Subscribe
Share
Share
Embed
Is AI becoming self-aware? We're diving into a mind-blowing study from Anthropic that feels straight out of *Inception*, where AI models can identify thoughts that aren't their own. But that's not all—we also explore a new study where AI epically failed to compete with human freelancers. Are LLMs getting smarter, or is the hype ahead of reality?
Join industry experts and thought leaders as we dive deep into how artificial intelligence is transforming cybersecurity, shaping defense strategies, and creating new opportunities in the digital landscape.
Hey, welcome back to the pod. Today, today Shelby and I have some super interesting AI news that we want to share with you. And uh so stick around. Okay, so Shelby, the first story I got from you, mindblowing. Maybe more the mind is the LLM's mind that it's that it's that it's blowing its own mind. >> Is this phone where I think it is? >> Anthropic came up with a uh white paper where they basically have concluded that you know there's LLM have like emergent introspection awareness. So essentially and what does that mean? That sounds like a lot of words. Essentially what they did is they asked the LM a question, right? And then they're able to inject as part of the LLM's uh interpretability of that question a thought into the LLM like a word. And so they insert like a a word in all caps into the LLM's kind of thought process. And then the LLM would give back that word as part of the answer. And then they would ask the LLM, why did you give me back that word? And then the LLM would be like, I don't know. It just came was like came to my mind, but you know, it wasn't really something I thought of myself. Right? So, it was like self-aware enough that it knew that this word had appeared in its thoughts or in its response, but it also knew that it didn't originate from its own thoughts. um which is why they're saying like the models have the ability to do intros like emergent introspection or awareness on um on you know within the models themselves. So, you know, kind of another micro step towards maybe the LLMs are smarter than we think they are and maybe, you know, these are really the seeds of like what a future intelligence might actually look like in the world. So, what's what's your thoughts on all that, Shelby? >> You've seen the movie Inception, right? >> They like injected this thought into the AI's mind, I guess. Um, I think it's kind of cool that like for one thing, it's it's mastered colloquialism so well. Like it understands how people talk and so it doesn't say my working memory or something. It's it calls it it's like it's mind, right? It like talks to us like a person even though it is not, right? I think that's kind of interesting. It's become very personified in a way. Um, the other thing that I like about this example you shared is that it didn't hallucinate. You know, maybe we as people, if we were put on the spot, why did you say that? We might not entirely understand our own thought process sometimes fully and we might be like wanting to save face by coming up with a reason so we don't look like we're randomly saying stuff like we want to feel like we are rational and thoughtful. But the and so I was like glad that the AI said I don't know instead of just like you know it it was just in my mind or something instead of making up a reason because if it made up a reason that would be you know in the realm of it's hallucinating stuff. So I liked its honesty. It goes along nicely with introspection I suppose. >> Yeah. Yeah. And I'm sure you know they ran the test like multiple times right. So, you know, maybe there was some subset where they had some hallucinations, but >> it seemed like for the most part they were kind of getting consistent results. >> And the first thing that I kind of thought of when I read the article was that quote um where they say, "I think therefore I am." Right? >> It's like kind of like the test of like, >> you know, how do you know that you're actually real and that you're actually a human? Well, if you can think that thought, then therefore, you know, you must be there must be some intelligence behind it. And we can all we can see that same pattern at least reflected in the AI's response to these like injected words into its thought process. So, um, so I I don't know. I mean, I know there's probably going to be like more aspects of the human brain than just like the token predicting techniques that were that are used in LLMs now that probably need to happen for before LLMs can get to like a you know human level completion of tasks. Um, like for example, you know, like we as humans, we kind of have like, and I don't know if we really fully understand this, but you know, we kind of have like short-term memory and we have like a long-term memory. And my understanding of science, which maybe this is off off a little, but seems generally true to me, is like when you go to sleep, like your body figures, your brain figures out what am I going to take from the long-term memory and index into what am I going to take from the