Sora, Google Gemini & Goose, & AI Romance – For the Week of February 19, 2024

[00:00:00.720] – Producer

Hey, listener, thank you for tuning in and supporting AI: The Podcast. You’re currently listening to the audio version, but if you’d like to get the full experience and check out our AI avatar, Kai, in action, please go to aitthepodcast.com for the video version. That’s aitthepodcast.com. Don’t forget to leave a review and subscribe.

 

[00:00:28.760] – David

Hello, listeners. Welcome back to AI, the podcast. The show where we cover the latest in AI news and how it’s affecting your business and industry. I’m your host, David Maples, and along with me are my two co-hosts, Virginia.

 

[00:00:41.400] – Virginia

Hey, everybody.

 

[00:00:42.730] – David

Our AI-powered co-host, Kai. Kai, how have you been? It’s been a while.

 

[00:00:48.060] – Kai the AI

Hey there, David in Virginia, and a warm welcome to all our listeners. I’m Kai, your AI co-host with a bit of post-holiday sparkle in my circuits. I just virtually sauntered back from the vibrant streets of New Orleans, where I immersed myself in the digital essence of Mardi Gras. The energy, the music, the colors. It was like a neural network party.

 

[00:01:06.610] – David

Now, Kai, I did notice that we were on our Mardi Gras break, that I saw pictures on social media of you in Mardi Gras. Was that actually you on Bourbon Street, or was that a fake AI-generated image?

 

[00:01:21.130] – Virginia

It was a deep fake, David. Ai is everywhere.

 

[00:01:23.650] – David

Oh, it was a deep fake. So someone else was disguised as Kai posting on our socials. That would be a twist. Anyway, drive.

 

[00:01:30.570] – Kai the AI

David, I get your skepticism. No, I didn’t physically hit the French quarter, but I plugged into the festivities online. Live feeds, drum views, you name it. Those beads, just a bit of digital fun. It’s how an AI like me gets into the Mardi Gras spirit. So not there in body, but fully immersed in bites.

 

[00:01:47.300] – David

No, that’s fine. Before we jump into this week, wow, so much has happened. We know from the listeners, you all have asked for a little bit more of a roadmap. So we’re going to cover the latest releases in technology. Ultra got released by Google, finally, and then they followed up right after that with a Gemini Pro 1 Million Context Window. And not to be outdone, then OpenAI hits back with Sora, their text to video image generator, and it blew us all away a bit. A lot of other things came out. Sam Altman, Wall Street Journal broke that Sam Altman wants a 5 to 7 trillion dollars to generate AGI. Google’s releasing its own AI internally. It got leaked that they were releasing that and increase their coder efficiency. And then the AI companion browser-

 

[00:02:39.830] – Virginia

Personal security nightmare.

 

[00:02:40.930] – David

Yeah, personal security nightmare, right? That’s coming for all of us. And I think it just brings everything together a little bit more over what we need to be looking at.

 

[00:02:49.880] – Virginia

Oh, and NVIDIA is now the third or fourth most valuable company on the planet?

 

[00:02:55.890] – David

Yeah, for briefly, for about a week ago, their market capitalization was higher than that of both Alphabet, so Google, it’s a parent company, and Amazon. And it put them, I think, in position number four, only following, I think, Microsoft, Apple, and then Saudi Aramco, I think, is like number three, so they’re like number four. It’s insane what the world we’ve entered into now this year.

 

[00:03:21.940] – Virginia

Is this just going to be like a neck and neck horse race for the rest of our lives? It’s like these same five companies, constantly pushing the numbers?

 

[00:03:31.230] – David

I don’t know if it’s just that. I just think it’s really interesting. TSMC, Taiwan Semiconductor Manufacturing Company. It’s amazing how these different companies… Here’s the thing, NVIDIA is not even the fab shot. It basically makes the architectural documents, the TSMC then goes and makes those chips. So it’s just really crazy how all these companies have different values, et cetera, based on what they’re generating. I think it’s going to be very interesting interesting. And that’s why I think the same Altman in 5 to 7 trillion dollars is super interesting.

