This week, we bring you an abundance of AI news in a special two episodes. Join us for Episode 93 as hosts Paul Roetzer and Mike Kaput explore Meta's powerful open-source models and AI assistant integration, Microsoft's impressively realistic VASA-1 video generator, and the intriguing concept of "Service-as-Software." Stay tuned for even more this week!
Listen or watch below—and see below for show notes and the transcript.
Today’s episode is brought to you by rasa.io.
Rasa.io is the ultimate platform for AI-powered newsletters. If you’re looking to transform your email newsletter into a powerful, engaging tool that truly resonates with your audience, rasa.io is the game-changer you need. Join the 500+ organizations already making their newsletters smart.
Visit rasa.io/maii today.
00:05:25 — Llama 3
00:31:18 — Microsoft’s VASA-1
00:37:29 — The Service-as-Software AI Paradigm Shift
00:47:56 — Adobe Adding Sora, Runway, Pika to Premiere Pro
00:53:26 — Boston Dynamics' Atlas Robot + The Robotics Revolution
00:56:24 — HeyGen’s AI UGC
00:58:19 — 20VC Interview with Sam Altman
01:00:55 — Drake Releases Track with AI Voices of Snoop Dogg and Tupac
01:02:59 — AI for Interviewing Job Candidates
01:06:55 — AI Fatigue
Meta Announces Llama 3
Meta just made some big AI moves this week. The company announced the release of Llama 3, the next version of its foundational open-source AI models. Meta says that Llama 3 outperforms other open source models like Gemma from Google and Mistral.
It also appears to outperform Gemini Pro 1.5 and Claude 3 Sonnet on at least some major benchmarks.
Today, there are two open source Llama 3 models that developers can now freely use, an 8B parameter model and a 70B parameter one. There is also a 405B parameter version of Llama 3 that is coming soon and is still being trained. (Though it’s unclear right now if this 405B version will be open sourced.)
Meta told The Verge that Llama 3’s training dataset is 7X the size of Llama 2’s, but didn’t use any Meta user data. Notably, it does include both a mix of what the company says is “public” internet data and synthetic AI-generated data.
Also exciting is the fact that Llama 3 will power the Meta AI assistant that is now being integrated into Instagram, Facebook, WhatsApp, and Messenger. Meta AI has been out for awhile, but Llama 3 will give it a serious upgrade in capabilities—and you can now access those capabilities in a standalone AI assistant if you go to Meta.ai.
Meta AI provides outputs just like ChatGPT and also allows you to generate images. It’s also worth noting that Meta AI can actually search the web for you as you interact with the AI assistant.
Microsoft’s VASA-1
Microsoft just released research on an AI model that can deepfake someone from a single photo. The model is called VASA-1, and it can use a photo to create a realistic video of a person talking or singing that syncs up with an existing audio track.
Reports Ars Technica:
“The VASA framework (short for "Visual Affective Skills Animator") uses machine learning to analyze a static image along with a speech audio clip. It is then able to generate a realistic video with precise facial expressions, head movements, and lip-syncing to the audio.”
Right now, it does not clone or simulate voices. Interestingly, the model was trained on YouTube clips. It used a dataset created years ago by researchers that contains data extracted from videos uploaded to YouTube.
Given the model’s apparent speed and power, it could possibly be used in real time applications like video conferencing—or power avatars that render locally. Microsoft researchers, wary of the ways this technology could be misused, are not openly releasing the code that powers the model.
The Service-as-Software AI Paradigm Shift
A possible AI-powered paradigm shift in SaaS is getting a lot of attention online. The concept is called “service-as-software” and it’s outlined in an essay from Foundation Capital, a VC firm that’s invested in notable names like Netflix and AI firms like Jasper.
Foundational Capital defines the “service-as-software” paradigm shift in the following way: AI is causing a transition from Software-as-a-Service to Service-as-Software.
What that means is this: In the past, you bought access to a software tool or platform, which you then used to achieve your desired outcomes in your work.
However, AI is changing this, by not just giving you access to tools, but access to bots that essentially act as brains.
In other words, when you buy an AI tool, you are buying access to a bot that can increasingly perform a service for you, instead of giving you the ability to perform that service better on your own.
In one example: Instead of buying QuickBooks in order to empower your finance professionals or an outside firm to maintain your books, you may simply be buying access to an AI accountant that does the job of the financial professional or outside firm entirely.
This is a change that has massive implications for both jobs and SaaS companies, according to Foundational Capital.
Says Foundational Capital: “The size of the opportunity for AI disruption is many multiples larger than Salesforce could ever be.”
In fact, they ask: “So how much work will AI + automation and the Service-as-Software model do away with? We believe this is a $4.6 trillion-dollar question.”
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: Why would Meta release something this powerful for free? Like why would they make this available to everyone? When Google's charging for it, OpenAI is charging for it, Like we've talked about it on show many times, is open source good or bad? Does it create, dangers to society?
[00:00:18] Paul Roetzer: Welcome to the Artificial Intelligence Show, the podcast that helps your business grow smarter by making AI approachable and actionable. My name is Paul Roetzer. I'm the founder and CEO of Marketing AI Institute, and I'm your host. Each week, I'm joined by my co host, and Marketing AI Institute Chief Content Officer, Mike Kaput, as we break down all the AI news that matters and give you insights and perspectives that you can use to advance your company and your career.
[00:00:48] Paul Roetzer: Join us as we accelerate AI literacy for all.
[00:00:55] Paul Roetzer: Welcome to episode 93 of the Artificial Intelligence Show. [00:01:00] I am your host, Paul Roetzer, along with my co host, Mike Kaput.
[00:01:03] Paul Roetzer: We coming to you, well, well, this releases on April 23rd. We are recording 10 30 m. April 22nd. I don't know about you, Mike, I have not looked at Twitter yet this morning, so I don't know if
[00:01:16] Paul Roetzer: anything has happened this morning that we needed to add, but it's crazy.
[00:01:20] Paul Roetzer: Like we, you know, again, I've shared how Mike and I do this, but basically, like, we sandbox articles, research reports, podcasts, videos, all
[00:01:29] Paul Roetzer: week long, and then Mike goes through on usually a Friday or maybe on a Sunday. Curates everything into the topics. I'll then go through and make sure he and I are on the same page with what are main topics versus rapid fire.
[00:01:40] Paul Roetzer: What's going to get moved to later. And
[00:01:43] Paul Roetzer: was sitting there having coffee this morning, like prepping for this. And I messaged Mike. I was like, dude, we, I think we got to do two episodes week.
[00:01:49] Paul Roetzer: I'm sort of like getting this out ahead of time. I, there was literally no way to cover everything in this single episode this week that we were Going to cover and we [00:02:00] honestly like move some things out.
[00:02:01] Paul Roetzer: so yeah, like, spoiler alert, we're actually gonna do a second episode this week because
[00:02:08] Paul Roetzer: there's just so much to talk about and I think so many important implications to the topics we're talking about.
[00:02:15] Mike Kaput: to go through
[00:02:16] Paul Roetzer: we got, we have a lot to go
[00:02:18] Paul Roetzer: today, we are, we are gonna do a second episode that should come out on, Thursday the 25th.
[00:02:24] Paul Roetzer: So it's the same format. We're not. You know, it's not a different format. It's basically going
[00:02:28] Paul Roetzer: to be three main topics and seven rapidfire or so, because, otherwise I just, I didn't know how we were going to get through everything. So, yeah, double up this week. We will be with you on Tuesday and Thursday this week.
[00:02:41] Paul Roetzer: and then everything else that happens, we'll talk about. Do our usual episode next week. All
[00:02:47] Paul Roetzer: So today's, first episode of the week is brought to us by rasa. io is the ultimate game changer for AI powered newsletters. Rasa. io's smart newsletter platform, tailors your newsletter [00:03:00] content for
[00:03:00] Paul Roetzer: each and every subscriber and automates tedious newsletter production tasks. We've known the team at Rasa for years now. We literally have, I they were one of the first companies started
[00:03:10] Paul Roetzer: talking to back in 2016, 17, when the Institute was created. I think their solution is well worth checking out. Mike and I have used it for years for internal purposes. We don't use it to send our external newsletters, but Mike and I actually use it as a research tool internally to kind keep up on,
[00:03:27] Paul Roetzer: on so we've, you know, been. Testing it out ourselves for years now. Um, so join 500 plus organizations leveraging rasa.io and get a demo today at rasa. That's r aa.io/mai. Uh, we also have a note today
[00:03:47] Paul Roetzer: that our 2024 State of marketing aI survey is now open and in the field. So we've been doing this since the. Third year, Mike, or fourth year we're doing
[00:03:56] Mike Kaput: think it's third year.
[00:03:58] Paul Roetzer: Third year. So last year [00:04:00] we had over 900 respondents take this. It's about a, about 20 questions, Mike? 21 questions, something like that.
[00:04:07] Paul Roetzer: So can take it in probably like four or five minutes. it's pretty we would really appreciate you taking the time.
[00:04:14] Paul Roetzer: to give us your responses, we're basically, we're partnering with Drift again this year. They've been our partner on this every year. So we appreciate Drift and their support.
[00:04:22] Paul Roetzer: but it takes a deep dive into how hundreds, hopefully this year, thousands of marketers are actually using and
[00:04:28] Paul Roetzer: applying aI in their work. By filling out the survey, you're helping the entire industry grow about AI. Only takes a few minutes, as I said. we will send
[00:04:36] Paul Roetzer: a copy of it when it releases, which Mike is slated for when? When do we drop the results?
[00:04:41] Mike Kaput: it's in late July,
[00:04:43] Paul Roetzer: okay, so coming this summer, so will survey will be open now for another month or so?
