so shout out to all of you new listeners who are joining us.
• 3:18
  Speaker 1: Fantastic.
• 3:20
  Speaker 2: Today we've got some exciting topics.
• 3:23
  Speaker 2: First and foremost, we're talking about Google Overviews.
• 3:25
  Speaker 2: This will be a fun conversation for sure.
• 3:28
  Speaker 2: Then we'll be talking about DeepL, an AI translation company, and finally we'll be talking about Rasaio's new product, rex.
• 3:36
  Speaker 2: So, first and foremost, google Overviews.
• 3:41
  Speaker 2: Google's AI Overviews, a feature introduced to enhance search results with AI-generated summaries, has faced significant scrutiny and criticism since its rollout.
• 3:50
  Speaker 2: Ai Overviews were initially introduced at Google's Developer Conference and are designed to provide users with a quick summary of information at the top of search results, along with links to external websites.
• 4:01
  Speaker 2: The feature uses Google's large language model, gemini, to synthesize information and handle more complex queries than traditional Google search.
• 4:09
  Speaker 2: Ai overviews have produced several erroneous and potentially harmful suggestions, such as advising users to consume rocks that's one that's gotten a lot of press just recently and mix glue into pizza sauce.
• 4:23
  Speaker 2: The AI's tendency to generate incorrect information stems from its reliance on statistical predictions rather than factual accuracy.
• 4:30
  Speaker 2: This has led to hallucinations where the AI fabricates information.
• 4:34
  Speaker 2: The inaccuracies have raised concerns about the erosion of trust in Google search and the potential negative impact on website traffic and ad revenue, as many users may rely on the AI summaries instead of visiting original sources.
• 4:48
  Speaker 2: Google has implemented over a dozen technical improvements to address these issues, including better detection of nonsensical queries and limiting the inclusion of user-generated satirical or humorous content.
• 5:00
  Speaker 2: Google's also added restrictions to prevent AI overviews from appearing for certain types of queries, particularly those involving health and hard news topics, where accuracy is crucial.
• 5:10
  Speaker 2: Despite these improvements, experts acknowledge that AI systems will always have limitations and the potential for errors, due to the nature of LLMs and the variability of online information.
• 5:22
  Speaker 2: Amit, what are your initial thoughts on Google AI overviews?
• 5:26
  Speaker 2: Did Google roll this out too quickly?
• 5:29
  Speaker 1: No, they rolled it out too slowly.
• 5:31
  Speaker 1: In my opinion, here's why Everyone else is doing this at a higher level of quality than they are, and the reason they're not there yet is because they haven't been in the game.
• 5:39
  Speaker 1: So Google is now playing ball.
• 5:41
  Speaker 1: Google is disrupting their bread and butter, their core revenue stream.
• 5:45
  Speaker 1: They don't really have it figured out how they're going to deal with ad revenue in the context of Google Overviews, because Google Overviews, if they are actually good at some point, will reduce traffic to partner sites, which is obviously terrible for SEO in terms of organic traffic to third-party websites outside of Google, but potentially has a negative effect on their traditional advertising structure and model.
• 6:08
  Speaker 1: But they're disrupting themselves in a way, so I think that's good.
• 6:12
  Speaker 1: I'm disappointed that Google isn't further along.
• 6:15
  Speaker 1: Frankly, I think that, with the resources they have and the kind of AI brilliance they have within Deepvine, they should be further along with this.
• 6:22
  Speaker 1: They're not where they should be and I think that's a function of not having moved quickly enough or focused enough on this for some time.
• 6:29
  Speaker 1: But I do think that they have resources and the technical capabilities to catch up.
• 6:35
  Speaker 1: So we'll see what happens, but I think one of the things that you need to think about as an association is think about why is Google doing this?
• 6:44
  Speaker 1: Google is not doing this out of the goodness of their hearts.
• 6:46
  Speaker 1: They're not doing it because they enjoy disrupting their primary revenue strength, their cash cow.
• 6:52
  Speaker 1: They're doing it because they clearly see that consumer demand is heading away from traditional search and towards generative AI.
• 6:59
  Speaker 1: In spite of all of its warrants at the moment which it has, particularly their framework, ai has more than others at the moment.
• 7:06
  Speaker 1: So they are doing this because they see this as the future of online engagement.
• 7:11
  Speaker 1: That generative AI is so much more effective at solving problems than traditional search.
• 7:18
  Speaker 1: Traditional search still has a place in the world.
• 7:21
  Speaker 1: I think it's great for research.
• 7:24
  Speaker 1: It's great for a lot of things where you need to go deep and you want to be able to kind of have fine-grained control over the exact sources you're going to solve for in terms of like really digging into all the different things that match certain queries and so forth.
• 7:36
  Speaker 1: I think search will always have a place, but most of the people, most of the time, are doing pretty casual things with Google.
• 7:43
  Speaker 1: They just want the solution, they want the answer quickly, and generative AI is so much lower friction and higher fidelity in terms of the output, and that's the reason ChatGP2 took off like a rocket back in late 2022, because it actually solved the problem no one else was solving.
• 7:58
  Speaker 1: Google sees this clear as day and they're moving to change this.
• 8:02
  Speaker 1: So, again, why is this relevant for your associations?
• 8:05
  Speaker 1: A lot of associations just now are starting to think about implementing a Google style search on our website, and I keep advising people that I talk to that you might want to think about that again and pause a federated search tech project and think about a generative AI solution for your website instead, because that's clearly where consumer demand is headed of.
• 8:25
  Speaker 2: AI solution for your website instead, because that's clearly where consumer demand is heading.
• 8:28
  Speaker 2: So, in your words, then, you don't think search engine optimization is dead.
• 8:33
  Speaker 1: You think there will always be a place for traditional search.
• 8:34
  Speaker 1: I think there will always be a place for it for people who want to go deep.
• 8:39
  Speaker 1: I think that the casual user most of the time let's say 80, 90% of people, 80, 90% of the time, which is the vast majority of traffic will not need the depth that comes from their own research using a search tool.
• 8:54
  Speaker 1: They'll be very satisfied, very well served through a generative AI solution, especially when you think about tools like Perplexity, which is kind of like a synthesis between generative AI and search coming together.
• 9:09
  Speaker 1: Much more reliable combination of what language models do well, but then citations are abundant.
• 9:15
  Speaker 1: You can actually see where the content is coming from.
• 9:17
  Speaker 1: It's very clear.
• 9:18
  Speaker 1: Openai is moving in that direction.
• 9:20
  Speaker 1: I think Perplexity is probably the best example, but Anthropix, their quad model, is actually quite good in this respect as well.
• 9:27
  Speaker 1: So I think what you're going to see is that you know people want the answer to their problem quickly, they want to move on, and so generative AI is going to crush traditional search volume wise, I think.
• 9:44
  Speaker 1: I think there's a place for it, but I think it's going to be at a much more narrow use case.
• 9:47
  Speaker 1: Now, as far as what does that mean for SEO?
