Christopher Long: Hello and welcome to episode ten of the Teak and Chris Podcast. My name is Christopher Long, and joining me today is
Pratik: I'm pretty quite aside.
Christopher Long: As always, I say that like there's gonna be someone else on the other side of this that that answers and says hello, but it is always my friend Teak.
Pratik: That's right. It'll always be me. Tell your audience.
Christopher Long: Unless w unless we have a third. Right? We are investigating, thinking about no, no, I'm not saying who or when or how or why, but like we are thinking about having some guests on soon. But that is for another day. I think you would like to talk about what you're feeling right now in the world of AI, because I think you may feel like this is a a different moment than what we've had
Pratik: You're giving you're giving away our secrets, Chris.
Christopher Long: Yeah, over the past two years.
Pratik: It does. It it feels it feels different. I think the only way I can put it is this feels like the month where AI's finally growing up. let's let's talk about a few things that I'm seeing, right? Regulations popping up everywhere. so that's, you know, off the top of my head, the Colorado AI Act, you know, the Great American AI Act, California's doing some things, EU is always doing things, Australia's looking into things like the The fact that regulation is finally catching up is really interesting. I think there's more of an emphasis in the market on infrastructure. Micron, has anyone looked at stock for Micron recently? Right. and Micron happens to be in my backyard, in in in Syracuse, New York. So we pay a lot of attention to what Micron stock is doing and and all of that. And then there's a lot of commentary on water usage around infrastructure and and there's a lot of commentary on
Christopher Long: Yeah. Yeah.
Pratik: you know, data centers and and now all of a sudden, you know, the US government is blocking models. What is that all about, right? And and obviously, ironically, it's July 1st when we're recording this. Mythos is back today, guys, if you if you're not aware, and by the time you're actually listening to this, you'll you'll be well aware. But Mythos is back today and and so what is it all about? Like there's just so much going on in terms of AI growing up from this thing that feels so disconnected.
Christopher Long: Mm-hmm.
Pratik: From commercial reality and regulation reality, and all of a sudden there's commercial pressures, there's infrastructure pressures, and there's regulation pressures. So much so that we we lost access to a model recently. So it just feels different. This feels this month feels different, this episode is gonna feel different.
Christopher Long: So where do we start?
Pratik: why don't we start with none of that? Let's start with let's start with ambient AI. I'm really excited about the the ambient AI. do you wanna do you wanna give our audience a little bit of context specifically on what Claude just launched and why we're why we're discussing Ambient, yeah.
Christopher Long: Okay, okay. Sure. Yeah, I think maybe it'd be helpful to tie the concept of of ambient AI to something tangible. So Claude tag would make sense. so if you're unaware, it was a big announcement this week. So if you're listening to this type of podcast, you are probably already aware. Claude made a new not a bot, but a new version of itself that you can have inside of your Slack environment. So they already had a Slack app. This replaces it a hundred percent. Like you're gonna have to, if you want to continue to have Claude inside of Slack, you're gonna have to move to Claude tag. And in a nutshell, if you've used Slack, you know that you can at mention somebody, which is also known as tagging someone, right? So I could at mention teak and ask a question. I could add mention teak on somebody else's message some post, right? So tagging in Slack is a way to bring attention to someone inside there, inside of your community. Okay. So this will now be clawed. And it's not a personal clawed instance. It's like clawed for your Slack. So if you have a small company and you have 50 people who all use Slack daily to communicate with each other with each other. Claude is now just one of those quote unquote people communicating with everybody. And you can tag Claude in conversations and ask Claude to do things. So where ambient AI comes in is this idea that you don't necessarily have to tag Claude. Like Claude is in Slack, Claude is paying attention to things.
Pratik: Yeah.
Christopher Long: One of the examples I saw was Claude could send you a message knowing that there's something in Slack that you care about and say, Hey, this thing was assigned. It's due in four days and nobody's commented on it, like nobody's interacting with it. Do you need to go like do something about it? so proactive. I mean, I will say always on. I'm not sure if that's exactly how it works, but that's the idea of ambient AI. Like it's there and it doesn't necessarily need you to show it the way. Like it's participating.