[00:04:49] Mike Kaput: couple months, we'll probably
[00:04:51] Paul Roetzer: We'll probably do about
[00:04:51] Mike Kaput: 60 days or so.
[00:04:52] Paul Roetzer: Okay. But don't delay. Go it some now so you don't forget.
[00:04:56] Paul Roetzer: go to stateofmarketingai. com. Again, that's
[00:04:59] Paul Roetzer: of [00:05:00] marketing ai.com. You can one download last year's report
[00:05:03] Paul Roetzer: you're there. So you can see the report from last year and there's a link to go and take this year's survey. So again, we appreciate you taking time to take that. And can't wait to share the results that later this summer. And again, thanks to Drift, our partner on that research.
[00:05:19] Paul Roetzer: Okay, Mike, llama 3 was all the rage last week. Let's start there.
[00:05:24] Mike Kaput: Yeah, let's just say
[00:05:26] Mike Kaput: been a big week for Meta because they announced the release of Llama 3, which is the next version of their foundational open source AI model, and Meta says that Llama 3 outperforms other open source models like Gemma from Google and Mistral. And it also appears on at least certain benchmarks to outperform Gemini Pro 1.
[00:05:48] Mike Kaput: 5 and Claude3 Sonnet. Now today, there are two open source Llama3 models that developers can freely use. There's an 8 billion parameter [00:06:00] model and a 70 billion parameter one. There's also a 405 billion parameter version of Llama3 that is coming soon and is still being trained. It's unclear right now though if the 405 billion version will be open sourced, but it is going to be in the wild at some point soon.
[00:06:22] Mike Kaput: Now, Meta actually told The Verge that Llama 3's training dataset is 7 times the size of Llama 2's. They also mentioned it did not use any meta user data in its training. Notably, the training data does include a mix of what the company says is, quote, public internet data and synthetic AI generated data.
[00:06:48] Mike Kaput: Now, on top of all this, what's exciting is the fact that Llama3 is now going to be powering the The Meta AI assistant, and that assistant is now being integrated across Instagram, [00:07:00] Facebook, WhatsApp, and Messenger. Meta AI as a tool has been kind of out for a while, but Llama3 is going to give it a pretty serious upgrade in capabilities.
[00:07:10] Mike Kaput: And the wider rollout means you're going to be seeing a lot more of it across. These core meta apps, and if you want to go try it out for yourself outside of those platforms, you can go to meta. ai and it looks kind of like a, you know, ChatGPT like interface where you can interact with the meta. ai assistant.
[00:07:30] Mike Kaput: Now it provides outputs just like ChatGPT. It's a general purpose assistant. You can ask questions, have it perform tasks, uh, for you that are, you know, research based. It also allows you to generate images. And, it's pretty noteworthy, Meta. ai can actually search the web for you as you interact with it as well. So Paul, first, I
[00:07:51] Mike Kaput: of want to put this into context for the audience. Like, we've been waiting for some time for Llama 3, and it [00:08:00] seems like it's quite a big deal because it is this extremely powerful open source model. Like, how big a deal is this release?
[00:08:09] Paul Roetzer: it's a really deal, so I've spent the last few days just trying to kind of like process this and And help put it in context for everyone.
[00:08:20] Paul Roetzer: So Mike, interject at any but I'm going to kind of go through some key points here that I think help understand, help everyone understand the significance of
[00:08:31] Paul Roetzer: And as I've said before, history matters here. Like the context of history here is very important to understanding what's going on.
[00:08:38] Paul Roetzer: So first a quick note, I don't, I think this may be what meta used, if not, it's just ironic, but huggingFace, announced on
[00:08:50] Paul Roetzer: April 21st, so that was Sunday, we have just released FineWeb, 15 trillion tokens of high quality web data we filtered and deduplicated, all common crawl [00:09:00] between 2013 and 2024.
[00:09:03] Paul Roetzer: So, I think HuggingFace has either made the training set that meta used, open, or just ironically, also a 15 trillion called FI web. So that's kind a breakthrough will enable the training of models moving forward. So, hugging face made that announcement.
[00:09:19] Paul Roetzer: another quick note just on kind of beginner's guide to lLMs. A parameter refers so we talk about seven, what it? 8B and 7DB, that's the billion parameters. So a parameter refers to a learned variable that the model uses to make predictions and generate, you know, outputs, generate text images.
[00:09:39] Paul Roetzer: Um, generally speaking, the number of parameters, , the more parameters there are, that gets into the size, complexity, and capabilities of the model.
[00:09:48] Paul Roetzer: So in general, more parameters means better performance. capabilities, natural language tasks, knowledge, retention, reasoning, all those things. So as the models get bigger, so we get 405 billion. [00:10:00] I don't think, at
[00:10:01] Paul Roetzer: in least my research again this morning, like I don't think openAI has ever confirmed how many people parameters were in gPT 4, but it's rumored to be anywhere from 500 billion to over a trillion.
[00:10:13] Paul Roetzer: So we don't know. Gemini 1. 5, again, I don't think disclosed the number of parameters, but it's also believed to be in that range. so so that's kind of just basics. So now the big question is why open source?
[00:10:27] Paul Roetzer: Like we've talked about it on show many times, is open source good or bad? Is it, you know, does it create, dangers to society? Um, and so I think like to understand
[00:10:39] Paul Roetzer: why would Meta release something this powerful for free? Like why would they make this available to everyone? When Google's charging for it, OpenAI is charging for it,
[00:10:52] Paul Roetzer: Mistral is playing in the open source world. They're releasing free things. So to actually understand it, you have to go back to the formation of free. [00:11:00] Fair, which is the Facebook AI research lab. So, this story is
[00:11:04] Paul Roetzer: actually told in GeniusMaker. So many times on this show, I've Suggested people read Genius Bakers by Cade Metz. So if you have read it, this will sound familiar you, but I think this sets up,
[00:11:14] Paul Roetzer: We'll go back to 2013 and then we're going to come back to today and I think everyone will understand exactly what's happening here. So you have to go back to 2013 when Zuckerberg
[00:11:25] Paul Roetzer: tried to buy DeepMind. So DeepMind, again, Demis Hassabis, Shane Legg, Mustafa Suleyman, founders
[00:11:32] Paul Roetzer: of DeepMind, 2010, they created a London based research lab. That lab eventually sold to Google in 2014. Well, Facebook tried to buy them. They failed. Um, so here's a couple of excerpts from genius makers that set the stage.
[00:11:48] Paul Roetzer: So in 2013, was a strange idea. The wider tech industry, including most engineers and executives at Facebook, hadn't even heard of deep learning and certainly didn't understand its importance. So deep learning [00:12:00] is basically this
[00:12:00] Paul Roetzer: idea of neural nets, which are, you know, decades old. but jeff Hinton had sort of coined the term deep learning around 2008, 2009, to sort of give a modern twist on this idea of neural nets.
[00:12:12] Paul Roetzer: But it was the idea that AI could have. understanding language, vision, capabilities. Give it a human like abilities is what deep learning enabled. So 2013, again, only 11 years ago, Facebook and Zuckerberg weren't really Paying that much attention to learning. It wasn't really a big thing yet.
[00:12:30] Paul Roetzer: it did not do the kind of research, so Facebook, it's referring to, again back to genius makers, did not do the kind of research DeepMind
[00:12:36] Paul Roetzer: aimed to do, which was more about exploring new frontiers than moving fast and breaking things.
[00:12:41] Paul Roetzer: But now, after growing into one of the world's most powerful companies, Powerful companies. Zuckerberg was on racing the
[00:12:47] Paul Roetzer: others. Google, Microsoft, apple, and Amazon to the next big thing. This is how the industry works. The largest companies are locked in a never ending race toward the next transformative technology, [00:13:00] whatever that might be. be.
[00:13:01] Paul Roetzer: Each is intent on getting there first, and if someone beats them to it, then they are
[00:13:06] Paul Roetzer: under immense pressure to, to get there too without delay. Zuckerberg was intent on bringing deep learning research to Facebook.
[00:13:15] Paul Roetzer: In the end, Zuckerberg and Schroepfer,
[00:13:18] Paul Roetzer: Shep, the CTO, Schroepfer I don't know how to say his name, his
[00:13:23] Paul Roetzer: name, made an unsuccessful bid for DeepMind. Hasabis told colleagues that he felt no chemistry with Zuckerberg. Again, is really, really important context. So Hasabis told colleagues
[00:13:34] Paul Roetzer: that felt no chemistry with Zuckerberg, that he didn't quite understand what the founder was trying to do with DeepMind, and that the lab wouldn't fit Facebook's growth obsessed corporate culture.
[00:13:47] Paul Roetzer: But the bigger issue for Hasabis Legg and Suleyman was that Zuckerberg didn't share their ethical concerns over the rise of intelligence in either the near term or the far. He [00:14:00] refused, being Zuckerberg, to accept
[00:14:02] Paul Roetzer: a contractual clause that guaranteed DeepMind's technology would be overseen by an independent ethics board. Quote,
[00:14:09] Paul Roetzer: we could have made more money if we had just been going for the money, Legg says, Shane Legg, the co founder, but we weren't. So it goes on
[00:14:19] Paul Roetzer: to say, Facebook faced the chicken and egg problem. It
[00:14:21] Paul Roetzer: couldn't attract top researchers because it didn't have a research lab, and it didn't have a research lab because
[00:14:27] Paul Roetzer: it couldn't attract top researchers. So Zuckerberg, based on
[00:14:32] Paul Roetzer: guidance, set his sights on recruiting Yann LeCun to Facebook to run this lab. LeCun initially shut down these overtures. So there's a quote in here, it says, uh, in a phone call with Zuckerberg, LeCun said, I can consult with
[00:14:46] Paul Roetzer: you, but that's about it. He had had similar
[00:14:48] Paul Roetzer: conversations with Schrepp, which is, you know, the short, for the
[00:14:53] Paul Roetzer: name,
[00:14:54] Paul Roetzer: in past, and his stance had always been the same. Zuckerberg, though, kept pushing. Facebook
[00:14:58] Paul Roetzer: was at another dead end. [00:15:00] Schrepp had approached several other leaders in the field, from Andrew o who we know is the founder of Coursera,
[00:15:07] Paul Roetzer: he was the founding lead of Google Brain, and Yashua Bengio, an AI legend, research legend. And still, the company didn't have anyone to run its lab.