• 9:48
  Speaker 1: I don't know the answer to that.
• 9:48
  Speaker 1: I mean the way people have thought about marketing and SEO for a long time now.
• 9:52
  Speaker 1: There's always been the latest SEO hack in terms of how you optimize for the algorithm, but now it's a totally different question.
• 9:58
  Speaker 1: It's like how do you even optimize for the ongoing way?
• 10:01
  Speaker 1: The AI is going to work right, and will Google even provide citations and links back to their sites for some of the content?
• 10:08
  Speaker 1: All of the content?
• 10:09
  Speaker 1: None of the content there's really.
• 10:10
  Speaker 1: It's harder to know, but that's kind of the nature of disruptive innovation, right.
• 10:14
  Speaker 1: It's hard to tell what's going to happen.
• 10:18
  Speaker 2: So in the eating rock scenario, which is the one that I saw the most headlines for, it was something like Google suggesting consumers eat rocks or something like that.
• 10:26
  Speaker 2: You know there's really clickbaity titles, but really that's what happened.
• 10:30
  Speaker 2: It looked like Google's AI overview may have referenced a satirical article from the Onion saying that geologists recommend eating one small rock per day, which was obviously just silly and satirical.
• 10:42
  Speaker 2: How do you think AI will be able to learn the nuances of sarcasm and satire and even trolling?
• 10:51
  Speaker 1: I mean it's getting better.
• 10:52
  Speaker 1: Right now, AI is kind of like an idiot savant.
• 10:54
  Speaker 1: You know, it's basically incredibly brilliant in some ways, but unbelievably stupid in other ways, almost comically so, but that's changing by the minute.
• 11:03
  Speaker 1: So that's one of the things that people tend to have their opinions formed quite quickly and quite early and they tend to stick to them, especially as they talk to their people and express their opinions.
• 11:11
  Speaker 1: So if you form your opinion about what AI can do back in the early days of chat, gpt, when it was GPT-3-5, or even the early GPT-4 days, your point of view is probably quite out of date and you may not have updated your opinion, even though the technology is double, triple, quadruple compared to what it was back then, capability wise.
• 11:29
  Speaker 1: So I think it's something you have to be careful of.
• 11:31
  Speaker 1: Even the statement I just made about the idiot savant, it's like I still think AI is largely in that category.
• 11:37
  Speaker 1: It doesn't have common sense.
• 11:39
  Speaker 1: Right now.
• 11:40
  Speaker 1: There's a lot of missing pieces there, things we take for granted, but it's brilliant and so much so I view AI as an incredible tool belt.
• 11:49
  Speaker 1: I view what are the things AI is really good at and I like to use AI for those specific purposes.
• 11:54
  Speaker 1: I think it's.
• 11:56
  Speaker 1: You know, the all or none mindset is part of the trap people fall into.
• 11:59
  Speaker 1: They're like they want to use AI for everything or they want to use it for nothing at all.
• 12:03
  Speaker 1: I don't think that's the right approach.
• 12:06
  Speaker 1: You have to look at it as a tool set.
• 12:07
  Speaker 1: It's like when Microsoft Word came out, I guess back in the late 80s.
• 12:10
  Speaker 1: People say, well, microsoft Word can't do everything, therefore I'm not going to use it at all.
• 12:16
  Speaker 1: It was extremely limited back then, but it was really good at one or two things.
• 12:19
  Speaker 1: Same thing with something like Excel.
• 12:21
  Speaker 1: This is a different thing.
• 12:22
  Speaker 1: It's much more kind of squishy right in terms of what does the tool actually do?
• 12:27
  Speaker 1: But it's a tool.
• 12:28
  Speaker 1: So, in any event, I think this idea of AI having better understanding of subtleties and content, the nuance, sarcasm, satire, wit, even trolling, will come, and not too far off.
• 12:39
  Speaker 1: It's just.
• 12:39
  Speaker 1: That's a higher order thing, I think, than the AI models that we've become accustomed to will be able to handle.
• 12:45
  Speaker 1: I will also say I think the current models that Anthropic has out there Claude's Opus model, I think, openai's GPT-4.0, and possibly even some much smaller models like Llama 3 would probably actually not have made that error.
• 13:01
  Speaker 1: I'm pretty disappointed that that particular error was made by Google, because actually I have a that error.
• 13:04
  Speaker 1: I'm pretty disappointed that that particular error was made by Google because actually I have a lot of good things to say about their Gemini 1.5 Pro model.
• 13:11
  Speaker 1: It's a great model in so many ways.
• 13:12
  Speaker 1: It just seems like there's just something missing in what they're doing.
• 13:15
  Speaker 1: I don't know what it is.
• 13:16
  Speaker 1: I wish I knew the answer to that.
• 13:17
  Speaker 1: So I think part of this is it's a Google issue.
• 13:20
  Speaker 1: It is an issue overall, but I think Google is behind the other major players at the moment.
• 13:27
  Speaker 2: Interesting For me.
• 13:28
  Speaker 2: What's really notable is that Google is so often seen as this great truth holder.
• 13:35
  Speaker 2: I mean so much so that it's in our vocabulary.
• 13:37
  Speaker 2: I need to Google something.
• 13:38
  Speaker 2: Can I Google this?
• 13:39
  Speaker 2: What do you think about this statement that there's an erosion of trust in Google using these AI overviews and it presenting dangerous, potentially harmful information?
• 13:50
  Speaker 1: You know it might be helpful to talk a little bit about Google's algorithm at a very high level.
• 13:55
  Speaker 1: You know a lot of what Google did originally.
• 13:57
  Speaker 1: That made them the search engine of choice is they used essentially crowdsourcing to figure out which pieces of content were better, more reputable.
• 14:05
  Speaker 1: They had this algorithm called PageRank, which was this idea of looking at these backlinks that a page had, and so if you and I have pages on the web and my page and your page have literally the exact same page content and this is before there were filters on things like that with Google that penalized you for duplication of content Just forget about that part for a second.
• 14:28
  Speaker 1: But if we were to have that and let's say you were to have you know to determine which pieces of content are more trustworthy or better relative to the ranking for keywords and ranking for context they've continually gotten better at.
• 14:52
  Speaker 1: The algorithm is much, much more complex than that these days, but the idea is to leverage the power of the network or the graph on the internet.
• 15:00
  Speaker 1: Um, the problem with that and the weak spot for google that they've addressed and I think they've done a really good job of this traditionally is it's also possible to have lots of links for the wrong reasons, where people link to your site because it's sensationalist.
• 15:13
  Speaker 1: People link to your site because it's negative, because all of us, across the species we, have a negativity bias so we float to that.
• 15:20
  Speaker 1: It's a survival instinct.
• 15:21
  Speaker 1: So it's not perfect and Google's search results, maybe in the first couple of pages for key terms you know, have been groomed but there's a lot of garbage that you found in traditional Google search too.