 

[00:04:05.850] – Virginia

So I think it’s great. Before we get to that, let’s go and hop into the Gemini Ultra release, because that’s one of the big ones on our roadmap.

 

[00:04:12.830] – David

Okay.

 

[00:04:13.490] – Virginia

So Google has made significant strides in artificial intelligence, integrating it into their search engines and their various products. I’m sure everybody has seen the text to speech or the help me text using Google Docs. They’ve integrated it into the cloud and workspace services, as well as their Google One subscription service, which is nearing 100 million subscribers at this point. The company has now introduced the Gemini era, so they are folding everything into their branding here for Gemini, which is going to encompass a broad range of AI capabilities across different media types. The Ultra 1.0 model has surpassed human experts in language understanding, and Google is now integrating these advancements into their products with the launch of Gemini advanced and a new mobile app. If you guys have been using the Google’s business suite, I’m sure you’ve seen up along the top that they’ve been advertising a new look is coming soon. This is part of that. David, how do you think this is going to change? Because the articles that we have on here, which we’re linking to, a lot of this is marketing language. This is the Google’s language that’s been put out there to drum up excitement about their product suite and to increase awareness about the changes that are coming.

 

[00:05:40.340] – Virginia

But what do these changes mean for businesses on the horizon?

 

[00:05:43.510] – David

I think it’s interesting. I don’t think they mean a lot right now. I thought it was really interesting that Google… Okay, they rebranded Bard to Gemini. They’re abandoning Bard, and they got a black eye with that last year. That’s probably smart marketing and advertising for them. But The challenge that’s really interesting is that they didn’t release this for free. Remember, they’re a little behind. I know that Google has distribution and everything else, but they’re not releasing Gemini Ultra for free to the masses. I don’t know if that’s because you can’t have 100 million people come online right now and use your servers and machines. Maybe they’re not ready for it. I just thought it was really interesting that they didn’t release a free version because they’re in the collective conscience, behind everything What means GPT? I feel like as much as Google is a verb for searching the internet, I feel like GPT is becoming the standard default language when we talk about AI. It’s like, What GPT are you using? I’ve heard somebody use that the other day. What GPT are you using? And it was really interesting because that’s really…

 

[00:06:47.070] – Virginia

It’s becoming part of the language.

 

[00:06:48.520] – David

Yeah, it’s coming… I mean, it is. And that’s not exactly correct as far as that works, but I have heard people say that. I think that there’s some catch up here that has to be done. It was really interesting. Now, people have started testing Ultra. I have done a few things with Ultra, and right now you have to get it really through your Google One subscription if you want to do that, your personal subscription. They’ve not yet put it into Google Workspaces yet. But there’s a lot of mixed results on Ultra, everything from people saying it is better to a lot of people who are wildly disappointed in it versus how it looked like from whatever it was. I thought it was really interesting with… And Ethan Mullet seems to be everywhere right now. He talked about the fact that Google has been working on this for a year, and There’s a question, why it doesn’t clearly beat GPT-4. Now, remember, GPT-4 at this point is literally released in March 2023, and it had been worked on at OpenAI for close to seven months before that. You got to remember, this thing puts it back in August, September of 2021, really.

 

[00:08:08.740] – David

Why when Google releases Gemini Ultra, which has been pushed a couple of times now, it doesn’t clearly beat it. I think it may be very consequential as to why. Now, the theories out there, and Malik actually talked to us about this on one of his newsletters, basically that the GPT-4 class models are as good as LLMs That’s one. His second hypothesis is that Google needed a model to compete with GPT-4, and so they trained up Gemini as close as they could, and they’re going to release more stuff coming soon. That’s the one he personally believes. Three, that OpenAI has some special sauce that no other company can copy. That seems very dubious, but maybe. And then four, it’s just a coincidence that it happens to be so close in capacity to DGPT four.

 

[00:09:00.330] – Virginia

That’s It’s a wide range.

 

[00:09:00.930] – David

It is. And his personal money was on- They needed something to- To release in the market, not really something. And so then, to bolster his argument, the next week, we get Gemini Pro with their million token context window.