[00:15:17] Paul Roetzer: Someone with the heft needed to attract the world's top researchers. So again, still in 2013, he to a dinner with yann LeCun. just the two of them, Zuckert explained his grand vision for AI Facebook in the future.
[00:15:31] Paul Roetzer: and this is really, really important. He told LeCun, interactions on the social network would be driven by technologies powerful enough to perform tasks on their own.
[00:15:41] Paul Roetzer: Sounds lot like AI agents, doesn't it? In the short term, these technologies would identify faces and photos, recognize spoken commands and translate between languages.
[00:15:52] Paul Roetzer: In the longer term, intelligent agents or bots would patrol Facebook's digital world, take instructions, and carry them out [00:16:00] as need be.
[00:16:01] Paul Roetzer: Need an airline reservation? Ask a bot. Order flowers your wife? A bot could do that. When LeCun asked if there were any areas of AI research that would not interest Facebook, Zuckerberg said, Probably robotics, but everything else, everything in the digital domain was
[00:16:16] Paul Roetzer: in bounds. Two other quick excerpts. The bigger issue was how Zuckerberg viewed the philosophy of corporate research. LeCun believed in openness. Concepts, algorithms and techniques openly shared with the wider
[00:16:31] Paul Roetzer: of researchers, not sequestered inside a single company or university. The idea was that this exchange of information accelerated the progress research as a whole.
[00:16:43] Paul Roetzer: Everyone could build on the work of everyone else. Open research was the norm among academics in the field, but not in big internet companies. Facebook, Zuckerberg explained, was
[00:16:54] Paul Roetzer: big, was a big they had built on open source. believed deeply in it. And so basically, [00:17:00] zuckerberg and LeCun agreed to pursue this together, but leCun's stipulations was
[00:17:06] Paul Roetzer: was not leaving New York, he was not leaving New York University, and basically Facebook had to agree to keep everything open.
[00:17:13] Paul Roetzer: Because in LeCun's opinion, that was how you were going to recruit the best researchers. So that's the context of
[00:17:22] Paul Roetzer: why is Meta so committed to open source. One, that is what Zuckerberg committed to LeCun, they were
[00:17:28] Paul Roetzer: and LeCun still runs. at Facebook. Two, it is the foundational belief. The thing that wasn't discussed though in Genius Makers and hasn't been probably as widely discussed is
[00:17:42] Paul Roetzer: the more likely scenario, or a third key component of this is the competitive element. Facebook has, according to their own reports
[00:17:55] Paul Roetzer: reports, most recent, so is 2004, [00:18:00] 2024. they have 3.07 billion monthly active users
[00:18:07] Paul Roetzer: on Facebook. So that doesn't even include WhatsApp and Instagram. I don't think so. That's just Facebook. So we've always talked about. Distribution being key. So
[00:18:16] Paul Roetzer: everything that we're going to talk about with Llama3 and what Facebook is doing is at, to this
[00:18:21] Paul Roetzer: point, open source and free as a user. So there was a great tweet I saw from naveen Rao, who's the of GenAI genAI Databricks. And he said, I don't think everyone has comprehended the massive disruption
[00:18:34] Paul Roetzer: and distortion that is going to happen in the GenAI market due to Llama3. What do you think? Motes will be destroyed and investments will go to zero.
[00:18:43] Paul Roetzer: Just like everything in gen AI, this will all happen fast. So now we kind of come to the Llama announcement. I don't know, Mike, if you have tried this yet, but have you gone to meta. ai yet?
[00:18:55] Mike Kaput: I have,
[00:18:56] Mike Kaput: Yeah,
[00:18:56] Paul Roetzer: it is. Okay. So meta. ai for everybody. If you [00:19:00] haven't tried it yet, go to it.
[00:19:01] Paul Roetzer: have to log into your, through your Facebook account. at least I had I couldn't do anything without it. so released meta. ai with the Llama 3 announcement. The thing that, again, Mike and I mess
[00:19:14] Paul Roetzer: this stuff all day, every day. I was blown away by the auto complete image generation thing. Did you, did you try that yet?
[00:19:22] Mike Kaput: I did, I did. was really, really cool. Yeah.
[00:19:24] Paul Roetzer: if you haven't seen this yet, the way this works is, if think about autocomplete, like in Gmail where it's like finishing your sentences, or when you're in Google search and it's like your search query, they
[00:19:35] Paul Roetzer: are doing that for image generation. So if you start, like, I was just playing around with like, my daughter's favorite thing
[00:19:40] Paul Roetzer: pandas, like her favorite animal. So I was like, image, imagine a panda. And so it creates a panda
[00:19:47] Paul Roetzer: And it's like on the moon and it's on the moon instantly, drinking a glass of scotch,
[00:19:52] Paul Roetzer: it should be playing cards with his friend and it's like now he's got cards front of him, and the friend is a red panda instantly. Like [00:20:00] it's just, and it's so fast.
[00:20:02] Paul Roetzer: And so it's now creating the image as you're typing. It's auto completing the image. It is
[00:20:09] Paul Roetzer: And I assume where they're going with this is like video audio completion
[00:20:13] Paul Roetzer: and like you can start to see the potential of this, but this is also baked right into facebook, Instagram. and I assume WhatsApp. I don't use WhatsApp, but I have tested it in Facebook instagram. And so
[00:20:26] Paul Roetzer: have. Now, your search bar in Facebook and Instagram is basically an Ask Meta A think is what it says. I have seen people complaining like, how do I get rid of this Meta AI thing? What is this thing? So,
[00:20:39] Paul Roetzer: like we've always said, like eventually this stuff trickles into society and now we have billions of with access to meta AI. And, you
[00:20:45] Paul Roetzer: know, your relatives and friends may be like, what in the world is this Meta AI thing? And you can now explain to them what it is. Um. So that is the real is they're
[00:20:56] Paul Roetzer: opening it, but by doing this, it is a [00:21:00] direct attack on OpenAI, Google, Mistral, like, Anthropic, everybody, because what Zuckerberg is basically saying is like, hey, we're going to spend whatever they're spending
[00:21:12] Paul Roetzer: a hundred billion, or eventually they'll spend a
[00:21:14] Paul Roetzer: hundred billion, they're probably spending a couple hundred million right now, maybe the 405B model is going to be
[00:21:20] Paul Roetzer: A half a billion to whatever 750 million dollar, run, a training run. And they're just basically like, yeah, whatever. Like we'll, we'll give it away for free because everybody else is trying to charge for it so they can basically undercut everybody and then they can make it,
[00:21:35] Paul Roetzer: you know, readily not only build on top of, but to build right into their very network. So that was, it
[00:21:44] Paul Roetzer: was amazing how they did it, but to start seeing
[00:21:48] Paul Roetzer: the technology and to seeing their plan to know that while they've been doing AI for 11 years, basically,
[00:21:53] Paul Roetzer: know, it's been key part with the research lab, and, you know, machine learning longer than that.
[00:21:59] Paul Roetzer: They've really been [00:22:00] only playing this Gen AI game since like late 2022. And here they are with Llama 3 that is, you know, apparently on par with at least GPT 3. 5.
[00:22:11] Paul Roetzer: not, you know, probably GPT 4 once the 405B model comes out. So I had one other thought on the Dwarkesh podcast, but I don't know,
[00:22:19] Paul Roetzer: any other thoughts from you, Mike, just like what you're seeing so far and kind of early thoughts on Llama 3?
[00:22:24] Mike Kaput: Yeah, I just think it's worth belaboring the competitive disruption point here. And I think you had surfaced, Paul, another tweet from a user basically saying, Look, by releasing this, Zuckerberg undermines competitors to the point of destruction. It's like if these companies were all Teams in a strategy game.
[00:22:45] Mike Kaput: Zuckerberg just nuked part of the map, I feel like, in turn, and prevented the others from benefiting from it. So, I think that's very fascinating. I think you have to maybe just consider the more cynical or self interested or realpolitik [00:23:00] motivations here if you're trying to understand where defensible moats are, where competitive advantage is going to lay moving forward.
[00:23:06] Paul Roetzer: Yeah, and you hear about these like wartime CEOs, like Zuckerberg's killer, but whether you like the guy or not, he, there is no issues with him doing what has to be done to win. Like he, that is what he has always done, where some of these other. You know, big
[00:23:22] Paul Roetzer: tech companies we talk about maybe just don't have that mindset. So yeah, that matters. So the other thing I'll note is, I suggest if you're really interested in this stuff, go listen to this podcast, but Dwarkesh Patel,
[00:23:35] Paul Roetzer: who we've talked about before on the show, he had one of the two interviews Zuckerberg
[00:23:39] Paul Roetzer: did. So the way they basically dropped dropped Llama3 was they put out a release it
[00:23:43] Paul Roetzer: on their own site and then Zuckerberg did two podcast interviews and then they just put it on social.