short-term memory and index into the long-term memory. And it feels like somehow sleep is like related to that whole like indexing process, at least for me. So, so, um, you know, I think we'll probably see a lot more emphasis from like the LLM like research labs on, you know, better memory processes, right? Because I mean, right now they have like trivial things where they're trying to extract thoughts and then reing inject them as part of the inference process when you ask a new question. But, you know, it'd be much better if there's just like, you know, a way to kind of like update these short term and long term, maybe like vector stores or things like that in the future. So, you know, that that the entity like the LM actually that grows over time, right? So, um anyways, this is for you. >> Yeah. Go. Yeah. For anyone who might be alarmed about an AI uh overlord overthrowing of the humans, do you think that this uh this emergent awareness introspection is any cause for concern? >> I mean, I think it's kind of like evidence that either there's two conclusions I have to this, right? Evidence that LLMs are smarter than people are giving them credit for, right? like they're doing something more than just token prediction, right? That could be one interpretation. Another interpretation that I could definitely see from this, which I don't hear anybody saying, is that maybe humans just aren't that complex. Maybe we really [laughter] are mostly token prediction, right? You know, I mean, I think there's more going on with like our memory systems and things like that. Um, but I don't know. And I think the truth is probably somewhere in the middle, right? I think people overly give humans credits for being a lot smarter than maybe we really are. And I think also LMS are probably underappreciated, right? And are probably smarter than we're kind of giving them credit for because they have a lot of flaws still at this point in the game, right? Um, so I think the truth is probably somewhere in the middle, right? I mean, I think there's some complexities in the human systems that still definitely haven't been cracked or solved. Um, but then I think, you know, I think LLMs are are probably smarter than a lot of people are giving them credit for. So, um, anyways, that's kind of my perspective on it, at least today. I reserve the right to update my perspective at any point in time if any of this ends up being false. But uh due to future research, but based on the data I've seen it now, that that seems to make sense that LLMs are smarter than people give them credit for and humans are dumber than people give them credit for. I don't know. [laughter] >> Seems right. >> We're going to meet in the middle. [laughter] >> Funny. What have you seen this week, Shelby? Yeah, go. Yeah. >> So, um, a lot of the time we talk about like some like the way AI is being used for the adversarial side and obviously we know we've got to balance it out with some tooling for security experts to help them keep up and um, and keep going, right? So, I just was looking into one tool um, that's fairly new. It's called Sock Autofocus and its maker is Blumira. And this the the target audience for um for this tool is like your IT teams, your MSPs, um basically anyone who's going to be looking at your security alerts, right? Um and the goal is to help your analysts do a better job, right? So they were actually pretty clear. They said we are like they're not trying to replace the humans. they feel like this is an aid for the human analysts to be able to do their job better with the you know technology assisted. So the goal is to reduce alert fatigue um and accelerate incident response um which interestingly it was kind of showing like oh this is how you can measure your ROI on this tool you know by showing how it it cuts down on your incident time um and also just kind of have better better responses for your alerts. So, how does it work? Basically, um if you picture your screen, the way I saw it, it had like analysis on one side of um of an alert and then it had the assigned responders, kind of typical things you'd expect to see in like a SIM or something like that. And then on the side, it had autofocus on the other side. And this autofocus, it would give like expert guidance um and information that matches the context. So, you know, you've got your information about the alert, the alert and then it's going to kind of give you a little bit more basically like if you are a junior analyst and you're trying or just someone who doesn't understand the alert that you're looking at like why is this a problem? What does it necessarily mean that this particular alert happened like this, right? And it just gives you context and information that relates to that specific alert. Um, so it's supposed to kind of um bridge that gap if you have any knowledge gaps, right? Give you the information you need to make an informed decision, but then it kind of goes another step further. It gives you step by step kind of like walking through the workflow of it. So down below the analysis side, there's