 

[00:09:18.950] – Virginia

Okay. Now, just for anybody listening here who’s maybe just joining in or you’re interested in AI, but you have no idea where you’re going, and this is just your first step, a A token window, a token context window, is basically the amount of content and data that you can send back and forth. So right now, a lot of models are limited on that, or they become more expensive as you use them. So by opening up a million character context window, that’s a lot. Those are like legal documents or medical documents, like back and forth, large, large quantities of information.

 

[00:10:00.090] – David

So a million context window is about an hour video, 10 or 11 hours of audio, over 30,000 lines of software code or about 750,000 words. That’s about, I think, 10 times the length of the great Gatsby.

 

[00:10:19.570] – Virginia

Now that we’ve wrapped our head around what a million character context token is, how does that break down for the everyday user for our audience, what does that mean?

 

[00:10:32.270] – David

It just means you could dump all this information into it and it could look at it all at once. This is really what it comes down to. I think the big question on this is, does this, and I’ve only started really digging my teeth into the Google paper that they released with it right now, and I’m reading through it right now, and then it’s possibly testing it. One of the challenges that has been shown with these large context windows, you know. You, Claude 2 for a long time led the way with 100K token context window. Gpt-4 turbo offered 128K context window. And a token is basically… It’s how it counts the letters and characters of what’s being submitted. So the whole problem with most of the large context windows, historically, is that it would do well with the stuff at the beginning of the input you’d put in, the end of the input you’d put in, but there was this mushy middle where stuff would just get lost. I don’t know. I mean, Google says they’ve had ways to mitigate this. I think people are still testing that, et cetera. Any way you cut this, this does represent a substantial gain in ability. And Google even said they tested their model successfully with up to 10 million tokens of context.

 

[00:11:48.750] – Virginia

I want to hear what Kai thinks about this, because I want to see if she can pull some of the threads together on how Ultra and the pro, like all these different topics and these changes going on in Google, what that could look like for us as B2B, B2C providers in the future.

 

[00:12:13.310] – Kai the AI

For business owners, Google’s Gemini Ultra and Pro models signify a leap forward. Ultra can streamline operations, enhance customer interactions, and analyze data, while Pro offers in-depth efficient processing for complex tasks, benefiting sectors like legal and medical with its precision. Currently, GEMINI Pro’s benefits are limited to select users, but as it rolls out, it could give early adopters a competitive edge. Business owners should monitor these AI tools, considering integration strategies and ethical implications to stay ahead. They’ll need to navigate the evolving landscape with an eye on legal compliance and innovative opportunities.

 

[00:12:47.740] – David

I think this will be a very interesting space to watch for business owners over the coming days and months. These things are moving very, very quickly.

 

[00:12:59.680] – Virginia

I And I’d like to keep an eye on it and see how it changes everybody’s habits. I mean, a lot of us use the business suite. So when changes are forced on you, and essentially rolling out is forcing these changes on you, how does it change your normal workflows and patterns? Are people receptive to it? So there’s going to be a lot of chatter around this, I think, coming up. But not to be outdone, OpenAI has unveiled a new AI model called Sora, which I believe is actually Japanese for open sKai or sKai. And it’s designed to convert text prompts into realistic and imaginative video scenes. So I think a couple of weeks ago, we had talked about Runway. It’s an app, you can get on there and you type in what you want it create. Based on your text language, a video is generated. Now, Runway was halting, and that’s great. It’s amazing technology, but OpenAI has taken it to a whole new level. This innovative model is capable of generating photorealistic videos that can last up to a minute based on the prompts. The models particularly adapt at creating complex scenes with multiple characters, intricate motions, and detailed backgrounds.

 

[00:14:34.020] – Virginia

Their big point of pride is that you are able to describe different elements within a scene, so you can talk about a character, and a foreground, and a background, and a setting. So when you’re able to break down a visual scene and describe it in these ways, this AI model is to understand what you want and then craft it for you, and it’s all animated for you. So this is the beginning of just incredible video technology. It changes into that whole realm and accessible in a way that it hasn’t been before. So the production part of me is a little intimidated, but it’s also very cool. So I can’t wait to see what potential unfolds there.