[00:23:47] Paul Roetzer: Like that was the launch plan for this entire open source model. So A few key topics that Zuckerberg addressed in this interview with Dworkesh. So one is 405B is in training now, like
[00:24:00] Paul Roetzer: that is actively going and it's already approaching GPT 4 level and the training isn't even done yet. So he said like, Hey, this thing is, It's going to be really, really powerful.
[00:24:10] Paul Roetzer: We're already seeing incredible results on standard
[00:24:13] Paul Roetzer: benchmark testing and we're not even done training it yet. Um, there was an interesting side note where
[00:24:18] Paul Roetzer: talked about, you know, we mentioned a few episodes ago how they were going to have the equivalent of like 650, 000, GPUs for GPUs for NVIDIA by the end
[00:24:27] Paul Roetzer: 2024, I think it was. And so he asked him about that. And apparently they pre bought lot of these things in 2022
[00:24:36] Paul Roetzer: when they were building out the Reels, uh, capability because they were so far behind TikTok. And he
[00:24:42] Paul Roetzer: said, ironically, some of the most important decisions we made were to catch up from previous screw ups. So he said, like, we, we were behind. like TikTok killed us.
[00:24:51] Paul Roetzer: So we had to come out with Reels. And so they envisioned a world where they were going to need do way more compute with Reels. And so bought double the amount
[00:24:58] Paul Roetzer: of GPUs they [00:25:00] had planned. to just prepare for it, which then ended up having those GPUs there ready to be used for AI.
[00:25:06] Paul Roetzer: Kind of interesting. Dworkesh asked him about AGI and like, hey, when did that become kind of part of your focus? And he said it's basically been a progressive
[00:25:15] Paul Roetzer: thing since they started the research lab 10 years ago.
[00:25:19] Paul Roetzer: he said the idea
[00:25:20] Paul Roetzer: was that along the way, to general intelligence whatever you want to call it, there are going to be all these different innovations and that's going to just improve
[00:25:26] Paul Roetzer: everything we do. so we didn't conceive it of as a product. It was more research group over the last 10 years that was infused into their product. But then he went on to say like ChatGPT the diffusion models really accelerated everything for
[00:25:39] Paul Roetzer: them. They then saw the potential. when he got into asking him like, You know, what does he think about when he's building the next versions?
[00:25:46] Paul Roetzer: Llama 4, Llama 5, Llama 6, whatever. There was a couple of things that I thought were interesting. One was, he said, So much of the human brain is
[00:25:54] Paul Roetzer: dedicated to understanding and understanding
[00:25:57] Paul Roetzer: expressions and emotions. I think [00:26:00] that's its own modality. You could say that maybe it's just video or image, but it's clearly a very specific thing.
[00:26:04] Paul Roetzer: Specialized version of two. So he talked a lot about emotions and expressions, which they obviously have
[00:26:10] Paul Roetzer: ton of data to build on, as being a part of where they're going there. And then he talked about assistants and, you know, agents and things that. and
[00:26:20] Paul Roetzer: then they got into Llama 4. And he said, one of the trickiest things in the world to plan around an exponential curve. How long does it keep going? I think it's likely enough that we'll keep going. I think
[00:26:30] Paul Roetzer: it's worth investing the 10 billions, tens of billions, or hundreds of billions in building the infrastructure and assuming that if keeps going,you're going to get some really amazing results.
[00:26:41] Paul Roetzer: So again, we keep hearing this from Sam Altman, for sure. Dario amodei at Anthropic. Anthropic. I don't know about Yann LeCun. He, I don't
[00:26:49] Paul Roetzer: know that he's always on board with this idea of the scaling laws, but now you're hearing from Zuckerberg. They're basically saying like,
[00:26:55] Paul Roetzer: yeah, we're going to invest hundreds of billions of dollars in energy and [00:27:00] data centers and chips because to all of their knowledge,
[00:27:04] Paul Roetzer: the scaling laws appear to be staying true, which means throw more compute.Give it more data, give more training time, and these models just get smarter. So that’s why we're hearing Amazon building data centers next to nuclear power plants. Why, Altman's raising, you know, supposedly
[00:27:20] Paul Roetzer: trillions of dollars for, building out energy infrastructure. So it's all kind of connected.
[00:27:26] Paul Roetzer: And then the last one I wanted to address here is, you know, again, Mike, you and I talk about Is open source good or bad for society all the time? And I'm always kind of shocked at how poorly they
[00:27:40] Paul Roetzer: all are prepared to answer this question. So, Dwarkesh, to his credit, pushed Zuckerberg on topic of the dangers of open source,
[00:27:48] Paul Roetzer: as you start getting into these smarter models.
[00:27:51] Paul Roetzer: And he basically said, They don't have a toxi He said, So
[00:27:55] Paul Roetzer: Dorcas said, when would you not open source something? Like, what would you need to see in a [00:28:00] future model that would stop you from open sourcing it due to concerns? and he said they don't have a taxonomy for that yet. That they do
[00:28:08] Paul Roetzer: in fact, have 18 or 19 categories that they look at for harmful things related to social media.
[00:28:14] Paul Roetzer: that they, theyplanned for, And said, basically, we probably need to do the same
[00:28:18] Paul Roetzer: thing for this. But then
[00:28:20] Paul Roetzer: he very quickly sort of just deflected the whole conversation
[00:28:23] Paul Roetzer: said, yeah, but centralizing AI could be worse. And this is the answer you get from all of them, whether it's Ahmad, formerly at Stability, or him,
[00:28:31] Paul Roetzer: or Dario amodei.
[00:28:32] Paul Roetzer: like, yeah, it could get really, really bad. But
[00:28:36] Paul Roetzer: we think it's better than
[00:28:37] Paul Roetzer: like three key companies deciding our future. So I don't know about you, but like, it just doesn’t well with me. Like I don't feel,
[00:28:46] Paul Roetzer: and we'll talk about the Dario amodei thing in the next episode, because I have a lot of, he just gave a pretty far interview as well, and he had a lot of opinions on this that I have on, but
[00:28:57] Paul Roetzer: I will just say Dario's thoughts [00:29:00] echo what Zuckerberg's saying, which is like, Yeah, it could go really, really bad, but as long as it's not centralized to like
[00:29:06] Paul Roetzer: companies that
[00:29:07] Paul Roetzer: we don't trust, then it's better than the alternative. be interviewed. We'll yeah, so it's, again, go
[00:29:14] Paul Roetzer: listen to the interview.
[00:29:15] Paul Roetzer: I don't know. Do you
[00:29:16] Paul Roetzer: have any other thoughts on this one, Mike? It's, I feel like we could talk Llama3 all day.
[00:29:20] Mike Kaput: oh my gosh, I know we could do a whole episode on it. Now, I think your concerns are well founded with open source. I would suspect we're going to see some type of backlash once some truly damaging application comes on. I mean, I'm sure in the dark corners of the internet it's already being misused, but there will be probably, my guess would be, some high profile examples where it's like, wait, Somebody just downloaded this thing from github or
[00:29:45] Mike Kaput: or wherever
[00:29:46] Mike Kaput: started doing whatever that it is that they've done So I would fully expect yeah, like we've talked about that wider societal potential backlash against AI in general I could see this being a component of
[00:29:57] Mike Kaput: that.
[00:29:57] Paul Roetzer: And again, like, I'm not. For or [00:30:00] against open source. Like, I'm not even taking a stand that I think bad for society. It is truly
[00:30:04] Paul Roetzer: one of these like unanswered things that every time I hear one of these massive advocates for open source talking, I keep waiting to hear the justification that I'm like, yeah, okay,
[00:30:17] Paul Roetzer: Actually I'm, I'm in the open source
[00:30:18] Paul Roetzer: No, I totally get what they're coming from. I understand their concerns around limiting. AI to a few companies. We talked about that
[00:30:26] Paul Roetzer: Altman got fired and we're like, Oh, we're trusting company, this like dumpster fire
[00:30:31] Paul Roetzer: of a company at that moment to, to shepherd an AGI. Like, okay, I kind of understand the need to have this
[00:30:37] Paul Roetzer: So I'm very much in the middle here just kind of like open minded to any conversation. But don't ever
[00:30:45] Paul Roetzer: hear a good answer with the, yeah, but you're open sourcing a really powerful thing that can be used for disinformation and can be used for persuasion and we can't assume everyone's going to be a good actor in this.
[00:30:57]
[00:30:57] Paul Roetzer: Um, they just kind of blow it off and say it's better [00:31:00] than, you know, Letting be controlled by the government or these other entities. So to be continued, I suppose,
[00:31:05] Mike Kaput: which tells me that it's inevitable we're going to get that future where it’s going to be allowed to proliferate however youwant it to.
[00:31:15] Paul Roetzer: I think we're, we're probably we're there. We're going to be there this year. Yeah.
[00:31:18] Mike Kaput: Yeah. Alright, so next up, Microsoft just released some research that details an AI model they created that can deepfake somebody using a single photo.
[00:31:31] Mike Kaput: So this model is called VASA 1, V A S A 1, and it can use a photo to create a realistic video of a person talking or singing that you can sync up with an existing audio track. So Ars Technica had this description. of the model. It says the VASA framework, short for visual effective skills animator, uses machine learning to analyze a static image along with a [00:32:00] speech audio clip.
[00:32:01] Mike Kaput: It is then able to generate a realistic video with precise facial expressions, head movements, and lip syncing to the audio. Right now, This model, which is only in research mode at the moment, does not clone or simulate voices, though obviously there's plenty of tools that do that. Interestingly, this model they say right in the reporting was trained on YouTube clips because it used a data set created years ago by researchers that has a bunch of data extracted from videos uploaded to YouTube.