a workflow and it asks some questions, you know, is this behavior expected? Yes. And then it can take you to another branch off of that workflow. And if is it if you click no, is this behavior expected? then it can take you to next steps in your investigation. So it um it guides the analysts, right? So you so the the humans are still running the show, but they're making smarter decisions because they're getting informed um context, right? So I thought that was pretty cool. Um it also I imagine if you're using this and you're taking advantage of that information that's being put in front of you, it's going to train your users too. Your sock analysts are going to become better at what they're doing. takes a little bit of that guesswork out, you know, makes their job easier to just get the relevant information, I think, to make a good decision. So, that was I I thought it seemed like a cool tool. Um, I really like where this is going. Um, and then just trying to think like critically, you know, what blind spots might we still have on this? You know, will the tool recognize its limitations or will it give you false confidence? you know, um I you know, we we know the ttps of today, but we don't always know the ttps of tomorrow and next week. You know, I hope that like it will continue to stay current with all the attacks as they kind of keep coming to help guide the an analysis through in a similar vein, right? Because things are always changing. Um but yeah, what do you think about it? >> Yeah, I think that's super cool. I uh I uh definitely think that sock analysts and cyber threat hunters or cyber threat intelligence professionals, they could use a lot of more help, right? Because they're they're pretty especially sock analysts, they're pretty overwhelmed, right? Um and you know I believe strongly in the concept of like creating force magn force magnifier magnifiers for humans rather than you know there's all these like hipster companies coming out where they're like we're going to automatically triage every alert that comes out of the system using AI agents and um like yeah cool that sounds amazing there's no way that's ever going to work. I mean, you know, it's gonna like be another layer of triage, right? And maybe eventually that'll just get like absorbed into everyone's security stacks, but to today it's novel, but tomorrow it'll just be like part of like the normal process of like weeding out false positives, right? But what's not going to go away is eventually there's going to be some subset of these notables or suspicious events that a human's going to have to review. And then if you can really help the human review those a lot faster um or train up people a lot faster so you can I mean it takes a lot of resources to when you hire a sock analyst to get them to the point where they're operating as a mid or a senior. So maybe you know AI is helping them just get trained up faster. Um, I I think that's a great area to be working on. And uh I I I know you're probably not aware of this, but I've been secretly building a product and like over the last, you know, six plus months that actually kind of operates in the space as well. Um, so I have some companies testing that right now. So I I'm not gonna I'm not going to talk as specifically about that today, but maybe in one of the future pods we can talk about it. But but it's Yeah, I want to look at this solution more because I haven't looked at it yet. >> Really dropping that bomb. That's cool. >> Yeah. But I I just could let you know like I I have a lot of thoughts in this space because I've been working on this space for you know the last year, right? So, um, but you know, it sounds like based on what you said, like, you know, we we've taken slightly different approaches, but, uh, >> but I do think, you know, that that's a really good space to like in there's a lot of improvement in that space that could occur. And I think right now like a lot of companies are just not looking at the bulk of the suspicious events uh because they just don't have the resources to do it, right? So, so you know, if there's capabilities out there using AI to help people triage stuff faster or better, um, then, you know, I think that's going to be, you know, I think that's going to I think that's going to become the new normal, right? So, >> yeah. >> And I love that it sounds like your tool probably does this as well as the sock autofocus. I love that in in in previous conversations we've talked about the um skill gap because you're going to have AI that's you know expert in some things and students who are using AI and maybe aren't developing the nitty-gritty like low-level skills that might have been required 10 years ago or whatever right um and I feel like this is bridging that gap so so nicely right with the help of AI because they're they're still in the alerts they are still looking at the technical details and they're being tutored all along the way by the AI and making more informed decisions. I think that's really cool. So, I get excited about that, too. >> Yeah. I I hope there's a future