Pratik: Yeah, i we're truly going from AI that you go to for answers, actions, consultation, to to moving the needle to AI that's kind of always present. And to your point, it might not always be on, like technically, that's a word that is very clear what it means. but and I think that's why everyone's using the word ambience, right? Like it's kind of trying to resemble the perception or the feeling that it's always present. And I think there's more examples that we can kind of highlight. It's not just Claude Tag. Apple just, you know, kind of talked about this new LLM powered Siri that focuses on screen awareness. So it knows what you're doing and it kind of interprets those actions and what you're doing and what you're browsing and your your motions and your where you stop where you s you know where you scroll and and it tries to help you.
Christopher Long: Uh-huh. Mm-hmm.
Pratik: Google search recently shifted to what they call 24-7 search agents. So you don't go to Google search, ask a question, get results, and you pick one. You go to Google search and say, This is what I'm trying to accomplish, similar in a search fashion, but I want you to actually go out and do the work. and it just kind of spins up a background errand agent. and then, you know, we're seeing this pop up in healthcare as well. Where doctors are walking around with you know phones in their pocket or wherever it is, and it's just listening to their to their everyday activity. and presumably it's got an integration back to Epic or whatever it is, and it's just writing it's their documentation, it's kind of auto auto charting. So just like always working. so this this presence, this AI presence is interesting. my question to you, Chris, is is this because we've hit the limits?
Christopher Long: Mm-hmm. Mm-hmm.
Pratik: of model training when you have to go to it and the next paradigm shift for model efficacy is going to come from just always listening and always being present and you know creating the next evolution of model means training off of humans in perpetuity.
Christopher Long: I I would hesitate to say we've we've reached some limits on training. I think with enough compute, like we will figure out how to train even more and train faster. there is a large discussion though over is that is that the answer for AI to get better and evolve. So maybe that's the lens I would I would kind of come at it from is you know what is actually helping because you need to see a model needs to know results and you can only train so far. So now if you are always watching or listening as a model or an agent, you can start to see the results of whatever. So you trained your model, you have a base understanding of the the certain things. Then someone starts to use the model. Now you get it gain an understanding of how effective the original training was, right? are there things that should be added to the training, changed from the training, but also you just start to see how people use it and the results. So there's another level of awareness and helpfulness that I think a model and an agent can get to once they start to see how it's actually used and if the results are.
Pratik: Mm-hmm.
Christopher Long: are valuable. it's interesting for something like Slack. I so that the Siri case on an iPhone and Slack or Claude inside of Slack both really fascinate me. I think that is going to be the next most valuable thing is to have AI in the place where all of your conversations live.
Pratik: Mm-hmm.
Christopher Long: Right. That's so the and this is something I want to talk to you about because we've talked a lot about doing things locally, training a local model, it never goes out. I but I'm building this agent and then I have to go to my computer and I have to run it on my computer, things like that. We're quickly, it would seem getting to a point where my digital life, wherever that is. So if I'm at work, that might be Slack. If I'm out and about, it's my f my mobile device. All of those conversations live on those devices. and Apple is also very much touting the idea that if it's private information, they have a model on the device that will run so it doesn't leave your phone and things like that. What is the point of all of us trying to build our Hermes agent or you know, whatever it is we picked? if we're rapidly approaching these things just being available where we work and live?
Pratik: Oof, that was that was a ha it was a it was it no, I I think that's a fair question. I mean I I I don't know. I don't know the point, right? I think I think the point is just just get smart on it 'cause we don't know where things are gonna work and where things aren't gonna work. Like are people gonna accept that there's just an AI locally on their iPhones going forward? I think I I think so, yes. But I not every one of those examples will be met with a yes, right? Like
Christopher Long: Was that too much? Do I need to calm down? That was a lot.
Pratik: I I just don't know and and and that's that's where it gets interesting because I do think there's gonna be room for people to create the non enterprise or the non like Apple Google version of things. but you know it's and it's crazy to to think that this wasn't obvious. It it it it almost seems obvious that the easiest way to get people to adopt AI was to just do two basic things. One, put it exactly in the interface where they're already communicating with text-based communication. So Slackbot, Claude Tag, Siri, like those th those areas where we're already texting, right? Like Meta has their you know Llama LM on WhatsApp, like it just is such an easy adoption.