[00:32:32] Mike Kaput: So given the model’s model's
[00:32:33] Mike Kaput: speed and power, this could actually potentially be used in real time applications like video conferencing, or it could power avatars that just render locally on your machine. Now, Microsoft's researchers, wary of the ways this technology could be misused, say they are not openly releasing the code that powers the model at this time. So, Paul, while Microsoft isn't really releasing the code here, [00:33:00] They're not the only AI lab working on this type of thing. we wanted to talk about it just because it is such a stark example of how quickly this technology has progressed, but I guess my question is, we basically, don't we just have to accept now that anyone can be deepfaked from a single image, whether or not Microsoft releases this particular version of this technology?
[00:33:22] Paul Roetzer: It's a recurring topic,
[00:33:25] Paul Roetzer: or theme on this podcast of like, we're just in a different place. And most of society is on blissfully unaware that this happens. So when I saw this went
[00:33:36] Paul Roetzer: back to our book, my marketing artificial intelligence. And so we released that in summer, 2022, and I'll just read a quick excerpt from it.
[00:33:44] Paul Roetzer: So we were in a, in a. A chapter we were talking about vision, AI vision technology and its implications. there was a section there called the danger of deepfakes to your brand. And we wrote vision can also be applied produce
[00:33:57] Paul Roetzer: deepfake videos in which a person in an [00:34:00] existing image or video is replaced
[00:34:01] Paul Roetzer: someone else's likeness. the prevalence and impact of deepfake videos is just beginning and having an understanding of the underlying technology will help you prepare your brand for its potential. Now, in this
[00:34:11] Paul Roetzer: we're talking about an image of a person that you just turned into a video. went on to say, this was, you know, me talking, I did a fair amount of crisis communications planning early in my career.
[00:34:21] Paul Roetzer: Basically, you envision different scenarios of what could go wrong, then put strategies in place for how organizations react. Then you hope of
[00:34:28] Paul Roetzer: it actually happens. Never did I imagine a day in which brands would be planning for deep fake videos of executives doing insane things never happen in real life. But here we are. But the
[00:34:39] Paul Roetzer: interesting part is, then went on to say AI has made it possible and relatively easy with the right resources
[00:34:44] Paul Roetzer: to create fake videos of people that and sound real.
[00:34:48] Paul Roetzer: According to the Department
[00:34:49] Paul Roetzer: of Defense, one of their people who is developing software to detect prevent the spread of deepfakes, this was a quote, it only takes about 500 images or [00:35:00] 10 seconds of video to create a realistic deepfake.
[00:35:03] Paul Roetzer: So that was 2022. You needed. 500 images or 10 seconds of video. Here we are with a single image and a single audio clip and we have deep
[00:35:14] Paul Roetzer: technology. So in a two year span, we have gone from the state of the art,the Department of defense probably had the most powerful stuff at
[00:35:23] Paul Roetzer: time, to, to this.
[00:35:25] Paul Roetzer: So yeah,
[00:35:26] Paul Roetzer: it's this recurring theme of. Oh yeah, Breakthrough happened, Sora from OpenAI as an example, we can now create minute long videos that maintain their consistency, or we can generate videos from a single image, whatever it is, And
[00:35:38] Paul Roetzer: this, we're not releasing it because, well, going back to the open source conversation, if they're doing it, then
[00:35:44] Paul Roetzer: it's either already in other labs, or it will be within three to six months, and there is going to be plenty of open source models who have no problem releasing because
[00:35:53] Paul Roetzer: they believe you release everything, and you put it in the hands and democratize it,
[00:35:57] Paul Roetzer: no matter how dangerous it [00:36:00] is. So if we hear about this from Microsoft, from OpenAI or whomever, assume within six
[00:36:06] Paul Roetzer: months, it will be open sourced by someone, and this stuff will be all over the web.
[00:36:12] Mike Kaput: And I think it's also worth noting here, there's just this inherent tension between the people creating the technology but also trying to release it responsibly. Not at all knocking Microsoft, I'm glad they're thinking about how it
[00:36:27] Mike Kaput: be misused, but like we've discussed, you get into this cycle or this game, this arms race, where it's like Even if they say they're not releasing it today, all we need is for someone else to make a breakthrough or to open source it or to Open AI says guess what?
[00:36:43] Mike Kaput: There's Sora free plans for everybody or whatever. Pick your category of AI and suddenly Microsoft's like Oh, yeah, we can't not commercialize this anymore
[00:36:54] Paul Roetzer: Yeah, you're right. just takes one person to put it out in the world and it's like, oh, okay. Well, somebody else it. Now we can go and see what happened with [00:37:00] ChatGPT.
[00:37:00] Mike Kaput: Yes.
[00:37:01] Paul Roetzer: Google had the technology. They weren't willing to put that kind of technology out into the world.
[00:37:05] Paul Roetzer: OpenAI had less to lose, so they did it. And then everybody else had to follow suit. So yeah, it's like,Ah, man.t's going to get so weird.
[00:37:13] Mike Kaput: Yeah, yeah, this is probably another area where I know we've seen some pretty crazy deepfakes, but I feel like we're, we've got at least a couple apocalyptic examples that are gonna happen soon, that people are gonna be like, wait a second, what? You can do what?
[00:37:28] Mike Kaput: Yeah.
[00:37:29] Mike Kaput: Yeah. All right, so our third big topic today is pretty interesting commentary around a possible AI powered paradigm shift that some people are starting to see in the SaaS space, software as a service space, and this is getting a fair amount of attention online, so we wanted to kind of unpack it.
[00:37:49] Mike Kaput: This concept is called Service as software, and it's outlined in an essay from Foundation Capital, which is a VC firm that's invested in some [00:38:00] notable companies like Netflix and some AI firms like Jasper. Now, Foundational Capital defines this idea of service as software as this paradigm shift that basically entails the fact that AI is causing a transition from software as a service to To service as software and what this means is, you know, in the past you bought access to a software tool or a platform.
[00:38:26] Mike Kaput: This is how SAS works today, how it worked before. You use this software to achieve your desired outcomes. in your work. However, they argue that AI is changing this because it's not just giving you access to tools that make you more productive. It is giving you access to bots that essentially act as brains.
[00:38:47] Mike Kaput: In other words, when you buy an AI tool, you're buying access to technology that can increasingly perform some of the work, the tasks, the services for you instead of giving
[00:38:58] Mike Kaput: you the ability to[00:39:00]
[00:39:00] Mike Kaput: to do the service better, faster, cheaper on your own. So like, as an example, instead of buying QuickBooks in order to make your finance team or an outside firm better at maintaining your books, you may in the future simply be buying access to an AI accountant that does the job of that finance professional or does the job of the outside firm.
[00:39:23] Mike Kaput: And this is a change that just has massive implications for both jobs and businesses. SAS companies, according to Foundation Capital. So if you think of something like the basic value proposition of software like Salesforce, a human rep enters the data, Salesforce helps them do their job much more productively, but suddenly with AI, and especially with AI agents, you unlock capabilities and autonomous actions that far more resemble a skilled human performing a service, not just some piece of software that a person uses.
[00:39:58] Mike Kaput: So, the whole [00:40:00] point here, they argue, is that the size of the opportunity for AI disruption of services, of jobs, of roles, of tasks, is far bigger than the opportunity that exists to disrupt existing SaaS tools. And so they ask, how much work will AI plus automation and the service as software model do away with?
[00:40:22] Mike Kaput: We believe this is a 4. 6 trillion dollar question. This is a number they get to. by looking at salaries of jobs globally and the amount spent on outsourced services and salaries. So, Paul, there's a lot to unpack behind this concept, but I kind of wanted to just start by talking about this thesis seems directionally correct to me, based on kind of what we know and teach.
[00:40:48] Mike Kaput: Like, in my mind, we have to acknowledge we're no longer just building static tools that make employees more productive. That's part of it. Part
[00:40:55] Mike Kaput: of what's being
[00:40:56] Mike Kaput: built in the world of AI, but we're also building [00:41:00] technology that fundamentally alters how work is done and Determines which work gets done by a human.
[00:41:06] Mike Kaput: Does that map at all to kind of how you're thinking about this?
[00:41:10] Paul Roetzer: Yeah, and I think it's a really delicate dance these
[00:41:13] Paul Roetzer: software companies. So I had to laugh. Like the first time I saw this service as software quote, I was is that meaning the software companies building AI are having to sell services because they're
[00:41:24] Paul Roetzer: making enough money? Hmm. That was what I assumed it
[00:41:26] Mike Kaput: I Thought that too. Yeah
[00:41:27] Paul Roetzer: and then I saw like Brian Halligan tweeted it and I was like, Halligan wouldn’t tweeting that if that's what meant, and so I like, started digging in a little bit more.
[00:41:36] Paul Roetzer: Um, so, yeah, think, like, the Delicate dance I'm referring to is, So again, if you
[00:41:45] Paul Roetzer: listen to the show regularly, you've heard me say, like, my agency, PR2020, that I sold in 2020 was HubSpot's first partner back in 2007.
[00:41:53] Paul Roetzer: So, we built Our entire agency around being a value added reseller and partner of [00:42:00] HubSpot's and so HubSpot customers that didn't want to do the work would come to my firm and we would build plans. We
[00:42:05] Paul Roetzer: create content. That's where Mike, Mike and I work together at PR And so that was what we did.
[00:42:12] Paul Roetzer: We, we added value to HubSpot software by providing services to their customers who did not have the internal capabilities to do these things. And
[00:42:24] Paul Roetzer: so now we're entering into this phase where whether it's QuickBooks or Salesforce or HubSpot or take your pick of SAS tool, The
[00:42:32] Paul Roetzer: The SAS companies themselves, can be thinking of their software as, part part of the org chart in essence, as outside
[00:42:43] Paul Roetzer: they're not just selling the software, the tools do the job. They're, they're selling you an agent that
[00:42:47] Paul Roetzer: do the job. So you
[00:42:48] Paul Roetzer: don't need an outside agency, in, in production. So this certainly aligns what we
[00:42:53] Paul Roetzer: talked about last week on episode 92, with Google positioning their AI agents [00:43:00] as customer agents, employee agents, creative agents, data agents, code agents, security agents.