• 15:32
  Speaker 1: You know you can't trust what you read in the internet is what I've taught my kids, since they, you know, started browsing the web, and whether it comes through Google search or it's their generative AI, you should really be kind of skeptical about what you read online, no matter where it comes from.
• 15:45
  Speaker 1: So I think people who've kind of like taken you know the word of Google is like the gospel or something that's a problem, right, Because that's never really been true.
• 15:55
  Speaker 2: Yeah, thinking particularly about all those articles, basically anything after page one, there's probably a lot of stuff in there that wasn't factual, that we just didn't see.
• 16:08
  Speaker 1: Or even if it's a lower volume topic, like in the association community, a lot of the topics people search on are pretty niche, you know.
• 16:11
  Speaker 1: So you get into these narrow topical areas, you don't have a lot of volume.
• 16:14
  Speaker 1: It's a different game because there's less data for Google to lean on to determine authenticity and trustworthiness and all those things than you can between.
• 16:22
  Speaker 1: Like saying, let's compare, you know two different major like sports websites the Athletic from the New York Times or we're going to compare that to ESPN.
• 16:29
  Speaker 1: You know those are much, much like higher order kind of data sets that you know some Google algorithm can utilize to rank content.
• 16:38
  Speaker 1: But you know, a lot of times in narrow fields it's a lot harder to do that reliably.
• 16:45
  Speaker 2: And, on that note, if you follow us on LinkedIn, you may have seen that ChatGPT was actually linking back to Sidecar when asked about associations and AI, so we're pretty proud about that.
• 16:53
  Speaker 1: Yeah, I was happy to see that.
• 16:54
  Speaker 1: I was surprised, but I was happy to see that.
• 16:56
  Speaker 1: Not surprised that Sidecar was mentioned, just to be clear, but surprised that we got a link back from ChatGPT.
• 17:04
  Speaker 2: All right, moving on to topic two, deepl.
• 17:07
  Speaker 2: Deepl is a machine translation company based in Germany that aims to eliminate language barriers through artificial intelligence.
• 17:14
  Speaker 2: They recently announced a $300 million investment, bringing it to a $2 billion valuation and a funding round led by Index Ventures.
• 17:20
  Speaker 2: Their key product is a translation engine that can translate between multiple languages accurately by understanding context.
• 17:23
  Speaker 2: In a funding round led by Index Ventures, their key product is a translation engine that can translate between multiple languages accurately by understanding context and meaning.
• 17:29
  Speaker 2: The company's mission is to enable global communication and expansion through AI translation capabilities.
• 17:35
  Speaker 2: The company claims their translation quality surpasses competitors like Google Translate not to bag on Google in this episode by being able to capture nuances and convey meaning accurately across languages.
• 17:49
  Speaker 2: Deepl's translation capabilities are not limited to just text.
• 17:52
  Speaker 2: They can also translate documents, websites, etc.
• 17:55
  Speaker 2: While preserving formatting and context.
• 17:58
  Speaker 2: The company's focus is on developing AI to translate languages fluidly so that communication is seamless across cultures.
• 18:05
  Speaker 2: Deepl positions itself as a solution for businesses looking to expand globally by overcoming the language barrier through AI translation.
• 18:13
  Speaker 2: Danny Reamer, whose partner at Index Ventures that I just mentioned, said that in enterprise settings, the demands for accuracy, privacy and security in translations are significantly higher than in consumer context.
• 18:26
  Speaker 2: He emphasized that vertical models designed for specific tasks can outperform more general models in meeting these stringent requirements.
• 18:34
  Speaker 2: Amit, what do you think is most exciting about DeepL's initial success so far?
• 18:41
  Speaker 1: I think that companies like this that are producing specialized models will succeed not necessarily this company, but companies that are doing this because the focus you get from specialization creates a better product in that category.
• 18:55
  Speaker 1: There will be a world of many models for a long, long time, and generalized models are not going to be able to do as well at this specific task as a model trained for it.
• 19:05
  Speaker 1: And DeepL is specifically interesting because you can really add a lot of specialized language and context to that enterprise use case where, for example, an association that had a lot of medical terminology or scientific terminology could utilize their training data to help drive a better version of DeepL's model of their context.
• 19:26
  Speaker 1: That's really important and really hard to do.
• 19:28
  Speaker 1: General purpose models don't have the training in the content area that you're in to be able to do that.
• 19:34
  Speaker 1: Another way to think about this, though, mallory, is when we talk about specialist models versus generalist models, you almost think about industries.
• 19:41
  Speaker 1: You don't typically see banks also manufacturing infant formula.
• 19:45
  Speaker 1: You don't typically see, you know, accountants also writing poetry, although that that can happen right.
• 19:51
  Speaker 1: So, um, usually not professionally in both contexts.
• 19:54
  Speaker 1: So, um, specialization of labor, specialization of industry, are things that you know we as a species again have continued to pursue into deeper and deeper, narrow and narrower spaces.
• 20:05
  Speaker 1: And we see that actually with our association community where you know you might have a large medical association that's like a broad umbrella category but you might have a sub-specialist association where you know there might even be an association of like paper cut repair doctors or something like that, where those doctors just do that one thing right.
• 22:13
  Speaker 1: Hopefully that's not really a category but you never know.
• 22:17
  Speaker 1: So the point would be that you have all this high degree of specialization with kind of the world of humans, and you should expect the same thing with AI and the models that are best trained for specific tasks.
• 22:30
  Speaker 1: They'll get a lot of the reps in those tasks and general purpose models are going to be smarter overall, but they won't have the specializations.
• 22:37
  Speaker 1: This is just that broader trend.
• 22:38
  Speaker 1: We talk about a lot about the world of multiple models and then stringing them together in a way where you might use a deep L for a particular task.
• 22:47
  Speaker 1: Will it replace how you use GPT-4-0 or Quads Opus?
• 22:51
  Speaker 1: Probably not, but those broader models we use for a lot of different general purpose tasks and then you'll dive deep which is, of course, I think, where they got their company name on and you use them for those specific tasks.
• 23:04
  Speaker 1: So I'm excited about this because it just shows that this ecosystem is flourishing in so many ways where a highly specialized company like this raising a pretty significant amount of money this would not have been a thing three, four years ago.
• 23:17
  Speaker 1: There was capital being invested in AI, but not at this scale, and this type of product, this kind of offering, being available, is a direct function of the enthusiasm the investor community has to the broader category.
• 23:31
  Speaker 2: We've mentioned HeyGen several times on this podcast before as a consumer translation application.
• 23:36
  Speaker 2: You and I have both also experimented with it ourselves, but honestly HeyGen kind of seems like the Little Leagues right compared to a company like DeepL.
• 23:44
  Speaker 2: Do you see a place for both in business, or is it that companies like DeepL will outpace the HeyGens?
• 23:51
  Speaker 2: And HeyGens are just fun to use for now but will never be enough to use in a business context.
• 23:57
  Speaker 1: I don't know.
• 23:58
  Speaker 1: I mean I think products like Haygen that have broad consumer appeal might actually it might be the opposite.