Christopher Long: Mm-hmm.
Pratik: shift to just open up a new chat with an LM instead. It doesn't feel any different than texting Chris. so so so that's one like that that feels like a a win for a lot of these companies. and then two make the model local, especially if you're gonna go into ambience. Now there's a value exchange that I don't necessarily know how how like how big companies reconcile that if the model's local, I'm Probably desiring that because I don't want you training on my data and then doing something with that. and if I have that, what benefit is it back to the company from a value exchange perspective? For an Apple, I'm spending a thousand dollars on a on a phone, I know where they're making money. but if I've got, you know, Google, which is free, and I want the model running locally or however they want to architect that.
Christopher Long: Mm-hmm. Mm-hmm. Mm-hmm.
Pratik: At the end of the day, they gotta make money somehow. So like those those commercial mod I don't know why my brain always goes to the commercial model, but but the commercial model will explain the behaviors and how all these models are installed, right? And whether they're local, whether they're trained on our data, whether they don't train on our data. So you know, my long-winded answer to your question is I don't know, but I think if we analyze commercial models and how they already make money, it'll become clearer on what might or might not work.
Christopher Long: Mm-hmm.
Pratik: Which is interesting. the the other part that kind of came up for me as you were talking is like is this now the unlock? You know, I think there's a diagram shared all over social media of like this many people are using AI and it's such a small subset and this this many people have just not adopted it. is is AI being finally in those workflows that you're already communicating with, is this the unlock for the masses? My experience is telling me yes. Like I'm I'm seeing it on a daily basis. I I really am. And I don't I I'm curious what our audience is seeing. I can't ask my audience this question right now. So Chris, what are your thoughts? Yeah, well what audience, what email us. What are you what are you thinking? What are you seeing? but
Christopher Long: Mm-hmm. Well you can ask him. You Yeah. Put a comment in in the video. Yeah, like let us know if you think you're using if you andor the people you know are using AI more on the or in these channels where they are most of the time. I mean Siri, I don't think that like I think the only way you can get it right now is if you run through the beta on the iPhone. so
Pratik: Yeah, please.
Christopher Long: There's still probably not a lot of folks with that experience, but I absolutely think like the the sheer number of people with mobile devices, if it is accessible on those devices, like adoption will skyrocket. but also at the same time, it's it's interesting because a lot of folks won't know they're not doing it intentionally. Right. They didn't pick up their device and say, okay, I'm going to do AI, like AI tasks today. They're just going to pick up their device. They're going to try and do whatever it is they need. And AI is going to be there seamless behind the scenes. now, then it becomes a matter of how does how do all these companies start to monetize it? Because we know it's way more expensive. Like I was thinking about that today. now in Gmail. You can have a start screen that's AI driven that just looked at all your email and and recommends, hey, you probably want to look at this and consider that. And like the business brain to me was just like, that is so much more expensive than just showing me my inbox. and I didn't like I'm not paying for Gmail. So they've have to have a play in mind eventually for that. Like what is that? Like
Pratik: Mm-hmm.
Christopher Long: Okay.
Pratik: I and I have to imagine, I have to imagine, right, back to my point, they're training on your data, right? Like there's no other reason to serve that to you for free. it's not out of the kindness or their hearts. So I have to imagine they want you to adopt, they want you to get an experience that's sticky, like it, feel good about it, use it a lot, and then they train on your data.
Christopher Long: Yes, yeah. Yeah. And what would be upsetting to me about that, and I think this is the case, is I don't recall giving consent in Gmail for that. So I'm just assuming it's probably s part of some larger agreement for just having a Gmail account that you, you know, you forfeit, you know, your rights to keep all this a hundred percent private or whatever it is.
Pratik: Mm-hmm.
Christopher Long: I wonder if there's a setting for all of these things. Like I can just be like, no, I don't want to do this. we'll see. There might be.