[00:43:05] Paul Roetzer: They're positioning these, these AI tools that can do that can take actions. In essence, as people, they're, they're, they're, they're able to do a bundle of which a
[00:43:16] Paul Roetzer: job is basically just a bundle of tasks. I found myself this weekend, revisiting the GPT's, our paper from
[00:43:23] Paul Roetzer: OpenAI, which they released when GPT 4 came out in 2023.
[00:43:28] Paul Roetzer: And that's the basic premise that these things are general purpose that are capable of, evolving. They're capable of being built on top of, and they're able to do that humans would do.
[00:43:39] Paul Roetzer: So. , uh, I, again, I think this is the thing that people just aren't talking enough about that as we follow these scaling laws we touched on earlier, that Zuckerberg believes in, and Altman believes in, and Anthropic believes in, like they all are believing in these scaling laws.
[00:43:57] Paul Roetzer: And if these laws hold, we will in the [00:44:00] not too distant future have relatively reliable agents that can function doing many of the tasks that would make up a job.
[00:44:09] Paul Roetzer: And again, I've said this before, like, I just don't hear from large enterprises who are addressing this, who are actually thinking about what means, and preparing for it.
[00:44:23] Paul Roetzer: And for agencies, like we talked couple episodes agoabout how much Accenture's making, 640 million or whatever in a quarter, just doing AI consulting work. You do have to look out to
[00:44:33] Paul Roetzer: future and say, well, that could keep scaling. They could be a 10, you know, billion or whatever, or, Or two years from now the agents are just going to do most
[00:44:41] Paul Roetzer: the stuff that they're making consulting money on. I don't know. Like I don't, I don't know where it goes, but this is definitely, as you kind of directionally
[00:44:51] Paul Roetzer: we're seeing is it's going to be very, Narrow to start. You're not gonna have this general agent that I just go and get one agent from HubSpot or Salesforce
[00:44:59] Paul Roetzer: [00:45:00] Adobe or or whatever. And it just does the job of all of my people, but you're going to have
[00:45:04] Paul Roetzer: tasks and maybe they're like, like think like of it
[00:45:07] Paul Roetzer: as like AI roles, like customer service, BDR, marketer, podcast producer, social media manager, whatever it is.
[00:45:15] Paul Roetzer: That you're training these agents to be very specifically tuned to do
[00:45:18] Paul Roetzer: a bundle of 10 or 15 or 20 tasks. And now all of a sudden, it becomes very disruptive to jobs. Um, right now we have distinct tasks that are able to be automated. In the not too distant we will have bundles tasks, and that's what a job is.
[00:45:37] Mike Kaput: I think also what jumped out at me is in this post with, from Foundation Capital, they're kind of trying to quantify, right, with this 4. 6 trillion question, they're trying to quantify. What is the size of the opportunity in terms of salaries being paid for different roles? And we talk about quite often, we've talked about since way back when, when we started reading the quants.
[00:45:59] Mike Kaput: It's like, what is [00:46:00] the opportunity to disrupt things? I think the equation is something to the effect of the technology being able to do it plus the size of the opportunity. So when I look at a chart like this, I see a ton of very hungry AI students. Startups that are going to start coming for salaried work, perhaps.
[00:46:17] Mike Kaput: I mean, I don't think it'll be that, like you said, all at once. But the incentives here give me some pause to think about.
[00:46:25] Paul Roetzer: Yeah. and I like the quants was one
[00:46:27] Paul Roetzer: book that you and I both read in the early days, and the other one that, you know, really started my path down AI was Automate This by Steiner.
[00:46:34] Paul Roetzer: I think that's where we originally saw that quote was like the potential to disrupt plus the reward for disruption. And that was actually
[00:46:43] Paul Roetzer: quote that led me to say, well, hold on a second. This is 2012. 2012. AI is going to come for marketing and sales and
[00:46:49] Paul Roetzer: It's not here yet, but like the potential to disrupt is massive. And what you're saying is the potential disrupt the service industry
[00:46:56] Paul Roetzer: is way greater than even the software industry. [00:47:00] And so it would make a ton of sense that that's where these research labs and tech companies
[00:47:04] Paul Roetzer: would be focusing on. It's like, Hey, if we can, coding's great. Like writing code is really helpful. But if we can actually augment or replace work, now they're not going to call it that. No one is going to say our job is to replace workers. Like, it's not going to be anybody's tagline.
[00:47:21] Paul Roetzer: They, that is what they are all thinking. Like, I cannot tell you how many times a month I hear this when I'm meeting with or talking to to people at companies. The thing no one's saying, but thing we’re being directed to is, do we need as many humans in the future?
[00:47:37] Paul Roetzer: It, again, it's not going to be an earnings calls, like, they're not going to talk about this, it is the thing they are all talking about, if that makes sense.
[00:47:46] Mike Kaput: Yep. Yep. That's exactly what I'm getting at. I think that's wise for people to keep in mind as we move forward and see what happens here.
[00:47:56] Mike Kaput: Alright, let's jump into some rapid fire topics. So, we [00:48:00] got some huge news this week related to AI for video editing. So Adobe just announced that they plan to add AI video generators, including Sora, Runway, and Pica to Premiere Pro, which is one of the world's most popular video editing programs.
[00:48:17] Mike Kaput: So in the announcement, Adobe said that OpenAI's Sora, RunwayML's Gen2, and Pica 1. 0 will would be made available via plugins to Premiere Pro users. Now right now there's no timetable for when this becomes available and it's still a bit unclear how this would work exactly given that some of the tools require paid subscriptions, Sora is not even publicly available yet, so we have to kind of see on that.
[00:48:43] Mike Kaput: However, the news has the video community. Pretty excited. Um, especially given how popular Adobe products are. I mean, there's 33 million paying Creative Cloud customers right now. So I thought one point jumped out at me from an AI [00:49:00] creator named Bilawal Sidhu who shared one reason that he's really excited about it.
[00:49:05] Mike Kaput: He said this is quote, amazing for creatives because to do anything compelling with AI video generation models, you need to bring them into a video editing tool. So Paul, that stuck with me because it seems like AI adoption has the potential to skyrocket the moment some of these powerful tools and capabilities get baked right into the stuff like Adobe.
[00:49:27] Mike Kaput: that people are using every day to do their existing work. So, what were your thoughts on hearing this announcement?
[00:49:34] Paul Roetzer: One, I would say, go watch the video. It was pretty, pretty impressive. Like about a minute and a half of how they're envisioning working where you're already working in Adobe pro and you a thing and you drop down, you want to extend a scene or you want to edit a scene and you pick the
[00:49:47] Paul Roetzer: model you're going to use. So it, My first reaction was, it seems like it functions very similar to Perplexity, where I pick what model I want to work with. AWS I think is envisioning what doing,
[00:49:57] Paul Roetzer: how Google, like they're all kind of making where,
[00:49:59] Paul Roetzer: yeah, we have [00:50:00] our own models, like Adobe, we got, we got our model, but if you're a Sora user or you want to have capabilities
[00:50:05] Paul Roetzer: PICA, like whatever, like it's all, you know,
[00:50:08] Paul Roetzer: kind of built right in.
[00:50:09] Paul Roetzer: So my first thought was, For these tools, great distribution. I don't know how many Adobe users there are. I tried to look up.
[00:50:16] Paul Roetzer: I couldn't find a, like a firm number on how many Adobe premier users there
[00:50:20] Paul Roetzer: are, but I would guess it's hundreds of thousands, if so it's a quick way for Adobe to level up. So for these tools, great distribution, you know, they get their tools in front of a bunch of people.
[00:50:32]
[00:50:32] Paul Roetzer: for Adobe. Quick way to level up and offer these capabilities. people
[00:50:36] Paul Roetzer: maybe stay within their platform. Makes, makes a lot of sense. Definitely seems to run counter to Adobe's. Current or previous position on, the licensing of training data.
[00:50:48] Paul Roetzer: If they're basically saying, Hey, yeah, we're being selective about our data, but everybody else just come
[00:50:53] Paul Roetzer: on in. We don't really care how you trained your models. So it seems like we're just gradually moving to the point where these companies are all just like, yeah, we[00:51:00] care. Like, we’re just going to keep using this training data until somebody tells you it's illegal.
[00:51:04] Paul Roetzer: And they're just going to gradually, you'll just all of a sudden these, uh, positions they took previously on the legitimacy of training data are just going to disappear t seems very obvious. That's where this going. is going. seems
[00:51:16] Paul Roetzer: really disruptive. Like if we go back to the previous conversation around AI agents, again, the marketing and positioning is all, you
[00:51:24] Paul Roetzer: sunshine and rainbows, like this great and creatives love it.
[00:51:27] Paul Roetzer: The other alternative, though, is like, people's jobs, like, are involved in this, like if you're able to
[00:51:33] Paul Roetzer: all of a generate B roll without having to hire a film crew and actors and, like, people make money doing what they're now gonna automate with a drop down model. I'm not, again, not saying it's good or bad, I'm just saying we should talk about the fact
[00:51:47] Paul Roetzer: this isn't all good. Like, is Every announcement is always like It’s just great and people love it. No, the people you're featuring in your news love it.
[00:51:58] Paul Roetzer: There's a whole bunch of people who [00:52:00] don't and I'm not saying we should slow anything down or it shouldn't happen. I'm just saying we should be more transparent about the negative impacts that this is gonna have on people.