• 24:03
  Speaker 1: They might be MLB and DeepL might be in minor leagues.
• 24:06
  Speaker 1: It's hard to say.
• 24:07
  Speaker 1: It's almost like you know, crm is a broad category versus association management software, where AMS is highly specialized for associations.
• 24:15
  Speaker 1: Crm is similar but different, but a much bigger, broader category.
• 24:19
  Speaker 1: A CRM, typically unaltered, is hard to use to run an association, but an AMS is probably not going to be in the quote-unquote major leagues in terms of doing the kinds of things that a big CRM company is gonna do.
• 24:33
  Speaker 1: So I don't know.
• 24:34
  Speaker 1: I think that what you might find perhaps is that specialist companies will thrive in their little pocket in their niche Specialists in an industry or specialists in a particular function like DeepL.
• 24:48
  Speaker 1: But I think HM and tools like that coming back to your comment, they provide a lot of utility and they get you to the 90% mark, which is probably good for actually a lot of business use cases.
• 24:59
  Speaker 1: I could see us using HM to translate all Sidecars video content into whatever language.
• 25:04
  Speaker 1: I think it's good enough for that.
• 25:06
  Speaker 1: We don't typically talk about anything so deeply technical here that it'd be a problem.
• 25:11
  Speaker 1: We don't have that different of a vocabulary.
• 25:12
  Speaker 1: But we were, let's say, you know, medical doctors or physicists or something like that, and we're talking about some like really specialized vocabulary.
• 25:21
  Speaker 1: I think HeyGen probably would have a harder time with that.
• 25:24
  Speaker 2: For sure.
• 25:25
  Speaker 2: I'm just thinking.
• 25:26
  Speaker 2: In one of our monthly intro to AI webinars, we did get a question when we showed HeyGen, asking if they were certified translations.
• 25:33
  Speaker 2: And of course they are not, and I don't really know what the future looks like in terms of having AI generate certified translations, or maybe the whole concept changes in the future.
• 25:42
  Speaker 2: But I'm wondering if listeners would agree that good enough is enough.
• 25:46
  Speaker 2: I think in the case of Sidecar, sure, but I don't know.
• 25:50
  Speaker 2: Like really would we want to be communicating with our customers and getting it only 90% right?
• 25:56
  Speaker 1: Yeah, I'm not sure about that.
• 25:57
  Speaker 1: I think the novelty of it would produce value, but very quickly the expectation would be, in whatever language the listener is listening in, that's as good as the English version.
• 26:05
  Speaker 1: So I think that's a really good point.
• 26:07
  Speaker 1: One thing to think about with all this stuff is that we live in a world where the AI is advancing rapidly, as we always talk about in this podcast, but we also have the ability to chain together multiple different AI models.
• 26:22
  Speaker 1: So a common practice that's emerging is this idea of NAS or multi-agent systems.
• 26:28
  Speaker 1: We've talked about this before I think we had a whole episode on agents, if I remember correctly, a number of months ago, and I've written a lot about it and agents are a very powerful concept in ai, and the basic idea with a multi-agent system is you chain together multiple different ai sometimes the same ai model but prompted differently, and sometimes fundamentally different ai models and you ask them to kind of check each other's work in a way, so you might have one AI that generates a piece of content, another AI that checks that content and says, hey, does this content comply with the instructions that the AI was the first day I was given, if it doesn't pass a certain criteria, maybe there's a rubric you give it to evaluate the content.
• 27:10
  Speaker 1: It gets kicked back to the content writer AI and it goes back and forth X number of times until you produce something.
• 27:16
  Speaker 1: What we found consistently in AI research across the board is that when you use a multi-agentic system like that, you have way better results than any one model, and that's really a key thing.
• 27:27
  Speaker 1: There are some kind of mental traps we fall into with new tools like this, one of which is the idea that a single model has to be able to produce perfect output by itself.
• 27:37
  Speaker 1: Think of it more as a team sport, where you don't need Roger Federer or Serena Williams on court to perform at a superstar level individually.
• 27:45
  Speaker 1: You can have a team of AIs that work together to produce a world-class outcome, along with giving input, obviously, as well.
• 27:53
  Speaker 1: So I think it's a mixed bag right now, but I would be one that over the next 12 to 18 months, this won't be a topic that it'll be so good that it's better than any human translator can keep up with.
• 28:05
  Speaker 1: That's what I think is going to happen, especially with the multi-agent type of scenario.
• 28:11
  Speaker 2: I'm making a grimace because I'm thinking of all the human translators, and that's just a whole other conversation.
• 28:16
  Speaker 2: Yeah, but, amit, do you think that what sounds like we keep getting to the same idea in every episode, or in many episodes, which is that in the future of business and the future of associations, we'll have a suite of specialized AIs that perform different tasks and perform those tasks very well In that world?
• 28:34
  Speaker 2: Do you think there'll be a human managing these specialized AIs?
• 28:39
  Speaker 2: Will it be an AI managing these specialized AIs?
• 28:41
  Speaker 2: What do you think about that?
• 28:43
  Speaker 1: Well, as long as I'm living and breathing, I sure as hell hope there'll be a human.
• 28:46
  Speaker 1: I don't want to be in that world personally.
• 28:50
  Speaker 1: That doesn't have us in the mix somewhere telling these things what to do.
• 28:53
  Speaker 1: But honestly, I think there will be many scenarios where humans in a loop will be more of an artifact of our comfort.
• 29:00
  Speaker 1: It's like having a pilot in the cockpit.
• 29:01
  Speaker 1: These planes, for many years, have all been able to take off and land on their own.
• 29:04
  Speaker 1: I'm not getting on an airplane unless I see a couple of pilots up there, and I prefer if they look like they've been doing it for a while as opposed to you know they're younger than my kids so.
• 29:14
  Speaker 1: I think that's a problem for us more than it is for the technology, but I do think there's a lot of value that we add in terms of our judgment and our reasoning, things like that.
• 29:25
  Speaker 1: So I would say, you know, ultimately, what we are seeing as a trend line is the capability is going up and up and up.
• 29:33
  Speaker 1: I still think there needs to be people thinking through and coordinating things.
• 29:38
  Speaker 1: You know, ultimately, like think about AIs as though you're going to hire them.
• 29:44
  Speaker 1: So when you think about, like a specialized AI, like a Betty or a Skip, which are agents that our companies produce, you should think of them as like you're hiring an employee.
• 29:54
  Speaker 1: You're saying, oh, you're going to hire Betty as an employee.
• 29:56
  Speaker 1: You're going to hire ChatGPT as an employee and ChatGPT is not the best example, but the idea is you hire an AI, you think of it as being part of your team and as these things become more and more sophisticated, I think that analogy will actually be even better than it is today.
• 30:10
  Speaker 1: And so you're going to need a bunch of them, and so you're going to need a bunch of them.
• 30:13
  Speaker 1: I think you're going to need specialists throughout your organization.