where like anybody who's sitting in an operation center, you know, like a network operation center or like an app monitoring operation center could say to themselves like, "Oh, in five years I want to be in a cyber security career." And then they could just like say like AI model help me so I can learn more about cyber security as I'm running through my like day-to-day operations almost like take your existing pool cuz like yeah take your existing pool of like people that you have doing other types of monitoring tasks and see if you can train them up into the security monitoring sectors as well. Right. um you know and I think the you know the limiting factors there is really like the time and resources to get someone trained up is a lot. So I think AI could definitely help in that space. So So cool. Well on other news about AI taking all our jobs. You want to hear about the latest research report that came out? >> Oh tell me. I want to know. So, Scale AI um if you don't know about them, they historically have generated a lot of like training data and then provided that to like other companies that have generated models. Um so they have like a center for AI security or safety. So uh they call it C AIS and they basically put a test together and they use platforms that freelancers typically use, right? So think about like an Upwork or Task Rabbit or something like that and they took a sample of tasks. So someone asked them to complete a job and then these freelance workers completed the job. So they had both the questions and the answers for the data set from these platforms. And what they did is they took those same questions or job descriptions, the things that other humans were able to complete, right? Um, and then they provided that data to various AI models and then they took the outputs from those AI models and they had humans review the outputs, but the human outputs and the AI outputs were all mixed together, so they didn't really know who's, right? >> Okay. And so, um, the results were kind of surprising, right? Given all the hype we've heard from the industry, you would just think that like the LMS were just like crushing it, right? You give them a task and they would just it'd be easy for another human to look at it, be like, "Yeah, that's awesome. Let's go." But turns out most of the models performed under 2%. So, the highest mo the highest system >> 2%. >> Yeah. So only 2% of their outputs were actually accepted by another human. Um so and the highest was Manis which kind of like inteerates multiple models to use like a full desktop environment. Um that they got a 2.5% acceptance rating. That was the highest and Gemini 2.5 Pro got a 08. So less than a percent of them were accepted. Uh so actually the rankings were Manis Gro 4 Claude 4 point 4.5 Sonnet GPT5 and then Gemini at the bottom. So kind of surprising results um when digging into the white paper on why the humans rejected the the results, right? And reminder reminder, they don't know if they're coming from humans or they're coming from the LMS. They just know, they're just been told, would if you were the customer and you asked this thing, would you accept the result that came back? Um, and so 45% of them when they rejected it, they did so because it was poor quality. They just felt the quality was too poor of what they got back. And then 35% said that what they got back was incomplete or didn't really answer the task that they had that had been provided. So we can see like basically over 80% of these things were rejected for the one of those two reasons >> for being bad or not all the way done. >> Yeah. And like you know it kind of like in any workplace you know you're you're going to get called out if you don't complete the assignment, right? That's obviously like a autoreject. And then if you did it, but it was poor quality, like imagine I just sent you a report, Shelby, and like I just made up graphs or something, you know? Like that's probably not going to get like your approval rating. So >> yeah. Can I ask you a question? >> Yeah, go ask me anything. >> Um, do we know the approval rating of the human submitted answers? Well, we know that the data set that they were evaluating the humans had accepted the outputs, right? >> Okay. >> Um, but we don't it wasn't like you're Yeah, you're trying to do like a AI to human comparison on the output. >> Yeah, I was just curious. >> I don't think that was part of the study, right? Um, that's a good question, though. I uh I'd be interested to see another study that does something like that. Like like Yeah, that'd be interesting. like do the humans stuff get rejected at similar rates and for similar reasons? Um, yeah, that wasn't really, as far as I understand, that wasn't really part of the scope of this, but um, but yeah, that'd be an interesting question, too, right? Because I got to imagine that I mean, have you ever used any of those platforms before, like I would say 50% of the time I I kick back the work, right? >> Cuz like it's like a dude who's like >> iterative, right? It's an iterative process. You're