Pratik: Yeah. There might be, there might be, but I'm sure you s when you first created your Gmail way back when you were in college, or whenever, right? yeah, you you gave consent for all future data. You gave it all up, right? And so I'm there might be t settings or whatever it is, but I think you gave it all up. This this is a good this is a good segue to let's go back to regulation. Right? I'm gonna read a sentence here.
Christopher Long: Yeah. College sounds right. Gave it up.
Pratik: And and I think this will kick off the segment on on regulation. As of today, this is yesterday, as of today, June thirtieth, the export controls on Fable Fi five and Mythos five have been lifted. So I'm not gonna read the entire article. but you know, the as everyone probably knows, Mythos is this powerful model that does really well on cybersecurity and kind of can bring back vulnerabilities and really expose those. so they actually released a commercial version of this called Fable Five with just a ton of guardrails that they believed would actually work. Amazon went out and said, here's a report that shows that Fable Five did not have enough guardrails and it it did expose some vulnerabilities. The article goes on to say or the the announcement from Anthropic goes on to say that actually it was it was a vulnerability that any model would have exposed, including Kimmy K Kimik and and ChatGPT and and they talked about how what they're doing right now and to to push the bounds of like safeguards is they're they're calling this this concept it's a spectrum, but there it's like a classifier boundary. And the classifier boundary, there's a section of the boundary where it's benign, then that's right next to safety margin, which is right next to ambiguous, which is right next to harm harmful. And on that spectrum, they're saying like we we push the bounds of benign further towards the safe spectrum, sp safe end of the spectrum when we release fables. So there's there's less you can do so that we can get this model out to the market. And pushing the bounds of that, you know, kind of boundary from benign to larger safety margin and getting into ambiguous and harmful things was Amazon saying like we we we jailbroke it.
Christopher Long: Mm-hmm.
Pratik: We we actually got to that ambiguous part of the spectrum on Fable Five. Okay. So what's the point of this? The point of this is the US government saw that report and said, stop the model. Well, one, there's a lot that's interesting. Let's dissect this. The US government didn't say we ran the report. We didn't analyze the model. We didn't do any testing. We didn't do any research. They got their feedback from researchers at Amazon who have a commercial competitive incentive.
Christopher Long: Mm-hmm.
Pratik: To run the research against a competitive model. They waited for that report to come in and said, Other competition, stop your model. That is the current version of regulation that's happening in this country. and yeah. Okay, go ahead.
Christopher Long: Well, let's talk about that, because that's not my understanding. well, let what you s everything that you said may be valid and and true, but what I heard that the government said to Anthropic, and I say heard, like doing research for this podcast, reading articles, all of those things. Nobody told me these things. I went out and read these things. that they came back to Anthropic and said You have to be able to guarantee that only US citizens can use this. And that was the thing where they were like, Well, right now that's impossible. and that's like Anthropoc Okay, we have like we're shutting it down because we can't meet that requirement. is that your like did you see that as well?
Pratik: I did see that. That is not how they explained it in the R in the letter that they released yesterday. In the letter that they released yesterday, yeah. So that that that's what I was saying, right? Like what they what they said yesterday is as simple as Amazon exposed that Fable Fives safeguards, let a penetration happen, what what called jailbreak happen into harmful outcomes. They showed it to the government, and the government said, Stop your model.
Christopher Long: Okay. Talk talk us through that. Yeah.
Pratik: Like that's how they're explaining it, that's how they're framing it. There's no discussion about access countrywise, yeah.
Christopher Long: Okay. Well, so that that's a good thing to understand. Like, maybe that's all they cared like maybe that's all Anthropic cared about was that piece. Yeah.