[00:52:08] Mike Kaput: To your point in this segment and also the previous one, I do read, and I've read this phrase multiple times for certain video editing tools. Anytime someone's saying you can do all this great stuff without filming additional footage, that feels like code to me.
[00:52:25] Paul Roetzer: It
[00:52:26] Mike Kaput: is definitely You know, I don't know enough about video editing and production.
[00:52:30] Mike Kaput: Obviously, there's gonna be new opportunities created. But yeah, that reads like code to me, like We're trying to say something else here.
[00:52:37] Paul Roetzer: Yeah, and all those extras in Hollywood that you got to sign off on their digital likeness Yeah, just recreate them in the
[00:52:42] Paul Roetzer: Like, you don't need to go bring in the extras Just you know take 50 extras from previous movies that you have owned the digital likeness for and just generate them using this tool and The stuff
[00:52:53] Paul Roetzer: isn't being said. Again, there's, and I think that's hopefully what one of the things take away from podcast
[00:52:58] Paul Roetzer: all the time is there is a [00:53:00] whole separate thing happening here that isn't talked about and
[00:53:03] Paul Roetzer: I think it's really important that people understand that and that we start talking about because it's the only way for us to be ready for the change that's happening.
[00:53:13] Mike Kaput: Yeah, for sure. I, you know, I'm giggling here, but like, I take it very seriously. I think we just need to have the discussion because you have to go in with your eyes as wide open as possible to the disruption that's coming.
[00:53:26] Mike Kaput: Alright, so next up, robotics company Boston Dynamics is turning heads with a new robot.
[00:53:33] Mike Kaput: They announced that they're transitioning to a fully electric version of their Atlas robot, which is a humanoid robot designed to perform tasks and labor in the real world. As part of this announcement, the company released this About 40 second video that shows the all electric Atlas go from lying on the ground to rising to its feet and walking around with surprising agility.
[00:53:57] Mike Kaput: Boston Dynamics says they're [00:54:00] partnering with a small group of customers. to test and iterate applications for Atlas over the next few years. And the first customer they're doing this with is Hyundai, the car maker. On this partnership, the company
[00:54:13] Mike Kaput: said, quote, in the months and years ahead, we're excited to show what the world's most dynamic humanoid robot can really do in the lab, in the factory.
[00:54:22] Mike Kaput: And in our lives. I thought it was also interesting they gave a nod to how quickly the field of robotics is moving, which we discussed at length, Paul, when you outlined your timeline of AI innovation in episode 87. Boston Dynamics said, quote, a decade ago, we were one of the only companies putting real R& D effort into humanoid robots.
[00:54:43] Mike Kaput: Now the landscape in the robotics industry is very different. Paul, what did you think of the Atlas announcement? How does this kind of play into what you see as the timeline for robotics? It's
[00:54:58] Paul Roetzer: is on track. I wouldn't [00:55:00] update my timeline currently. I think we are continually moving, if anything, over time.
[00:55:04] Paul Roetzer: I think some of the things I said in that timeline will just shorten, but I think this shows we're aggressively heading in a direction where, uh, humanoid robots are
[00:55:11] Paul Roetzer: going to be key. Uh, second thing is watch the video. It's creepy as hell, man. It's so weird. If you haven't seen it’s laying flat all folded up and then the head pops up and turns and then the body can twi It's really
[00:55:26] Paul Roetzer: It’s like sci very sci fi like. third thing is, I wonder if Google's gonna regret selling Boston Dynamics. So, So, for people who don't know, Google owned Boston They sold them to SoftBank
[00:55:39] Paul Roetzer: 2017. So, 2017, when year the transformer was invented, we didn't have the breakthroughs in intelligence we're seeing now. You That we talked about a few episodes ago, how figure and OpenAI are now working to embody
[00:55:51] Paul Roetzer: models in the robot. Well, Google had the leading
[00:55:54] Paul Roetzer: robots at the time, and in theory, they could be now dropping Gemini into robotics, [00:56:00] but here we are, you know, however many years later, and it looks like Hyundai actually owns the company.
[00:56:04] Paul Roetzer: Boston Dynamics now, which I did not know,
[00:56:07] Paul Roetzer: Uh, they completed the acquisition in June of 2021 from SoftBank. So yeah, I know Google could have been a major player, but at the time we just had, you know, robot dogs racing around and robots doing flips on things, and now we have embodied intelligence and robots.
[00:56:24] Mike Kaput: Alright, so next up, the popular AI avatar generator, a tool called HeyGen, just released a demo of an incredible new feature that allows you to create user generated content, UGC, using an AI spokesperson. So all you need is a URL of a product page on Amazon, Shopify, etc. And HeyGen will use it to create for you a video of a hyper realistic AI avatar Promoting your product or service.
[00:56:52] Mike Kaput: Now, as part of this process, HeyGen will generate a script based on details that you input and then allow you to select [00:57:00] which avatar you want to use and what format you want your video in. So, within a couple minutes, you can literally generate a full video with an AI avatar that is reading your script.
[00:57:12] Mike Kaput: Now, Paul, this is obviously a
[00:57:14] Mike Kaput: product. HeyGent's releasing it as a demo right now. It has bugs, it's not perfect. But I was pretty impressed with just how good some of this looks. Like, should we be expecting an influx of AI generated promotional content online? Like AI influencers influencers and ad spokesmen?
[00:57:30] Paul Roetzer: Yeah, I, I feel like this kind technology, one, it's not only HN is going to have it. Everybody's going to have this. I feel
[00:57:38] Paul Roetzer: like by this time in 2025, this stuff's just going to be everywhere. Uh, you could definitely see early adoption, like SaaS
[00:57:46] Paul Roetzer: industry, like the early adopting industries are going to be using this stuff crazy. I don't think your,
[00:57:53] Paul Roetzer: Your average bank or hospital or manufacturing company is necessarily going to be. Yeah. powering their sales and marketing with [00:58:00] this stuff, but you could see this take off and
[00:58:03] Paul Roetzer: of those early adopters. and I think could become pretty prevalent, which I'm not necessarily a fan of, I'm a bystander uh, commenting on it it doesn't matter
[00:58:14] Paul Roetzer: if I like it or not. I feel like this is the kind of technology that will just snowball at some point.
[00:58:19] Mike Kaput: Alright, we also got a new interview with Sam Altman that just dropped courtesy of the popular podcast 20VC. In it, VC and host Harry Stebbings interviews Altman and OpenAI COO Brad Lightcap about a range of topics, from OpenAI's operational challenges and opportunities, to how to invest and operate in a world where OpenAI appears to be dominating.
[00:58:43] Mike Kaput: Parts of the AI arms race. So Paul, we both listened to this. You had, a couple of things jumped out to you. It sounds like about the interview.
[00:58:50] Paul Roetzer: Yeah, it, unlike the Zuckerberg interview, I wouldn't say there was as many like noteworthy things. It was just kind of a lot of like
[00:58:57] Paul Roetzer: and forth and, but some interesting stuff. [00:59:00] But the one quote that I will say it doesn't have quite shock value of the episode 86 and 87 quote that we talked about, but I will read this excerpt anyway.
[00:59:11] Paul Roetzer: So when being asked about GPT 5 and, you know, what of companies. could exist, What
[00:59:17] Paul Roetzer: of moats exist, he said, and I quote, Here are two strategies, or there are two strategies to build on AI right now. There's one strategy, which is to assume the model is not going to get better. And then you kind of like build
[00:59:32] Paul Roetzer: all these little things on top of it. there's an, so basically we would call it wrappers. then there's another strategy, which is to build assuming that That open AI is going to stay at the same rate of trajectory, and the models are going to keep getting better at the same pace.
[00:59:49] Paul Roetzer: It would seem to me that 95 percent of the world should be betting on the latter category, but a lot of startups have been built in the former category.
[00:59:58] Paul Roetzer: When we just do our [01:00:00] fundamental job, because we have a mission, we're going to steamroll you. GPT 5 will be on par with the leap from three to four. Hmm. So again, all indications have been
[01:00:12] Paul Roetzer: and continue to be That GPT 5 is going to be a massive leap forward that is really hard to project right now and to comprehend. But Sam talks with an
[01:00:24] Paul Roetzer: awful lot of confidence of someone who has probably seen what is likely going to be possible.
[01:00:31] Paul Roetzer: And I feel like he is increasingly being vocal to prepare people for the amount of disruption that is coming.Again, Sam's history isn't to say things until he's ready. and confident about what
[01:00:46] Paul Roetzer: has to say. and he is being increasingly direct about the need for people comprehend how much change is coming.
[01:00:55] Mike Kaput: Alright, this next one's not nearly as weighty or [01:01:00] impactful, but it's near and dear to our hearts, Paul. A new song has just been released by Drake that's getting attention for its use of AI, and that's because the song features full AI generated verses from both Snoop Dogg and Tupac. So, on this track, you're literally hearing highly realistic voice clones of both artists rapping, and then Drake himself wraps up the song with a verse.
[01:01:24] Mike Kaput: Now, I don't really know if I have too much to say other than this was just wild to me to hear as a fan of these artists. I mean, what was your reaction when you heard this?
[01:01:33] Paul Roetzer: You know what, I think I spent more time to understand why the diss track being released. There’s a lot of
[01:01:39] Paul Roetzer: Rick Ross had to do with any of this. Like, I
[01:01:41] Paul Roetzer: was watching literal, like, videos, like, dissecting how all the people had all these beef with people. my first reaction was, is that illegal? Like, that's
[01:01:49] Paul Roetzer: literally my first reaction was, does Snoop Dogg, like, know this happened? And I'd actually went Snoop Dogg’s Twitter account and he posted video at like 2 a. m. that was kind of playing it [01:02:00] off as though maybe he knew this happening. Tupac
[01:02:03] Paul Roetzer: obviously release a video saying if he knew it was happening. So, yeah, I don’t know, but I did see a lot of tweets. Like, did we just open floodgate? Like, has Drake basically just, ushered
[01:02:14] Paul Roetzer: in the where, you just do whatever you want AI and it's just the accepted norm. And , I feel
[01:02:22] Paul Roetzer: like that also is kind of very near where people just do whatever they want with AI and apologize later, basically.