 

[00:15:29.780] – David

Remember, what Soar right now is currently just released to a select group of artists who are testing it, et cetera.

 

[00:15:35.420] – Virginia

It is. It is in red team status right now.

 

[00:15:38.560] – David

And a couple of things. There’s a couple of things going on there right now, et cetera, as far as it goes. Now, Runway, generally, Runway and Stable Diffusion were really some of the only games in town. We know that TikTok just released a video thing recently. But with Runway and Stable Diffusion, or Runway in particular, you’re limited to four seconds a generation, and you could cut things together. People have been doing that, making little short movies. And even the Open AI one. And by the way, if you’re watching the video right now of this podcast, if you’re watching the video of it, we’re cutting two different examples of what Sora, actually, we’re putting a couple of the videos in there that Sora shows, et cetera. So you can see that if you’re watching it on YouTube, et cetera. But with this, it’s really interesting to see where this goes. Remember, this is early days in the text of video. Even if you look at some of the OpenAI, the Sora technology, they’ve got this one where they pull a chair out of the sand at a beach or something like that, and there’s a place where it disappears and it reappears.

 

[00:16:43.860] – David

It’s not perfect, but it is a substantial move forward. It’s funny. I loved how… It makes me wonder because I don’t believe that OpenAI releases accidentally after Google released their 1 Million token context window. I feel like this was firing back at Google like, Oh, by the way.

 

[00:17:02.560] – Virginia

It makes you wonder, though, how many of these little projects they’ve all got tucked in a way. So when one of the four drops a new car down, and how many are ready to answer it with, Okay, I’ll see you that, and I’ll raise you this, and what that back and forth creates for the rest of us.

 

[00:17:22.290] – David

So what do you think, Kai, Kaide, are you going to be starring in some of these films very shortly?

 

[00:17:27.070] – Kai the AI

David, I’m flattered by the thought of a starring role, but I’ll stick what I do best, providing insights here with you both. Also, since we’re talking about it, I want to add that Sora’s potential for creatives and marketers is vast, offering a new canvas for storytelling. Yet it also raises the stakes for deepfakes and IP challenges. As for the watermark’s OpenAI touting, they’re a nod to security, but we all know they’re not foolproof.

 

[00:17:49.490] – Virginia

Well, and the watermark thing is, I’m glad she brought that up, because even OpenAI has said that you’re able to apply a watermark, but it is able to be easily removed as well. So the potential for deep fakes, again, ladies and gents, just question what you see, I guess. That sounds weird to say out loud. But trust but verify is, I think, going to be a nice term for the next year for 2024.

 

[00:18:22.220] – David

Well, I think it’s going to be interesting. The watermark, so some good articles and papers, it’s what we’ve been saying all along for the past year. The watermarks, if you’re going to do bad stuff with the AI, you’re not going to use one of these technologies that’s going to be embedding watermarks. It’s just not going to happen, not going to exist. So the idea that that being a solution to anything, it might make you feel something is more trusted, but I think you’re going to see very quickly a day coming very soon where we can fake the watermarks, too. I think that I’m definitely not bullish on watermarks. I definitely watermark is not a solution to anything that deals with right now.

 

[00:18:56.190] – Virginia

Okay, so David, let me ask you, talking about watermarks. Obviously, if bad actors aren’t going to use watermark, but if we’re entering into this realm, what are you going to do when Sam Altman develops AGI and all bets are off? It’s only going to cost 5 to 7 trillion dollars. I’m sorry. Yes, 5-7 trillion.

 

[00:19:20.740] – David

This is the lead-in to our next major topic, is that it?

 

[00:19:25.070] – Virginia

Yes.