Pratik: Fr from their lens, probably, right? Because they are a global company. They want other countries to use their their model. At the end of the day, whether it was don't let this other country use that jailbreak or don't let anyone use that jailbreak, I think from Anthropic's perspective, they don't want anyone jailbreaking it, right? That's the whole point of the safeguards. so that's how they're framing it, and I'm going entirely off their letter here to the masses. But I s I still think the point remains, regardless, is that the report was prepared by a competitive company, given to the US government. The US government took it and said, stop your model. Right? Like, yes, there are regulations coming. You know, again, I mentioned a few Colorado AI Act. Trump is working on a a kind of a whole act with the NSA NSA the Great American AI Act through Congress. New York is working on stuff, California is working on stuff, EU is always working on stuff. There are regulations coming, but the reflection of the regulations that come out of the government, forget about AI for a second. Let's talk about digital. They were always in the digital world six to nine months behind. Right. And then one of the moments that comes to mind, I remember Mark Zuckerberg getting grilled by Congress, and a congresswoman said, something along the lines, or woman,
Christopher Long: Mm-hmm.
Pratik: You know, t tell us about those cookies. Tell us about those cookies. Right? And you're laughing, Chris, because cookies at that even at that point, you don't even know year what year was, but you know that it was Mark Zuckerberg. So you know the fact that they were asking about cookies just so archaic. Like you you put that together in your head. But for our audience, like at the time of Mark Zuckerberg being in front of Congress, cookies were so ten years ago, right? The amount of technology in terms of privacy and all that stuff that was happening at that time.
Christopher Long: Right.
Pratik: Cookies was irrelevant. So the point remains of Congress was so far behind on their even their questions they were asking. Right. And and for all intents and purposes, I know that Congressperson did their preparation and their team did their preparation, but that's where that's the gap between where technology was at that time and where regulation was at that time. AI, I can't even keep up with AI. Every 20 minutes they're doing something new. We get we gave up, right? We gave up. So how how does REG
Christopher Long: Mm-hmm. No, we gave up trying to stay current. Yeah, I mean to the moment.
Pratik: Yeah, so how does regulation in a world where laws and policies take so long to get through, how does regulation keep up? I don't know, but is the answer competition doing research on competition and then getting a stop order?
Christopher Long: It's it it's philosophical. It's a philosophical debate because like in the lack of regulation, do you or do you not take action when there's evidence presented that points to perhaps action action being needed. So that's philosophical, right? Like it's okay. All right, so let's let's say And and people look at AI with the potential to be Earth shattering. Like there's a potential for it to just cause so much chaos and disrupt so much that it it could be dangerous. There's also the other side of that spectrum, super helpful, curing diseases, like all of those things. But in the lack of regulation, if you were to look at it as potentially harmful.
Pratik: Mm-hmm.
Christopher Long: Then there's a philosophical debate about okay, if there's some kind of moral imperative where ba it doesn't matter that this research came from a competitor, like if there's the chance this thing could get out and cause harm, do we or do we not act? And like we could have that debate for yeah.
Pratik: In in a world In a world where so much of what's right and wrong, good or bad is gray. Who determines morality? And
Christopher Long: That's the that's the philosophical debate, my friend.
Pratik: Right. And and so so like I I We're currently debating who is regulating. But then let's say we settle who is regulating. There's a further question on who determines what's right or wrong, who determines what's good or bad, who determines morality. and and the problem with I think the lack of regulation is that you're gonna have all these random players pop up. And and in a world where those players are incentivized for dollars and cents, morality gets tied to those dollars and cents. In some loose fashion, maybe it's direct, maybe it's indirect, but there's there's I think there's a tie. I don't think I want a tie between morality and dollars and cents as it relates to regulation. I know we need to wrap up here. So Maybe we stopped there. I I don't know. We didn't answer any questions.
Christopher Long: we can't end on that bummer of a note.
Pratik: One thing I want to add on is yeah, there's a lot of doom and gloom in the conversation we just had. one inspira inspiring read, the Pope. Why am I listening to the Pope? Well, one, he is the voice for billions of America sorry, billions of humans around the world. so I'd love to hear what they're hearing. but two, he happens to be now Commenting on AI, and I think that's interesting. And so if you have not written or read the letter that the Pope wrote, do so. And maybe we talk about it later on a different episode. But it is inspiring. He talks about the safeguarding the human person in the time of artificial intelligence. And he he kind of forces us to philosophically think about what is innovation for? Innovation in history and innovation as it relates to AI. why why do we innovate and what what is our mandate to innovate and and for for whom so it is an inspiring read it definitely put a little bit of a smile on my myself as as I was thinking through what he was saying and I I recommend it for anyone that hasn't read it
Christopher Long: Very good. So you're offering that that what would you call it? Is it a thesis? Is it just an essay?