[01:02:31] Mike Kaput: Yeah, we talk a lot about how the AI arms race between big tech companies is pushing the envelope, and I don't think we spend enough time thinking about how beef betweenrappers might
[01:02:41] Mike Kaput: just, like, throw the rules out the window, you
[01:02:43] Paul Roetzer: And the, the relevance of Taylor Swift to it all because apparently it had something to do with her dropping her album so he waited a day to drop it. I don't know, like it was, I had to like dive into pop culture and hip culture a couple days last week figure out what in the world was going on with everybody.
[01:02:59] Mike Kaput: [01:03:00] Alright, couple final topics here. A company called MicroOne has just released what it's calling, quote, the first ever AI interviewer. It's called GPT vetting, and it's an AI tool that automatically interviews candidates for software engineering roles. So what you do is you tell the tool what skills you want the interview to focus on, and it automatically asks candidates verbal questions.
[01:03:23] Mike Kaput: and conducts coding exercises with them. The entire process, it sounds like, takes about 20 to 30 minutes. You then review a report with an AI assessment of the person's skills, a trust score about how valuable they think, the employee would be, and the company claims that in beta, this tool has already conducted 13, 000 AI interviews, saving 10, 000 hours that would otherwise be spent conducting technical interviews.
[01:03:51] Mike Kaput: Now, The publicly available examples of this are all focused from what I could find on software engineering for hiring coders, but the company's website [01:04:00] also claims the tool can, or will, be able to be used for other interviews, like for project managers, designers, etc. Salespeople and more. So, Paul, like any of the tools we talk about, we have to emphasize this is still really early.
[01:04:14] Mike Kaput: We're going off just public claims made by the company. So take everything with a grain of salt. However, if this works as advertised or could sometime soon, it seems to indicate the world of HR is about to
[01:04:28] Mike Kaput: change. What do you think?
[01:04:31] Paul Roetzer: Yeah, not just HR, I mean, ops, finance, sales, service, everything. So,
[01:04:39] Paul Roetzer: you know, it's funny, I, you brought up the quote, I go back it, the opportunity to disrupt plus the reward for disruption.
[01:04:45] Paul Roetzer: The automated, automate this premise.
[01:04:49] Paul Roetzer: to think about this is there are only so many people in the world capable of building AI tools right now. now that's being democratized, so like you and I can build our own AI tools a year or two from [01:05:00] now when we have an idea. But for right now, there's only so many people capable of building it.
[01:05:04] Paul Roetzer: So, you look across industries and you say, which industry is most ripe disruption from generative AI technology? and then you prioritize those and you go after them. Or, you're a domain expert in HR and you say, I gotta go find somebody to build a better way to do this, and you go find somebody willing to do it.
[01:05:21] Paul Roetzer: But those, those people have lots of options of what to be building right now and where they can make the most, the most money. so so I feel like this is one of
[01:05:28] Paul Roetzer: This is a very viable thing. I am again, not commenting on whether I think this is actually good idea or not. It seems very viable, but I will say again,
[01:05:36] Paul Roetzer: the allure to do these things is going to be. Immense. When you have people who understand what these things are capable of doing, and they look at a challenge, like how many
[01:05:46] Paul Roetzer: hours, how many hundreds of thousands of hours are spent interviewing people every year, and they say, well, hold on a second, there's nothing they're doing that the
[01:05:55] Paul Roetzer: AI can’t do. And maybe even better when you figure out they can read emotions for videos, they can analyze people's writings. They can under make, you know, make predictions about the behaviors based on their writings.
[01:06:06] Paul Roetzer: You start realize that AI is actually quite human level or beyond at a lot of the things that we do and it just takes one person or one company to build that thing and then make it available to a lot of other people and disruption all of a sudden just happens.
[01:06:25] Paul Roetzer: So yeah, this one makes a ton of sense. I looked at it a little bit, seems pretty good. like a viable use case.
[01:06:31] Paul Roetzer: Not saying it's good or bad, but it is. And I think that’s more and more the reality of like, oftentimes we’re commenting on here is, listen, this is just the reality. Like it or don't like it.
[01:06:42] Paul Roetzer: You just got to figure out how to live with it and what to do about it
[01:06:46] Paul Roetzer: Because we can't stop this sort of stuff from coming. again, whether or not it should be used in some of these instances is up for debate.
[01:06:55] Mike Kaput: This actually segues perfectly into kind of our final topic [01:07:00] here, because Paul, you had wrote this week about a concept that we're kind of starting to see more of in business and marketing, and it's AI fatigue. And you wrote, I've been hearing a lot lately about businesses, and. Business and marketing leaders who have AI fatigue.
[01:07:15] Mike Kaput: They're overwhelmed by existing responsibilities and goals and see AI as one more thing they have to worry about and learn, which may lead many to delegate AI understanding and adoption in their companies. I think this is the wrong approach. Could you walk us through what you mean here?
[01:07:30] Paul Roetzer: Yeah, so the analogy that I used in there is like, if we go back to 1994 and the internet is You know, getting ready to, toreally burst onto scene and change the future
[01:07:41] Paul Roetzer: of society and business and communications and relationships and everything, like truly just transformed everything. there
[01:07:49] Paul Roetzer: are a lot of people in this world who feel like AI will be 10 to 100 times as transformative
[01:07:56] Paul Roetzer: the internet when all is said and done over the next 5 to 10 years. [01:08:00] If, if that is true, or even remotely close true, let's just say on par with the internet, for argument's sake,
[01:08:07] Paul Roetzer: If you're a CEO, CMO, VP, director, like
[01:08:11] Paul Roetzer: whatever, if you're a leader in a company, Should you
[01:08:14] Paul Roetzer: really be Letting other people figure this out and you just stand on the sidelines and read the reports about what to do with it.
[01:08:23] Paul Roetzer: My argument would be absolutely not. Like this defines the future of your career, defines the future of your company, your industry.
[01:08:31] Paul Roetzer: And so the point I was making is like, Hey, listen, Iget it. Like this is hard. There is lot to process here.it can be overwhelming when we all have full time jobs, but the only option to me moving forward is you have. to figure out a way to stay on this. If it's just listening to this podcast every week, great. Like
[01:08:51] Paul Roetzer: hopefully that is what we're helping do for
[01:08:53] Paul Roetzer: a lot of you is give you that one hour a week where you can really think about this stuff. But my point was[01:09:00] you have to make time to learn. If it is podcasts,
[01:09:03] Paul Roetzer: books, online courses, attending events, subscribing to newsletters, Or just having a small group of trusted people that can vet this stuff for you and tell you what it means. and then the second thing was like, you have
[01:09:14] Paul Roetzer: experiment with it. Like I said earlier, like go, your assignment for this week, go to
[01:09:18] Paul Roetzer: meta. ai, connect your Facebook account, play around with the image generation tool. Like, Experiment with it, see what's possible, because every company is going to need AI savvy practitioners and leaders.
[01:09:31] Paul Roetzer: And the opportunities are going to be so immense for the people that, again, fight through this
[01:09:36] Paul Roetzer: period where it just seems overwhelming and much, and there's too many other priorities going on. This has to be One of those priorities. Like, it's, can't be a topic just push away for months at a time.
[01:09:47] Paul Roetzer: Too much change will happen in a short amount of time to just stand by.
[01:09:51] Mike Kaput: Amen. Alright, that's a wrap this week. as a
[01:09:56] Mike Kaput: reminder, Well,
[01:09:57] Paul Roetzer: today we'll be back this week.
[01:09:58] Mike Kaput: sorry, that's a that's [01:10:00] that's a reminder, yeah, this week you're gonna get another episode, episode 94, on Thursday, April 25th. So twice, twice as nice this week with the Artificial Intelligence show. I would also encourage you if you're getting value out of the podcast, it would help us immensely if you would go ahead and leave us a review on your podcast platform of choice that helps us get the show into the hands of more people.
[01:10:24] Mike Kaput: I would also remind you, like we talked about at the top of the show, we have the State of Marketing AI Survey out in the field for 2024. Your input on this survey, which takes just a few minutes to fill out, really helps us understand better AI adoption in marketing and helps everyone in the industry grow together.
[01:10:43] Mike Kaput: So go to stateofmarketingai. com to take that survey. And last but not least, tune in to the Marketing AI Institute's newsletter, which goes out every week and summarizes all All the news you need to know, including stuff we don't cover or can't get to on [01:11:00] each
[01:11:00] Mike Kaput: episode of the podcast, go to MarketingAIInstitute.
[01:11:03] Mike Kaput: marketingaiinstitute.com
[01:11:04] Mike Kaput: forward slash newsletter. Paul, thanks so much and get ready to run it back.
[01:11:09] Paul Roetzer: Yeah. Thanks everyone for
[01:11:11] Paul Roetzer: listening. we'll talk to you again in a couple of days.
[01:11:14] Thanks for listening to The AI Show. Visit MarketingAIInstitute. com to continue your AI learning journey. And join more than 60, 000 professionals and business leaders who have subscribed to the weekly newsletter, downloaded the AI blueprints, attended virtual and in person events, taken our online AI courses, and engaged in the Slack community.
[01:11:37] Until next time, stay curious and explore AI.