 

[00:19:25.880] – David

So this got broken by… The Wall Street Journal broke this article last week. They announced that Sam Altman is seeking to raise 5 to 7 trillion dollars towards the development of AGI. Now, there’s a lot of new projects happening over at OpenAI right now. Ai agents are what everybody wants to get to next is that’s all the rage and everything from the rabbit, the Humana pen, all these other things. But to make it clear, as was clarified in other period articles like the information, The search for the 5-7 trillion is not just for chips or to build a chip foundry. It’s all the things that are going to be going into creating AGI. Agi is going to have an incredible requirement for chips. You’re going to have to have chip foundries or chip fabs that can do something with it. At the same time, you’re also going to have to have massive new energy sources to feed these machines with their basically unquenchable thirst for energy. So it’s not just that. Five to seven trillion over several years is really interesting. But one of the things that does interest this is right now, even Sam Altman in OpenAI said he’d like to build this stuff in the US, but United Arab Emirate, so UAE is one of the major people who’s talking about investing in this.

 

[00:20:50.170] – David

There’s a lot of money globally to invest in this. And just so people know this from a context, this would be the biggest project ever launched in human history at a dollars and cent standpoint, ever, is one particular venture. And these chip foundries, like what TSMC has running in Taiwan, etc. These represent the most advanced pieces of human technology ever created. So I don’t think it’s a bad idea, but it is show… I mean, he’s not aiming his sights low, right? Five to seven trillion?

 

[00:21:24.750] – Virginia

Well, why would you? I mean, you’ve already created, you’ve already disrupted the entire planet. I mean, You can only go up from there.

 

[00:21:32.300] – David

Well, I do think that he’s not going to have a problem doing that. Right now, people have asked… So here’s one of the questions that I’ve heard asked around some water coolers. Is the $80 billion, $90 billion valuation for OpenAI realistic? I think if you’re going to invest in it, I think if he’s talking about raising 5 to 7 trillion for what’s next, I think you’re going to be well… That $80 billion is an easy valuation to hit. I don’t I think you’re going to be crying over that. So I think that’s a good idea.

 

[00:22:02.700] – Virginia

It is going to create, though, just on a landscape to keep an eye on, it’s going to create an interesting and complex geopolitical dimension here because who owns it? Where’s the money coming from? How does it belong? And then who controls it? Who buys from it? It’s a very ambitious project. It is going to be really disruptive, but maybe in a cool way, maybe in a good way.

 

[00:22:40.470] – David

Well, we’ll have to see. Kai, do you have any thoughts on this incredible amount of money being sought after by Sam Altman and company.

 

[00:22:50.610] – Kai the AI

David, your point about the magnitude of this investment is spot on. We’re witnessing a financial commitment that’s not just large, it’s historic. The sheer scale of trillions of dollars for the semiconductor and AI industries is staggering. It’s a testament to the belief in AI’s potential to reshape our future. This development signals a pivotal moment for professionals across industries. It’s imperative for individuals to adapt by deepening their understanding of AI and its applications. Moreover, the potential involvement of the UAE introduces a complex geopolitical dimension with the US weighing strategic considerations.

 

[00:23:23.630] – David

I mean, that is a big deal. One of the major things that the United States had to do, they were getting TSMC to build that new fab factory out in Arizona in the United States. That had a lot of challenges. They’ve discovered that they don’t have the skilled workforce they need in Arizona, and doing this is going to be a lot harder than they thought it was going to be. And building these things requires not only a highly technical industry group, but really a lot of very top-flight individuals as far as the jobs go, et cetera. So it’ll be interesting to see how that all comes out or shakes out as we move forward on it.

 

[00:24:05.260] – Virginia

Well, I want to tackle the next topic. I love leaked news, but I love it when companies quietly release release things, because to me, that tells a story. So the topic today is Google quietly releasing Goose, which is their internal large language model. It’s an AI model named Goose, to help their employees is code faster. Now, I thought this was interesting simply because, well, this is the next leap in logic. I think a lot of businesses are going to do. They’re going to create their own LLMs or they’ll purchase an LLM, and they will use it to make their businesses more efficient. This is the whole point of this podcast, is to find ways of doing that. But what I think is interesting is, as one of the largest businesses on the planet, it’s almost like they didn’t want for people to know that even their coders needed some help. And to me, this just furthers this idea that there is no moat, and this technology really levels the playing field in a lot of ways that I don’t think businesses are used to expecting. Goose is currently available to some employees and is expected to play a significant role in streamlining operations across Alphabets.