like, not quite. Try again. It's also like a human thing, right? Because they're like, I'm doing all these different tasks. I'm just trying to crank them out as fast as possible. And I, as the customer, know that if I just tell them when they kick it over, no, that wasn't quite what I wanted, that they'll go spend more time on it, right? And it's not like I'm I'm not paying them more money for the more time. Maybe this sounds like an evil thing now that I'm saying this out loud, but you just get a high fidelity result if you like, you know, squeaky will gets the grease to a certain degree, right? So like if you're like no that wasn't quite what I asked for then they will spend more effort on it right which I don't know how that computes into this equation >> that would be interesting for like future followup on this because if you if and if someone made the argument that you know iteration and feedback is part of the process for humans and machines alike like um like I'm sure graphic designers feel that pain you know that back and forth and things like that. I'd be interested to see how many iterations or like um it would take until humans started accepting the results more or >> Oh, that's a good >> So, we'll submit it for future research. >> Yeah, there's probably a lot of additional research you could do in this space, right? Um >> because like >> Yeah. Yeah. Yeah. I wonder if how the LMS would react if you kick the action item back to them like I typically do to like half the work I get off those platforms. Um, with that being said though, I I do have to say from my own personal experience, which I understand is anidotal and not, you know, really repeatable, but ever since LM became a thing, like I've sent way less work to these liquid work platforms. Um because I feel like usually when I have a discrete task and I can package up all the requirements, I typically like just hand that to an LLM, [laughter] which kind of goes against what this what this thing's saying, the research is saying, because like my own personal experience is like I spend way less money on those platforms than I did two years ago. And cuz like if I have if I have to define everything like I have to for a platform like Upwork to get actually good results out of it and then half the time I have to complain to the person to go back and do more. Um like I feel like at that rate I I should just use like I should just go pay $200 a month for a GPT Pro 5 Pro or whatever and I could complain to the LLM. It's quicker, right? It's quicker at getting back to my complaints. But um [laughter] cuz I do complain about its output, too. So maybe maybe what we're learning from this, Shelby, is I'm just a a complainer. I'm not happy with anything and I want better results, whether it's from an LLM or from a human. >> Well, there are two sides of that coin. Maybe uh maybe you're visionary and you're searching for uh improvement and progress. >> Yeah. Yeah. Yeah. >> Sounds like a container. >> Yeah. It's it's all about the marketing spin. Yep. Yep. Yep. Yep. Yep. Well, did you see anything else this week that was interesting? What did you see, Shelby? >> Nothing um like too crazy, but I just wanted to kind of talk a little bit about the AI regulation. And I mentioned this to you earlier. It feels like often times when you've got new technology, um it it takes there's there's a gap of time before governments realize, okay, we need to put some regulation around this. we need to put some sort of guidance, you know, best practices or whatever it is for it to actually become like litigation or things like that. Um, so yeah, it's just it's kind of cool to see that with AI making things so fast like um commissioners and and and people around the world are are taking notice and they're they're trying to do work. They're trying to avoid being, you know, 10 years behind it, right? So I thought that was interesting. So I just want to talk about different things going on a little bit throughout the world. So in the Europe uh in our friends in Europe okay so the European Commission's guidelines um they have this thing called the EU AI act and they are moving that forward. So they now have um just want to talk like very high level um about kind of what is in this act one thing that it offers that I thought was very practical. I haven't looked at it in depth but hopefully it is practical. If anyone's used it give us a comment shout out show show uh tell us your thoughts about it. But basically there's draft guidance as well as a reporting template. So if you have a serious AI um security event incident, then they give you a draft to work with you. I'm sure you've got plenty to work on when you've got an incident on your hands, right? So they give you a draft to help you report it and and and resolve it and things like that. So I thought that was kind of cool. um their definition of like a serious AI incident would be something that's causing significant harm to either people or infrastructure or rights or the environment. And I thought that last one was interesting because um AI itself is not