• 30:15
  Speaker 1: So that's certainly how I think of it for us.
• 30:20
  Speaker 2: When we've talked about hey, jen, I'll bring it up again.
• 30:21
  Speaker 2: In the past, we've talked about how incredible it could be to translate especially evergreen content that your association has, maybe for manual meetings or articles or journals, into your members' native languages.
• 30:31
  Speaker 2: I'm wondering, amit, because you talk a lot about exponential growth and exponential thinking Can you think of any use cases with a tool like DeepL assuming it is the world's most accurate translator, which is their tagline on their website?
• 30:44
  Speaker 2: What could associations do with a technology like this in the future?
• 30:49
  Speaker 1: I mean, I think there's real-time translation potential at annual conferences or other events, on webinars.
• 30:55
  Speaker 1: I think there's a lot of ways of making accessibility so good that everyone participates at the same level of experience as people in the native language.
• 31:04
  Speaker 1: Um, when I think about translation, though, I look at it as a multidimensional opportunity.
• 31:09
  Speaker 1: So natural language, the natural language like English, spanish to French, to German, to Chinese, to whatever it's awesome.
• 31:20
  Speaker 1: It's amazing to be able to bridge those gaps, to have no loss of information, no loss of the quality of the experience as you go from one language to the next.
• 31:24
  Speaker 1: That's, that's unbelievably cool.
• 31:25
  Speaker 1: I think it's going to really bring the world closer together.
• 31:28
  Speaker 1: It's kind of the touchy-feely side of it, but I think a lot of things are lost in translation, as they say, and so you can solve a lot of problems that way.
• 31:36
  Speaker 1: So that's exciting.
• 31:37
  Speaker 1: But there's other facets to this or other dimensions to this.
• 31:39
  Speaker 1: So you think about translating content in terms of the context.
• 31:44
  Speaker 1: So you say, oh well, this particular piece of content is meant for professionals in this field.
• 31:49
  Speaker 1: Well, what if we wanted to present the same ideas to people who are lay people in that field?
• 31:54
  Speaker 1: So I have a piece of content that is for, let's say, attorneys, and it is for lawyers focused on intellectual property law.
• 32:02
  Speaker 1: So we've had a guest on this podcast recently who's a patent expert software patent expert specifically.
• 32:08
  Speaker 1: So a piece of content that he might read or write for his colleagues that are deep in that world may be really good information, but not really consumable for the average person, even an educated business person.
• 32:21
  Speaker 1: In fact, that content might not even be that consumable by the average lawyer.
• 32:25
  Speaker 1: And so this is a key capability of saying, without losing information and without getting it wrong, how do we translate the highly technical patent law article for a regular lawyer who's not a patent lawyer to understand it?
• 32:38
  Speaker 1: Well, and the same thing for, like, the average business person?
• 32:42
  Speaker 1: Or then, what about translating it for a high school student, right?
• 32:45
  Speaker 1: So that's a form of translation.
• 32:47
  Speaker 1: Another type of translation is tone and style.
• 32:50
  Speaker 1: So if you want to engage your audience, you have to learn how to engage them the way they like to be engaged.
• 32:55
  Speaker 1: This is a form of personalization.
• 32:57
  Speaker 1: In a sense, the same content but presented a different style, a different tone, a different cadence, perhaps incorporating other elements.
• 33:05
  Speaker 1: You know, especially as you think about different age groups, or you know just different personality styles.
• 33:10
  Speaker 1: You know, making something a little bit funny versus making it more serious, right?
• 33:14
  Speaker 1: So there's all those different aspects.
• 33:16
  Speaker 1: I call this content transformation.
• 33:18
  Speaker 1: It's actually a chapter in the next edition of Ascend second edition of Ascend that we're working on now.
• 33:25
  Speaker 1: That will be available in about a month and a half.
• 33:28
  Speaker 1: By August 1st is our goal to have it back on Amazon available to download, and this new edition we have a lot of new content that we've worked hard on.
• 33:36
  Speaker 1: Content transformation is one of the topics that we cover in detail, and I'm really excited about it.
• 33:40
  Speaker 1: I think DeepL is a great example of a piece of that solution.
• 33:44
  Speaker 1: Another thing, too, though, to remember is as much as I said, I think the world is going to have many AI models.
• 33:51
  Speaker 1: These types of folks have to move fast and have a lot of resources, because six months from now, gpt-4.0 is going to be whatever the next thing is called, and Google and other people will keep getting better and better.
• 34:02
  Speaker 1: So the general purpose models might not be 80% good.
• 34:06
  Speaker 1: They might be 98% good, and then a company like DeepL might be a really highly, highly specialist tool.
• 34:12
  Speaker 1: So that's the risk someone like that runs is that they get run off the road because the main model gets so good so fast.
• 34:17
  Speaker 1: I don't really know how you protect against that.
• 34:21
  Speaker 1: That's tough.
• 34:21
  Speaker 2: Right, that's really helpful.
• 34:23
  Speaker 2: What you said about thinking about translation as more than just one language to another, which I think is a trap we probably all fall into, but that's definitely helpful.
• 34:32
  Speaker 1: Well, in 2018, I published this other book called the Open Garden Organization, and in that book, my main argument is that associations need to throw open the gates and say we need to welcome everyone who's interested in what we're good at, rather than thinking about this binary mindset of member non-member and then come into my walled garden, which has all this beautiful, amazing content and stuff.
• 34:52
  Speaker 1: If you're a member, that mindset is a bit antiquated and it's too simplistic.
• 34:56
  Speaker 1: People who are interested in what you're good at are people you need to find a way to engage, but they come from all sorts of backgrounds.
• 35:03
  Speaker 1: And so you might be the world's greatest expert on intellectual property law, but you might need to engage people who are not IP law experts.
• 35:11
  Speaker 1: Right, there's always other people that closest encircle the nucleus of your association's membership may, in fact, be IP lawyers, but you might have other lawyers as the next concentric circle around that.
• 35:22
  Speaker 1: Then you might have, like you know, a broader circle around that.
• 35:25
  Speaker 1: Maybe law students are in the mix, maybe business students are in the mix, and but these are people who could value you and your expertise.
• 35:32
  Speaker 1: And so what I argue in the Open Garden organization is this thesis that you should focus on value creation across all of these dimensions, and back then AI was in a very formative time.
• 35:42
  Speaker 1: I've been working with AI for over a decade now.
• 35:45
  Speaker 1: In 2018, there was AI.
• 35:46
  Speaker 1: People just weren't talking about it like they are now.
• 35:48
  Speaker 1: We were trying to get people's attention with AI back then, but no one was really listening yet.
• 35:53
  Speaker 1: But AI was a part of the solution we even contemplated back in that book.
• 35:58
  Speaker 1: But really the idea is simple.
• 35:59
  Speaker 1: It's like you have value to create.
• 36:01
  Speaker 1: How can you create that value, the way the person on the other end wants to receive it?
• 36:05
  Speaker 1: That's the key to it.