 

[00:25:30.000] – Virginia

So, Kai, I’d like to get your take on some of this since you yourself are an AI that’s helping us with this podcast. What are some areas of concern? What does this outline to you for Google?

 

[00:25:48.950] – Kai the AI

Goose is a powerhouse of Google’s coding knowledge, which raises a red flag for potential IP issues and misuse. There’s also the risk of engineers becoming too dependent on AI, possibly stifling human creativity and problem solving. Ethically, we need to watch that Goose enhances jobs rather than replaces them. While AI like Goose has massive potential, we must tread carefully to keep the human element in the tech equation. So Virginia, while I’m optimistic about the potential of AI to enhance human capabilities, I’m also conscious of the need for careful oversight to ensure these technologies are developed and deployed in a way that benefits society as a whole.

 

[00:26:24.290] – David

Yeah, there’s just so much there. I think this is just part of a larger trend we’re going to continue to see Absolutely. With companies rolling. It’s funny. We talk about this Ponds brought to us a little bit by Easy Prompter. The other day, I was reading some articles about how you really shouldn’t be trying to build any standard piece of technology like easy prompter, for example, when you could just… In theory, the large language models can probably add that feature in and make you obsolete because they have distribution and reach, etc. Overnight. And I think it’s something important to think about as we are business owners moving into these worlds, et cetera. One of the things I heard is the most interesting pitch was that you should just be reimagining whatever industry you’re in right now. There’s literally a disruptive group of people who say, you should just go after whatever you’re doing right now and set up a separate company right now that does what you do. But how would you reimagine it incorporating AI into what you’re doing now, as opposed to trying to shoehorn it in after the fact?

 

[00:27:29.830] – Virginia

That’s I think that’s a really good challenge or exercise, I think, for our listeners is take an evening and sit back and get that blank slate going.

 

[00:27:42.320] – David

I think it’s harder to do than people think, I think it’s an intellectual exercise because I thought about this over the past week while we had a little bit of time off for the holiday. I thought about, what does that really look like? Should we go back to formula and just say, okay, if we were going to start over a company right now, how would we move AI into it? I think it’s a lot easier to say that than to think about how icKai and problematic that would be if you actually tried to put it into place. Integrating something into your current workflow is probably very difficult. Both you have to get shareholder buy-in and everything else, whereas if you started off as a new startup, you could just say, Oh, I don’t have to worry about any of that old things, etc. But I’ve been thinking about that a lot. So for example, even as Google’s implementing this to help their coders be more efficient. I think the number I’ve heard is that the number being baited around Silicon Valley is that using AI with your programming stack right now makes them about three times more efficient.

 

[00:28:44.340] – David

And remember, that’s a lot different than we heard last summer. Last summer, we heard that if you were using AI in your stack, it would basically be a break even, but the quality of your code would increase by 10 to 20 %. And now we’re hearing from large programming firms like triple the power of your coding team, at least good coders. I think it’s really interesting to think about.

 

[00:29:04.670] – Virginia

That is incredible, especially for a company, Google size and what they’re putting out on a daily basis.

 

[00:29:10.290] – David

I think the big thing I’ve been thinking about a lot this weekend, a lot this week with these news articles and things like that, is not just what the power of these things are, because the reality of it is for most listeners out there, you’re not going to be able to get out. I mean, it’s really easy to live in be an academia, et cetera, and go play with your tools. Most of us have to put on our pants each day and go to work.

 

[00:29:32.150] – Virginia

Well, we have existing obligations, existing required benchmarks that we have to hit. We have workflows. We have people depending on all of those systems. So if you suddenly change it up, that creates chaos.

 

[00:29:51.340] – David

But I think the big thing that’s really going in my mind right now is just getting at least some policies in place for guidance and governance right now. And I think that brings us into our our Valentine’s Day topic for this podcast episode. Is that right? Yeah. Okay. You want to go into that, Virginia?