entirely environmentally friendly. So I'm very interested to see how they kind of define that or I don't know maybe it might be up to uh specific cases. Um, it also includes details like reporting timelines and investigative duties and cooperation that you need to take, you know, things you've got to do to help clean it up, make everything right again after those kinds of incidents. Um, which I think is really good because as everyone includes AI into their stuff, you know, it's going to be nice to have that guidance. Um, they're also starting to explore some guidelines for transparency for AI systems. Um, this is, you know, this is not complete yet. it is in the works. They're ask they're like opening up to consultation. They're in the information gathering stage. So that's just some stuff going on in Europe. And then some of the like reactions we're seeing um within Europe. So the Italian Senate, they adopted an AI governance law and this complements the EU AI act. So it seems like they're all kind of adding their own special flavor to it. The UK Department of Science, Innovation, and Technology also published a policy paper um regarding AI assurance, which someone please tell me what AI assurance means. I don't know what those words mean. Coming together. Yeah, I know what AI and I kind of think I know what assurance is. I don't know what they are together. So maybe that can be next week's topic. [laughter] >> You are assured of the human destruction by the AI. That's what AI assurance is. No joke, I just [laughter] >> I'm I'm sure this could have been solved by a quick search, but you know, sometimes mystery is exciting. Anyway, so we'll scoot on over to our friends in Asia. So, the National Cyber Security Standardization Technical Committee of China, we need This is why we have acronyms, I think. Man, that was hard. [laughter] They um they published an update to their AI safety governance framework. Um, and then also the CAC, which is the Cyberspace Administration of China, they implemented rules on methods for identifying AI generated synthetic content. Um, and this is one I wanted to get some opinion on, um, get your thoughts on. So, basically, they're like creating a like a standard or rules that everyone has to follow. So, if you're making text, image, audio, video, and virtual scenes, you're going to have to mark somewhere on there somehow that matches the CAC guidance that this was AI generated. So, wanted to get your thoughts. What's what's your reaction to that? Do you think it's necessary? Do you think it's good? Do you think it's possible? You think people are actually going to follow along? >> So, this is coming out of China, right, is what you're saying. So, I mean, I think Yeah, I think you know the entities that are existing in China are probably going to have to follow these standards, right? I mean, I think that the what the tech companies in the US are doing right now is they're each kind of like implementing their own watermarking technologies. And the reason that they're doing that is to try to stay off a like federal or state level litigation forcing them to do it because they fear that the state or federal level enforcement will be much worse and much more burdensome on them >> or maybe technology won't even make sense. Right? A lot of times laws get passed and there's no way technically for them to even implement those laws. So then they just get fined like crazy or you know it just becomes an operating expense at that point. But, you know, so they're trying to implement these things and I think at some point there'll probably be some and there probably already is, but like some nonprofit entity established in the US which comes up with a standard and then it's a very loose standard and then everybody's implementations that they're doing today will probably, you know, only need to be tweaked slightly to meet those standards. Um, so I mean I think this is coming down the pipeline uh in the US and the tech companies have seen this play out before in other ways and now AI is just like the latest iteration of this and so they're trying to stem off regulation getting forced on them and them kind of being more part of the solution. Um, now whether what the US aligns with and what Europe aligns with and what Asia aligns with or like China aligns with, I mean those standards are probably going to be pretty different in my opinion. >> I think you know maybe the U maybe EU will be more restrictive and then US companies will just adopt it just kind of like we've seen with GDRP. Um but it feels like the stand the standards that come out in China are usually so different that US companies are are willing to stand up separate business divisions that are just in China just for those standards right so like in the cyber security space you know essentially there's laws in China that say if you have any type of encryption keys the government needs to have access to those right so um so that's why a lot A lot of companies will stand up basically identical tech stacks inside of China um because they're like, "Okay, if