Pratik: It is a it is a letter, that I think is intended to be a letter to the people as a conduit to the corporations.
Christopher Long: Interesting. Okay. All right. Very good. Okay. So man, we covered a lot of ground today. We kind of went yeah, we kind of went everywhere today. So yes, if you're listening, please let us know any comments or questions, anything you'd like us to pick up again or dr or dig into again. as always, we're available on YouTube.
Pratik: We did. We did. In a very short amount of time.
Christopher Long: Spotify and Apple Music. So if you're listening to this and you want to see the video version of this, check out YouTube. we really appreciate your time as always and just wanna say thank you. Have a great week. we will talk to you as soon as we can.
Pratik: Thank you.
Pratik: I'm pretty quite aside.
Christopher Long: As always, I say that like there's gonna be someone else on the other side of this that that answers and says hello, but it is always my friend Teak.
Pratik: That's right. It'll always be me. Tell your audience.
Christopher Long: Unless w unless we have a third. Right? We are investigating, thinking about no, no, I'm not saying who or when or how or why, but like we are thinking about having some guests on soon. But that is for another day. I think you would like to talk about what you're feeling right now in the world of AI, because I think you may feel like this is a a different moment than what we've had
Pratik: You're giving you're giving away our secrets, Chris.
Christopher Long: Yeah, over the past two years.
Pratik: It does. It it feels it feels different. I think the only way I can put it is this feels like the month where AI's finally growing up. let's let's talk about a few things that I'm seeing, right? Regulations popping up everywhere. so that's, you know, off the top of my head, the Colorado AI Act, you know, the Great American AI Act, California's doing some things, EU is always doing things, Australia's looking into things like the The fact that regulation is finally catching up is really interesting. I think there's more of an emphasis in the market on infrastructure. Micron, has anyone looked at stock for Micron recently? Right. and Micron happens to be in my backyard, in in in Syracuse, New York. So we pay a lot of attention to what Micron stock is doing and and all of that. And then there's a lot of commentary on water usage around infrastructure and and there's a lot of commentary on
Christopher Long: Yeah. Yeah.
Pratik: you know, data centers and and now all of a sudden, you know, the US government is blocking models. What is that all about, right? And and obviously, ironically, it's July 1st when we're recording this. Mythos is back today, guys, if you if you're not aware, and by the time you're actually listening to this, you'll you'll be well aware. But Mythos is back today and and so what is it all about? Like there's just so much going on in terms of AI growing up from this thing that feels so disconnected.
Christopher Long: Mm-hmm.
Pratik: From commercial reality and regulation reality, and all of a sudden there's commercial pressures, there's infrastructure pressures, and there's regulation pressures. So much so that we we lost access to a model recently. So it just feels different. This feels this month feels different, this episode is gonna feel different.
Christopher Long: So where do we start?
Pratik: why don't we start with none of that? Let's start with let's start with ambient AI. I'm really excited about the the ambient AI. do you wanna do you wanna give our audience a little bit of context specifically on what Claude just launched and why we're why we're discussing Ambient, yeah.
Christopher Long: Okay, okay. Sure. Yeah, I think maybe it'd be helpful to tie the concept of of ambient AI to something tangible. So Claude tag would make sense. so if you're unaware, it was a big announcement this week. So if you're listening to this type of podcast, you are probably already aware. Claude made a new not a bot, but a new version of itself that you can have inside of your Slack environment. So they already had a Slack app. This replaces it a hundred percent. Like you're gonna have to, if you want to continue to have Claude inside of Slack, you're gonna have to move to Claude tag. And in a nutshell, if you've used Slack, you know that you can at mention somebody, which is also known as tagging someone, right? So I could at mention teak and ask a question. I could add mention teak on somebody else's message some post, right? So tagging in Slack is a way to bring attention to someone inside there, inside of your community. Okay. So this will now be clawed. And it's not a personal clawed instance. It's like clawed for your Slack. So if you have a small company and you have 50 people who all use Slack daily to communicate with each other with each other. Claude is now just one of those quote unquote people communicating with everybody. And you can tag Claude in conversations and ask Claude to do things. So where ambient AI comes in is this idea that you don't necessarily have to tag Claude. Like Claude is in Slack, Claude is paying attention to things.