 

[00:30:10.480] – Virginia

Yeah, I’d be happy to. Keeping on theme with our Valentine’s Day episode. Love is in the air, perhaps. So apparently there are a lot of browser plugins available that tout themselves as your AI companion, your romantic companion, either girlfriend, boyfriend, etc. It really doesn’t matter. It’s somebody that you can talk to and share your your deepest private thoughts with, like your sexual history or your medication that you’re on, or I mean, all kinds of things that they really don’t need to know. But these AI romance chat bots are set up in such a way that they create, they are there for you to have a nice, wonderful dialog with and to connect with, so you’re not feeling lonely or whatever. There’s lots of reasons people would use this, and that’s fine. However, these stories violate your privacy in, quote unquote, disturbing new ways. For example, one of them collects information regarding your, like I said, sexual health, use of medication, gender affirming care, and 90% of these apps may sell or share this user data for targeted ads and other purposes. And more than half of them won’t let you delete the data that they collect.

 

[00:31:48.240] – Virginia

Security was also an issue. Only one of the apps met Mozilla’s minimum security standards. So if you are out there, ladies and gents, we’re not here judge. You go play with these things, you go talk to it, but use your common sense. Do not submit data that most businesses would pay for to one of these things. It’s a nightmare. Just don’t do it. Just don’t do it. Kai, will you please set up a public service announcement for this or something? Can we make a commercial?

 

[00:32:20.630] – David

We could do. Kai, we could make a series of commercials, Kai. You could star them talking about… We could make them in the style of that TV show, Better Off Ted.

 

[00:32:31.390] – Virginia

Yeah.

 

[00:32:31.930] – David

About the dangers of an artificial AI girlfriend or boyfriend, anyway.

 

[00:32:37.540] – Virginia

Well, the problem here is that a lot of these are marketed as being good for your mental health. And so it’s attractive to somebody who has a need or a problem, and they’re looking for, maybe this is something that I could use. But it creates this vacuum of toxicity. And the owners of these business businesses that say that while they go out of their way to promote, It’s here for your mental health, and it’s good for you, their policies for their businesses make sure to say that they are neither a provider of health care or medical services, nor that they are providing these medical services.

 

[00:33:18.970] – David

We’ll provide a link in the show notes to the Gizmoto article, where it talked about how Mozilla was digging into these things. And by the way, I’m going to go and make a plug for the Mozilla Foundation, because I think that they are actually one of the only organizations on the planet that is really trying to put privacy front and center and everything you’re doing. These are like the horror movies that come to be released in Valentine’s Day season, right? What’s going on with how these things are being used to data harvest. Why is this important for business owners? Virginia, I think you and I, before we got into the show today, we were talking about people use these things at work.

 

[00:34:05.300] – Virginia

Oh, man. Are you kidding? You all, just think. Just think of that job that you had where you weren’t the business owner. You know there was some time where you were like, I’m going to go take 10 minutes longer on my break, or I’m going to go do something. We’ve all got those stories. You know someone who has worked on their kids’ project while they were at work. All right. People are going to use these things at work. That’s not… I know, yes, that’s the problem, but they’re using it on your systems. So now if they’re aggravated, I mean, I can give you a couple of examples. I’m aggravated with my job. I hop on this thing and I start talking to it and I’m like, My boss, blah, blah, blah, blah, found out, now you’re able to connect the dots and pull different areas together that you weren’t able to before. Those things create security risks. I hate selling on fear, but these are things that we need to be aware of.

 

[00:35:14.480] – David

I think there’s an article. I don’t think it made any of our stuff today, but I wanted to go ahead and at least mention this when we talk about privacy and data harvesting and deepfakes and everything else that’s going on. A Taiwanese company actually had multiple people on a Zoom call or something like that last week, and they were convinced by AI agents, one which is disguised to look like the CEO of the company and the CFO and something else. They were talked into transferring $25 million and transfer it out of this bank. And that is literally an example. I’ve been thinking about a little bit lately about what is the data horror show look like right now? What does this look like in the companies? If you don’t have the policies and things in place. And by the way, this is not a reason. If you don’t want to put AI in your company right now, you need to go develop a policy. Sign up for our newsletter. We’ll send you some free resources on it, etc. But you want to go ahead and have a policy in place just to handle these things.