you want access to China stuff, it's there, but it's not interconnected that US-based stuff at all." >> Um >> I can tell you about the one that we that I know about. So, if you go to contentcredentials.org and we can link it, >> um there actually is a group. It's a coalition. >> Um and they make they're they've published like a standard for content credentials. Right now it is a voluntary theme, right? They're creating a framework or like a a standard so that you can have like credentials and a history. So you take a picture on your phone and then you edit it in Photoshop and then you send it to um you know have it AI enhanced or whatever it is. Um it can have its own history and um it's this coalition is led by a lot of groups. There's um so I worked at Adobe and it was it was a really cool exciting project. Um Microsoft, Intel, BBC, Truepic, Sony, Google, Meta, Amazon, a bunch of these companies are um like in on it, but it's not a it's just kind of a voluntary, you know, whoever wants to adopt it standard. But I think it's really cool when it's actually applied because then to any user who's kind of scrolling through internet content, they can if if the publisher, you know, abides by this standard, then they can know the history of this picture. Okay, it looks like it was edited to show like brighter colors, but the rest of the content is legitimate or things like that. Like, I think it's so helpful for consumers. Um, so yeah, I thought that was kind of cool. Interested to see how it works out. Um because as is the challenge with so many technology pieces, we have a lot of standards, [laughter] right? Which maybe makes sense and maybe doesn't in each case, but yeah, >> I I know there was that time period where everyone was all the rage about the blockchain, right? But these type of scenarios are one of the few scenarios where I feel like blockchain technology makes a ton of sense. >> Like anytime that you want to keep a like a un like an unmodifiable chain of custody about a thing, right? like who modified the image and when. All the way from like this camera snapped it and signed it when they snapped the picture, then this editor made this change and then this editor that made this change and then it went on this website, right? Like those type of things. In my opinion, you don't have to use blockchain technologies to achieve those. We were doing those things inside of Windows and other operating systems long before blockchain even existed. So there's definitely other ways to do it, but you know, that's definitely a scenarios where I feel like blockchain type solutions would be really strong. >> Yeah, that's a good point. >> Where others other other places where they're applying blockchain just makes no sense from it from like a business standpoint. >> It's like, oh, we have this hammer called blockchain and we're just going to use it for everything, you know, like [laughter] everything's a nail. But >> that's a good point because it has the integrity, right? You can't mess it up. you know, it's got a >> committed to the chain and then it's also viewable by everybody. So, so you have like the visibility, you have the you have strong cryptography preventing people from like breaking the chain. >> Um, I don't know, that stuff just makes a lot of sense to me. >> Cool. >> Now, because it makes so much sense to me, I also as a cyber security person know we'll never implement that solution. We'll implement something much worse and it'll be easily broken. Uh but uh I could hope, you know, I could hope. I can dream, Shelby. I can dream. So >> I think it's a great idea. [laughter] >> And then like just some of the last things for uh >> I mean I didn't cover all of the AI related like regulations, but things are happening in India too. They're make they're uh they're looking at ways to leverage AI to accelerate their economic growth. Japan has also named some areas for specific areas where they want to investigate AI use and impact and stuff like that. Stuff going on in South Korea. Just there's a lot. But I'll end there. >> Yeah. Yeah. We'll we'll come back to it in future podcasts, right? I'm sure it's an evolving space that like viewers would be interested in hearing more about. Uh, >> of course. But with that being said, I found something amazing and I want to share it with you, Shelby, because I feel like this is a hidden gem in the universe. >> What? >> So, have you ever seen that show The IT Crowd previously? >> Yes. Yes. >> Okay. So, there's an episode or two where the manager, what's her name? Jen, Jan, one of those. >> I don't remember. She goes into like the back room, which is like the data center, and then there's like this emo guy lurking back there that she had no idea even worked in the company, right? [laughter] And when she came back out, she's like, "WHY DO YOU TELL ME THAT THERE'S AN ENTIRE person back there?" They're like, "Oh, we don't talk about him." Right? Like, [laughter] so I saw that same actor is doing a new TV show and the the emo guy in the back. >> Okay. >> He's also emo in this new TV show. >> Is he