Pratik: Yeah.
Christopher Long: One of the examples I saw was Claude could send you a message knowing that there's something in Slack that you care about and say, Hey, this thing was assigned. It's due in four days and nobody's commented on it, like nobody's interacting with it. Do you need to go like do something about it? so proactive. I mean, I will say always on. I'm not sure if that's exactly how it works, but that's the idea of ambient AI. Like it's there and it doesn't necessarily need you to show it the way. Like it's participating.
Pratik: Yeah, i we're truly going from AI that you go to for answers, actions, consultation, to to moving the needle to AI that's kind of always present. And to your point, it might not always be on, like technically, that's a word that is very clear what it means. but and I think that's why everyone's using the word ambience, right? Like it's kind of trying to resemble the perception or the feeling that it's always present. And I think there's more examples that we can kind of highlight. It's not just Claude Tag. Apple just, you know, kind of talked about this new LLM powered Siri that focuses on screen awareness. So it knows what you're doing and it kind of interprets those actions and what you're doing and what you're browsing and your your motions and your where you stop where you s you know where you scroll and and it tries to help you.
Christopher Long: Uh-huh. Mm-hmm.
Pratik: Google search recently shifted to what they call 24-7 search agents. So you don't go to Google search, ask a question, get results, and you pick one. You go to Google search and say, This is what I'm trying to accomplish, similar in a search fashion, but I want you to actually go out and do the work. and it just kind of spins up a background errand agent. and then, you know, we're seeing this pop up in healthcare as well. Where doctors are walking around with you know phones in their pocket or wherever it is, and it's just listening to their to their everyday activity. and presumably it's got an integration back to Epic or whatever it is, and it's just writing it's their documentation, it's kind of auto auto charting. So just like always working. so this this presence, this AI presence is interesting. my question to you, Chris, is is this because we've hit the limits?
Christopher Long: Mm-hmm. Mm-hmm.
Pratik: of model training when you have to go to it and the next paradigm shift for model efficacy is going to come from just always listening and always being present and you know creating the next evolution of model means training off of humans in perpetuity.
Christopher Long: I I would hesitate to say we've we've reached some limits on training. I think with enough compute, like we will figure out how to train even more and train faster. there is a large discussion though over is that is that the answer for AI to get better and evolve. So maybe that's the lens I would I would kind of come at it from is you know what is actually helping because you need to see a model needs to know results and you can only train so far. So now if you are always watching or listening as a model or an agent, you can start to see the results of whatever. So you trained your model, you have a base understanding of the the certain things. Then someone starts to use the model. Now you get it gain an understanding of how effective the original training was, right? are there things that should be added to the training, changed from the training, but also you just start to see how people use it and the results. So there's another level of awareness and helpfulness that I think a model and an agent can get to once they start to see how it's actually used and if the results are.
Pratik: Mm-hmm.
Christopher Long: are valuable. it's interesting for something like Slack. I so that the Siri case on an iPhone and Slack or Claude inside of Slack both really fascinate me. I think that is going to be the next most valuable thing is to have AI in the place where all of your conversations live.
Pratik: Mm-hmm.
Christopher Long: Right. That's so the and this is something I want to talk to you about because we've talked a lot about doing things locally, training a local model, it never goes out. I but I'm building this agent and then I have to go to my computer and I have to run it on my computer, things like that. We're quickly, it would seem getting to a point where my digital life, wherever that is. So if I'm at work, that might be Slack. If I'm out and about, it's my f my mobile device. All of those conversations live on those devices. and Apple is also very much touting the idea that if it's private information, they have a model on the device that will run so it doesn't leave your phone and things like that. What is the point of all of us trying to build our Hermes agent or you know, whatever it is we picked? if we're rapidly approaching these things just being available where we work and live?