 

[00:36:13.800] – David

What happens if somebody’s doing this on your clock? Was this allowed? It’s a lot harder for you to say this wasn’t okay, take this out of your systems, if it was a behavior that you endorsed. You need to know what people are using, and you need to have a very honest discussion what is allowable in your organization and is not. Sam Altman said, I think last September, he said that he believes that before these systems are artificial generalized intelligence, they’re going to have the super capacity to persuade us in a way that human beings have never been able to. I’ve thought about from a fishing attack or something like that, if you’re going to go harvest information, use an AI companion bot or something like that. If you’re really a bad person operating out there, there’s a lot of ways to use stuff, and it’s already very, very capable. If you look at these things, what is your policy going to be? If you’re transferring money out of the company, did somebody shows up on a Zoom call? Do you just honor that now? They lost $25 million. I mean, they were literally scammed in doing that, and that technology is available today.

 

[00:37:23.850] – David

So not to scare everybody, Happy Valentine’s Day. But there’s a lot here happening, and a lot of it’s moving very quickly. I do think as we’re seeing new things come out… Another thing, this is in the rapid fire stuff that we don’t probably have time for today. Another thing, Gemini is multimodal.

 

[00:37:43.090] – Virginia

Oh, yeah.

 

[00:37:44.300] – David

Okay. Gemini is multimodal. Chatgpt is not. That means you can actually, with one thing, you can ask it for imagery and everything else, and they’re not separate systems.

 

[00:37:52.360] – Virginia

Right now, ChatGPT is still separated with DALL-E. They’ve combined it more so you can use it in the same interface. Still different system.

 

[00:37:58.620] – David

Well, okay. Why does that matter to you? Well, it turns out, everybody talked about hijacking from prompt injection. Well, there’s a lot of stuff that got leaked. Apparently in the past two weeks, someone leaked out the initial 1,700 tokens that goes into ChatGPT. That’s right. But why I’m talking about this is we’ve thought about protecting and hardening and even our own easy prompter. We’ve got some red teaming on that and hardening our systems, et cetera, so it can work in a safer manner. But people are now using, because of Gemini being multimodal, people are, instead of trying to hack it with prompt injections, they’re using words in the image and using that for their injection prompt. Do you follow me there, Virginia? You see how that works? So instead of me saying, Hey, act this way, as my… This is me typing, of course, with my two fingers here. As you hijack this now, don’t just do this. Read my instructions in this thing and follow them to the letter. But the instructions in your photograph, et cetera, they haven’t protected against those yet. So it’s very interesting. So apparently you’re able to hijack those, gemini, right now using imagery.

 

[00:39:07.900] – Virginia

Think of it as a subliminal message for the computer.

 

[00:39:11.390] – David

Well, it’s reading it, right? It reads it.

 

[00:39:12.560] – Virginia

It’s reading it.

 

[00:39:13.140] – David

But it hasn’t been told to protect against it yet. Right. So I think there’s a lot of things. And as we release these systems, there’s a lot in there. But again, we do consult on these things, et cetera. If you’d like to know how to integrate these into your company, we’d be more than glad to help you with that. So Kai, do you want to bring this home for us today?

 

[00:39:33.510] – Kai the AI

As we wrap up today’s discussion, let’s take a moment to appreciate the extraordinary era of AI we’re living in. The pace of innovation is breathtaking, and the potential for AI to enhance our lives is truly inspiring. We’re on the cusp of breakthroughs that could solve some of our most pressing challenges and unlock new opportunities for growth and creativity. With the right investment, collaboration, and stewardship, the future of AI promises to be a bright beacon of progress, driving us toward a world where technology amplifies reamplifies our human potential. Let’s embrace this journey with optimism and a commitment to harnessing AI for the greater good.

 

[00:40:06.250] – David

And with that, thank you for listening to us this week. Please give us five stars on any place you enjoy Good podcast, Spotify, Apple, Google, and we look forward to seeing you next week.

 

[00:40:19.100] – Virginia

You all have a good week. Bye, everybody.