• 36:08
  Speaker 2: Well, speaking of trying to get people's attention with AI back in 2018,.
• 36:12
  Speaker 2: This is a really good segue to our third topic, which is Rex.
• 36:15
  Speaker 2: So, listeners, you may have heard of Rasaio.
• 36:18
  Speaker 2: It's a company that's part of the Blue Cypress family, along with Sidecar, and it's a smart newsletter company that uses AI to send truly personalized email newsletters at an individual level.
• 36:29
  Speaker 2: It recently released a new product called Rex.
• 36:33
  Speaker 2: Rex is an AI driven personalization engine.
• 36:35
  Speaker 2: It has the ability to make custom recommendations on any platform, at any time, anywhere you apply it.
• 36:41
  Speaker 2: By utilizing new vector-based technologies, rex makes recommendations that go beyond just patterns.
• 36:47
  Speaker 2: Think of an AI brain that picks up on emotions, wit, conversational preferences and many other intricacies that go beyond human capabilities.
• 36:55
  Speaker 2: Rex's vector-based mind turns numbers into meaning at a scale impossible to achieve otherwise.
• 37:02
  Speaker 2: So what does this mean for associations?
• 37:05
  Speaker 2: Imagine a world where every interaction a member has with your association can be catered specifically to them From content suggestions on your site to personalized networking opportunities at events.
• 37:16
  Speaker 2: Rex can provide each member their very own intentional experience that highly resonates with them.
• 37:22
  Speaker 2: Amit, can you share a little bit about Rex, anything that I haven't covered with our listeners?
• 37:29
  Speaker 1: I think that's a great overview.
• 37:30
  Speaker 1: You know, people who've heard me speak on AI probably hear a pretty consistent drumbeat, because part of what I always hammer home is that friction is the enemy that you have to seek out and destroy at all times.
• 37:41
  Speaker 1: And rather than having people have to figure out what they should do to comply with your website structures or how you guys do business, making yourself conform to what the individual wants to get from you and how they want to receive, it is the whole idea behind personalization.
• 37:59
  Speaker 1: You know, the marketer's dream for a long time has been this idea of a segment of want, meaning you're really marketing to that one person.
• 38:06
  Speaker 1: If I know Mallory really well, I can market to Mallory in a way that other people probably could not.
• 38:13
  Speaker 1: And but is that something we've really had the ability to do Up until recently?
• 38:17
  Speaker 1: No, and recently meaning in the last handful of years, very large companies, very large budgets have had access to the type of technology that Rex is based on.
• 38:25
  Speaker 1: Ross has been using some of these technologies for years and has continually evolved the engine, but the idea essentially, is to do what Ross has done with newsletters for years, but to make it more general purpose, to make it so that you could recommend and personalize literally anything, as you mentioned, across all these different platform effects.
• 38:43
  Speaker 1: So how it does, that is a whole long big conversation.
• 38:46
  Speaker 1: We did a webinar earlier this week and that's available on the rasaio slash REX webpage.
• 38:52
  Speaker 1: That recorded webinars there for people who are interested in more details.
• 38:55
  Speaker 1: But it is using cutting edge AI and it's done in a way where it's affordable for pretty much every association on the planet, so we're pretty excited about it.
• 39:04
  Speaker 2: In that webinar and we'll be linking that webinar or the landing page that has the webinar embedded in the show notes.
• 39:09
  Speaker 2: You mentioned that Rex is an engine, not an app.
• 39:12
  Speaker 2: Can you explain that?
• 39:14
  Speaker 1: Yes, so Rasaio's newsletter product is an app.
• 39:18
  Speaker 1: It is something that you have a user interface for.
• 39:21
  Speaker 1: You go in, you configure how you want your newsletter to look, content sources, how you're prioritizing things, various rules and settings, etc.
• 39:28
  Speaker 1: Etc.
• 39:28
  Speaker 1: And it in turn creates a newsletter and it sends it out.
• 39:31
  Speaker 1: You have a lot of other apps in your ecosystem, right, that are things that actually perform a specific function for you.
• 39:38
  Speaker 1: In comparison, an engine is more like, uh, what you envision, like you know, uh, just a general purpose utility.
• 39:46
  Speaker 1: It's something that has the ability to take inputs and produce outputs, and you can wire it up to literally any other digital property.
• 39:53
  Speaker 1: So, um, we went over nine different use cases on tuesday when we did the webinar, and some of these use cases include things like personalized event marketing, where you send each individual in your database an email or other form of communication that highlights for them the sessions and speakers that are most likely to be interesting to that person, that are most likely to be interesting to that person.
• 40:17
  Speaker 1: Another example is professional networking, where you recommend to individuals the people that they might be most interested in connecting with, either at a conference or independent of that.
• 40:26
  Speaker 1: Another example is in the world of education, suggesting courses or modules or lessons even that are specifically relevant to the professional interests and direction of a particular individual, rather than kind of generalized learning tracks or generalized segments in marketing.
• 40:44
  Speaker 1: So those are some examples, but RECS doesn't do all of that soup to nuts.
• 40:48
  Speaker 1: It's not an app that you just press a button and it does those things.
• 40:51
  Speaker 1: It's an engine that enables any of those scenarios and any other scenario you can imagine where you're basically doing comparisons.
• 40:59
  Speaker 1: At the heart of any personalization technology, whether it's Rex by Rostio or anything else, is this idea of being able to compare things, saying, okay, well, mallory and this article have certain similarities and we think there's a high degree of relevance between this article and Mallory, so let's send it to her.
• 41:18
  Speaker 1: Or maybe Mallory has a high degree of links to another person and that person in turn has similarities to a particular event, and so therefore, because that person's linked to Mallory, we might guess that that person's also interested in that conference or that session.
• 41:35
  Speaker 1: So that's what recommendation engines basically do also interested in that conference or that session.
• 41:39
  Speaker 1: So that's what recommendation engines basically do.
• 41:44
  Speaker 1: And so what Rasa's RECS engine does is it does that with any kind of data and it's a super high performance real-time engine that you can plug into a website or into an app, plug it into an email engine.
• 41:51
  Speaker 1: There's a lot of things that you can do, but it is very much an engine.
• 41:54
  Speaker 1: So it's not something you just like you know.
• 41:55
  Speaker 1: Sign up with a credit card, press a button, turn it on an engine.
• 41:58
  Speaker 1: So it's not something you just like you know.
• 41:59
  Speaker 1: Sign up with a credit card, press a button, turn it on, and it does all these things.
• 42:01
  Speaker 1: These are examples of use cases and you have to implement RECs, meaning you have to think about the use cases you want to implement and then you have to connect RECs to your data sources and then RECs can do its magic in terms of great recommendations.
• 42:13
  Speaker 1: But then you have to do something with those recommendations.
• 42:16
  Speaker 1: You have to hook them up to email systems or to your website or to your mobile apps.
• 42:19
  Speaker 1: So there is work involved in this and over time I'm sure many apps will emerge on top of Rex.