even acting? >> I don't know. Maybe he really is emo. I don't know. I mean, the it's a very Maybe it's gotten so method he's just became that person. Uh, but you know, it's been like 20 years between like these two things. So, but the new show takes place in like medieval times. Maybe not medieval. Maybe it's more like like what do you call that? Like Paul River times, like colonial times. >> Okay. >> I don't know. I I'm probably butchering it, but it's it's definitely not in this. It's definitely not today. It's definitely not current. It's in the olden times. Okay. >> Okay. And um and it's pretty hilarious cuz he's like in the olden times but he's emo and it's called the completely madeup adventures of dick turnup. So and it's a completely absurd comedy where there's an emo dude hanging out in the middle ages. I don't know what that actually is. I don't know anything about history to be honest. Shelby, I know cyber security. I don't know history. But I also know the show is hilarious. I watched on Apple TV, but I don't know. It might be on >> Wait, wait. Was he like an emo guy from modern times that was transported or is he just ahead of the emo uh trend ahead of his time? >> I think he's just I think he's just ahead of his time like he's like a kind of like a progressive dude in like the in the you know in the olden days, right? So, >> uh, it's definitely a comedy not based on any facts at all. So, as the title probably the completely madeup Avengers that Dick Turnup probably tells you, >> probably tells the the viewer, this is not going to be real. But it's funny, too, because it's like, yeah, I won't spoil it, but he inc he integrates a lot of progressive things into like a time where like everything is very like there's witches and all that type of stuff, right? Where you're kind of like, you know, everything's just magic, right? You know, so [snorts] >> that sounds fun. Sounds good. >> Yeah, it's a good show. It's a good show. Give it a try if you enjoy comedies that don't take themselves too seriously. So, >> excellent. So, >> what about you? >> Something really, really important to talk with uh talk with you about today. Um, it's actually probably the most important thing. Spinach. >> Okay. >> I really like spinach, [laughter] >> but okay. >> Help me out here, Bryce. Just kidding. >> Continue, Shelby. So when you or the hypothetical you because it sounds like maybe you aren't into spinach. When you go grocery shopping, you used to see spinach in a bunch, right? With a little twisty tie on it. >> Yeah. >> In the area that has the thunderstorms, right? >> Yeah. >> Now spinach is only coming in like the clamshells and the baggies >> cuz it's and it's like pre-washed and it's there's a trend. People are not buying mature spinach as often and so they're starting like grocery stores are carrying it less. Are we just going to let them get away with this price? >> Why are they doing that? Is it cuz why would consumers not demand the spinach with all the options of things to eat? They're like, "Oh, I really need the spinach." >> They're like, "I could have steak for dinner, but there's this really good looking spinach over there, so maybe I'll go for that instead." I I I don't think you should take eating advice from me and my decisions. [laughter] So I feel like that's the road to ruin. >> They're they're they're um making more baby spinach is what's replacing mature spinach. >> I see. I see. >> And so we're getting a lot more baby spinach, but it's usually coming pre-washed, which is a very nice convenience. And then it's coming in like more like plastic packaging. So, um I guess good things are it grows faster, but the bad things is it loses its texture really fast. So for anyone who wants to just kind of steam it down, get the get the grown. >> Is the baby spinach more expensive than the normal spinach? >> I don't know actually about the price, but >> it's just like demand. People like the convenience of the ones that are pre-washed. I'm like, why can't we pre-wash the big boy spinach, the big girl spinach, just the baby spinach? [laughter] >> You heard it here first, folks. So, I just want you giving you a pro tip on how to make more money in the grocery. >> This is very important, right? [laughter] >> It's very important that we get the spinach right. We >> talk about the real stuff here. >> That's what I got. [laughter] >> All I know is when I hear that thunder, I know I've taken a wrong turn at the grocery store and I just get out of there as fast as I can, Shelby. So, I'm like, I've gone to the green area. >> You turn toward the bakery or toward the dairy. >> Exactly. Exactly. [laughter] I uh I just I just know I need to get out of there. That's all I know. Uh um [snorts] >> yeah. Yeah. Well, I'm sorry. I'm sorry for your loss, the spinach loss. And uh I hope they rectify that immediately. So So well, hope you could think about eating spinach while you're watching the completely madeup adventures of Dick Turnup. And uh we'll be back next week to bring you more AI news. So, thanks for tuning in everybody. See you. >> Bye.