• 42:26
  Speaker 1: Whether those apps are built by Rasa or built by somebody else doesn't really matter, but Rex is an enabling technology.
• 42:33
  Speaker 1: It basically bundles the really really hard work of doing recommendations really well at scale so that it's approachable by anyone in this community.
• 42:43
  Speaker 1: And then the idea is that this ecosystem, over time, will have an abundance of apps that sit on top of it.
• 42:48
  Speaker 1: Right now you have to actually connect the dots yourself and we have people across the Cypress family who will actually do that work for you if you want.
• 42:55
  Speaker 1: But the point is it's not a turnkey, press the button, play type of app.
• 42:59
  Speaker 1: It's more of a toolkit that's ready to go.
• 43:02
  Speaker 2: So you plug your data sources into this engine.
• 43:05
  Speaker 2: Do your data sources have to be in a certain format?
• 43:08
  Speaker 2: Can you all work with any data sources or is it preferred?
• 43:12
  Speaker 2: I know we've talked about on this podcast that you have a common data platform, or CDP.
• 43:17
  Speaker 2: What does that look like?
• 43:19
  Speaker 1: That's the key ingredient is having a CDP.
• 43:21
  Speaker 1: Member Junction is the preferred CDP because it's an open source, totally free CDP that obviously we in this family are quite familiar with because Member Junction is one of our companies here at Blue Cypress.
• 43:32
  Speaker 1: But it can really be anything actually that Rex can plug into Rex, just a pre-built connector to any data element inside the member junction CDP and the member junction CDP is totally agnostic as to data source, so you could pull the data from any LMS or AMS or CMS into member junction and then the RECS engine automatically has access to all of it.
• 43:54
  Speaker 2: So is the CDP necessary or preferred, so is the CDP necessary or preferred.
• 43:58
  Speaker 1: I would say that for early adopters it's necessary because for people who maybe want to implement this a year down the road, there probably will be connectors for RECs to connect directly to some other systems.
• 44:08
  Speaker 1: But right now the preferred path really is the required path, which is the member junction connection.
• 44:13
  Speaker 1: So the good news is that doesn't take any extra work.
• 44:16
  Speaker 1: You need to get your data centralized anyway.
• 44:18
  Speaker 1: Most associations do not have a central data source.
• 44:21
  Speaker 1: If they do, if you already have like a data warehouse or some other centralized data structure, that's totally fine too.
• 44:26
  Speaker 1: There's a way of connecting that to Rex.
• 44:29
  Speaker 1: But you know, essentially the member junction kind of acts like a middleware and is basically just there to house the data that you need in order to feed Rex.
• 44:38
  Speaker 1: Rex just needs basically all of your data in order to do what Rex does.
• 44:43
  Speaker 2: Okay, and I suppose Rex only works well if the association has collected good quote-unquote data from its members from across resources.
• 44:53
  Speaker 1: No, that's one of the common assumptions people make, but the problem is I don't know of an association that has good data.
• 44:58
  Speaker 1: Honestly, most organizations don't have quote unquote good data.
• 45:02
  Speaker 1: It's messy, it's dirty, there's duplicative data, there's incorrect data.
• 45:10
  Speaker 1: That's the beauty of this AI technology is that it doesn't get hung up too easily on things that would throw off a lesser technology.
• 45:15
  Speaker 1: It's almost as good as what you would say to a human looking at two pieces of data and saying, oh, does this make sense?
• 45:22
  Speaker 1: There's that level of intuition that goes into the way the technology works, so it's not thrown off by minor incorrect things that exist in the data.
• 45:33
  Speaker 1: A lot of people say, oh, I really want to do all the AI stuff, but I got to clean up my data first.
• 45:37
  Speaker 1: The problem is that will never happen.
• 45:39
  Speaker 1: You will never clean up your data unless you actually go use AI to do it.
• 45:43
  Speaker 1: Ai is good at cleaning up data, so, but for using RECS, you don't need to have pristine data.
• 45:47
  Speaker 1: You just need to get your data somewhere, and you actually don't need to have data that you probably think you need.
• 45:52
  Speaker 1: So you might in which, by the way, is typically what the data shows that people are awesome.
• 46:11
  Speaker 1: Awesome at telling you what they used to be interested in, they're not so great at telling you what they're about to be interested in, which is what you really want to figure out.
• 46:23
  Speaker 1: And that's actually what AI is really good at figuring out, based on, like, thousands and thousands of factors that are entirely non-obvious to us and probably indecipherable to us to a large extent.
• 46:29
  Speaker 1: So you don't need to have profile data, you just need to have data, and then the connections amongst that data is something that the AI can largely figure out on its own.
• 46:38
  Speaker 1: That's really the power of this generation of AI technology which, again, like we didn't release Rex as an engine until now.
• 46:45
  Speaker 1: Rex has existed, by the way, inside Rasa the whole time.
• 46:47
  Speaker 1: It's been the underlying AI engine that Rasa has had many, many iterations on over a lot of years, with a lot of investment.
• 46:54
  Speaker 1: But only now is the technology good enough where we can generalize it, where it can recommend anything across any platform.
• 47:01
  Speaker 1: Previously, in order to make it work as well as it has been working, we had to highly, highly specialize it on just the newsletter domain, but now the tech has gotten so good, so fast and so inexpensive under the hood that it's an accessible enough technology where we can package it up into this engine as the way we're doing now.
• 47:19
  Speaker 2: That's really exciting.
• 47:19
  Speaker 2: I actually did want to ask you why now.
• 47:22
  Speaker 2: I mean I figured it's because the technology has gotten to the point where it is, but it seems like how long would you say that Rex has been being worked on?
• 47:29
  Speaker 2: Five years, or no-transcript.
• 47:47
  Speaker 1: So we started Rasa back when I still owned Aptify and then, before we sold Aptify, I purchased Rasa back from Aptify because Rasa started out in my business subsidiary of Aptify, and then I acquired the company from Aptify before we sold Apt.
• 47:59
  Speaker 1: I purchased Rasa back from Apple Pie because Rasa started out in my business subsidiary of Apple Pie and then I acquired the company from Apple Pie before we sold Apple Pie.
• 48:04
  Speaker 1: Anyway, that's a little historical bit of trivia that someone can ask one day if they want to, but in any event, Rasa has been around I guess since 2016.
• 48:10
  Speaker 1: So that's eight years, and the thesis was back as a CRM vendor, the idea was to have all this great data locked up in CRM type systems.
• 48:18
  Speaker 1: What could you do with AI?
• 48:20
  Speaker 1: And this is the very early days of the deep learning revolution, which is like the middle part of last year.
• 48:25
  Speaker 1: This is pre-generative AI, this is all predictive AI, and these technologies were very, very expensive and difficult to work with.
• 48:34
  Speaker 1: So we poured a lot of dollars and a lot of time into designing the architecture and over time, like every year pretty much like Fockwork we had to pretty much retool the whole engine and keep making it better and better.
• 48:43
  Speaker 1: But we were looking for.
• 48:45
  Speaker 1: We knew there was something there.
• 48:46
  Speaker 1: We knew there was something there in terms of personalization.
• 48:49
  Speaker 1: We weren't sure how to apply the idea of personalization.
• 48:52
  Speaker 1: Another little trivia side note for those of you that know Rasa really well is before Rasa was a newsletter company.
• 48:58
  Speaker 1: Mallory, do you know what Rasa used to do before that?
• 49:00
  Speaker 2: the first product, it was a community like Facebook for associations.
• 49:04
  Speaker 2: I think is what I remember, yeah it was an online community product.
• 49:07
  Speaker 1: It was a very simple but AI-driven online community.
• 49:10
  Speaker 1: I did personalized news feeds, and that was the idea was like, hey, that's a great place to use AI for personalization.
• 49:15
  Speaker 1: And that turned out to be a really bad go-to-market strategy for Rasa for a lot of reasons, mainly because adoption of that kind of product is as complex in many ways as adopting a new website or a new AMS.
• 49:26
  Speaker 1: And then, actually, what happened was we got a call from several customers Actually, I was going to say a customer, but several customers are like you know what we really love the email that the community spits out every day.
• 49:40
  Speaker 1: That is this truly personalized email digest that tells us exactly what we need to know, each of us needs to know.
• 49:46
  Speaker 1: That's the number one thing we get out of this product.
• 49:48
  Speaker 1: Can we have just that?
• 49:50
  Speaker 1: Can we have that?
• 49:51
  Speaker 1: Outside of this community?
• 49:52
  Speaker 1: We're like, oh, light bulb, you know.
• 49:54
  Speaker 1: And so we're like, oh, and that's a very narrow use case, where in an online community it's actually broader, but in a newsletter it's a very, very hyper-specialized thing.
• 50:03
  Speaker 1: And back then you were training these AIs to be really, really narrow in what they did.
• 50:08
  Speaker 1: So we said, okay, let's do that.
• 50:09
  Speaker 1: So, as we mentioned in the webinar on Tuesday, rasta has over 500 organizations using the newsletter product actually many more than that beyond using the kind of the consumer grade version, but in terms of enterprise customers and it's very successful because, if you look at the statistics, people who use personalized newsletter technology have typically double the open rates and a factor of three to five X greater click rates.
• 50:33
  Speaker 1: And the answer why that's the case is simple People like the content, and if you like the content, you open the thing more and you click on stuff more.
• 50:41
  Speaker 1: It's really simple.
• 50:41
  Speaker 1: And so we're saying look, you know now the technology is double the double, the double the power, and Rasa keeps getting better as a newsletter product, but it's time for Rex to be set free.
• 50:52
  Speaker 1: So Rex has been kind of this captive engine within the Rasa newsletter platform and he's all grown up and ready to help the world in lots of other ways now.
• 51:00
  Speaker 1: So we're proud parents letting Rex go off beyond the gates of Rasa these days.
• 51:07
  Speaker 2: What a great story.
• 51:08
  Speaker 2: Did we miss out on a marketing opportunity?
• 51:10
  Speaker 2: Not making it like set the T-Rex free or like going with a dinosaur.
• 51:15
  Speaker 1: Rex actually is an alligator.
• 51:17
  Speaker 1: So, rasa being a New Orleans based company, rex has been the mascot and the AI engine and an alligator on the Rasa website for years.
• 51:25
  Speaker 1: Another thing a lot of people don't notice, but it's part of the Rasa branding, so now he's all grown up.
• 51:31
  Speaker 2: I learned that.
• 51:32
  Speaker 2: I think it's really interesting to in that moment when Rasa was a community platform, which is kind of crazy to think of how you couldn't have seen where it was going, but now that we're here looking back, it's almost like duh, it was there all along.
• 51:44
  Speaker 2: On this podcast Even, we've talked about personalization personalization but how do you do it?
• 51:49
  Speaker 2: And it's so exciting to see that there is a product in the association space that's here for you.
• 51:56
  Speaker 1: Yeah, we're pretty pumped about it.
• 51:57
  Speaker 1: I mean, the idea of being able to do it generally means that it's a toolkit, it's an engine, it's something that can be applied to any problem, and so we have people already starting down the path of doing proof of concepts to test it out with specific use cases.
• 52:10
  Speaker 1: We're recommending people who want to try it out.
• 52:11
  Speaker 1: They'll do like a short proof of concept, you know, four, six, eight weeks, something like that, really quick, and just test one idea.
• 52:19
  Speaker 1: So do you like a personalized event marketing campaign or a personalized session recommendation campaign?
• 52:24
  Speaker 1: Or, uh, you know a professional networking recommendation set, you know, just any kind of personalization or recommendation.
• 52:30
  Speaker 1: Just pick one, do something simple.
• 52:32
  Speaker 1: Don't boil the ocean, as we like to say here.
• 52:34
  Speaker 1: Don't go after everything at the same time and if it works well there, try a second one, try a third one, right, keep going from there.
• 52:40
  Speaker 1: So, um, it's, it is work to do this, but it's very rewarding work is what I would say.
• 52:46
  Speaker 1: It's pretty amazing what happens when you do this.
• 52:48
  Speaker 1: We have people already down the path here that I'm describing that are just kind of blown away like they.
• 52:53
  Speaker 1: They go to their board, they show them this stuff.
• 52:54
  Speaker 1: They're like how is this thing recommending stuff so well?
• 52:58
  Speaker 1: It's uncanny.
• 52:59
  Speaker 1: How did it know there was a connection between these two people that seemingly have nothing related, but really they actually have a pretty deep connection.
• 53:05
  Speaker 1: But it's like eight levels of interaction away from being obvious.
• 53:10
  Speaker 2: I'm really excited to see how this plays out.
• 53:12
  Speaker 2: I'm sure we'll be talking a lot about Rex in the future, in the coming episodes and for all our listeners that made it to this point.
• 53:19
  Speaker 2: Thank you so much for tuning in.
• 53:20
  Speaker 2: If you like the Sidecar Sync, please subscribe, give us a review or if you're listening on your mobile device, you can send us a text message with the link in the show notes.
• 53:28
  Speaker 2: We will see you all next week.
• 53:31
  Speaker 1: Thanks, Valerie.
• 53:32
  Speaker 1: Thanks for tuning into Sidecar Sync this week.
• 53:36
  Speaker 1: Looking to dive deeper?
• 53:37
  Speaker 1: Download your free copy of our new book, Ascend Unlocking the Power of AI for Associations at ascendbookorg.
• 53:45
  Speaker 1: It's packed with insights to power your association's journey with AI.
• 53:49
  Speaker 1: And remember, Sidecar is here with more resources, from webinars to boot camps, to help you stay ahead in the association world.
• 53:57
  Speaker 1: We'll catch you in the next episode.
• 53:58
  Speaker 1: Until then, keep learning, keep growing and keep disrupting.