Pratik: Oof, that was that was a ha it was a it was it no, I I think that's a fair question. I mean I I I don't know. I don't know the point, right? I think I think the point is just just get smart on it 'cause we don't know where things are gonna work and where things aren't gonna work. Like are people gonna accept that there's just an AI locally on their iPhones going forward? I think I I think so, yes. But I not every one of those examples will be met with a yes, right? Like
Christopher Long: Was that too much? Do I need to calm down? That was a lot.
Pratik: I I just don't know and and and that's that's where it gets interesting because I do think there's gonna be room for people to create the non enterprise or the non like Apple Google version of things. but you know it's and it's crazy to to think that this wasn't obvious. It it it it almost seems obvious that the easiest way to get people to adopt AI was to just do two basic things. One, put it exactly in the interface where they're already communicating with text-based communication. So Slackbot, Claude Tag, Siri, like those th those areas where we're already texting, right? Like Meta has their you know Llama LM on WhatsApp, like it just is such an easy adoption.
Christopher Long: Mm-hmm.
Pratik: shift to just open up a new chat with an LM instead. It doesn't feel any different than texting Chris. so so so that's one like that that feels like a a win for a lot of these companies. and then two make the model local, especially if you're gonna go into ambience. Now there's a value exchange that I don't necessarily know how how like how big companies reconcile that if the model's local, I'm Probably desiring that because I don't want you training on my data and then doing something with that. and if I have that, what benefit is it back to the company from a value exchange perspective? For an Apple, I'm spending a thousand dollars on a on a phone, I know where they're making money. but if I've got, you know, Google, which is free, and I want the model running locally or however they want to architect that.
Christopher Long: Mm-hmm. Mm-hmm. Mm-hmm.
Pratik: At the end of the day, they gotta make money somehow. So like those those commercial mod I don't know why my brain always goes to the commercial model, but but the commercial model will explain the behaviors and how all these models are installed, right? And whether they're local, whether they're trained on our data, whether they don't train on our data. So you know, my long-winded answer to your question is I don't know, but I think if we analyze commercial models and how they already make money, it'll become clearer on what might or might not work.
Christopher Long: Mm-hmm.
Pratik: Which is interesting. the the other part that kind of came up for me as you were talking is like is this now the unlock? You know, I think there's a diagram shared all over social media of like this many people are using AI and it's such a small subset and this this many people have just not adopted it. is is AI being finally in those workflows that you're already communicating with, is this the unlock for the masses? My experience is telling me yes. Like I'm I'm seeing it on a daily basis. I I really am. And I don't I I'm curious what our audience is seeing. I can't ask my audience this question right now. So Chris, what are your thoughts? Yeah, well what audience, what email us. What are you what are you thinking? What are you seeing? but
Christopher Long: Mm-hmm. Well you can ask him. You Yeah. Put a comment in in the video. Yeah, like let us know if you think you're using if you andor the people you know are using AI more on the or in these channels where they are most of the time. I mean Siri, I don't think that like I think the only way you can get it right now is if you run through the beta on the iPhone. so
Pratik: Yeah, please.
Christopher Long: There's still probably not a lot of folks with that experience, but I absolutely think like the the sheer number of people with mobile devices, if it is accessible on those devices, like adoption will skyrocket. but also at the same time, it's it's interesting because a lot of folks won't know they're not doing it intentionally. Right. They didn't pick up their device and say, okay, I'm going to do AI, like AI tasks today. They're just going to pick up their device. They're going to try and do whatever it is they need. And AI is going to be there seamless behind the scenes. now, then it becomes a matter of how does how do all these companies start to monetize it? Because we know it's way more expensive. Like I was thinking about that today. now in Gmail. You can have a start screen that's AI driven that just looked at all your email and and recommends, hey, you probably want to look at this and consider that. And like the business brain to me was just like, that is so much more expensive than just showing me my inbox. and I didn't like I'm not paying for Gmail. So they've have to have a play in mind eventually for that. Like what is that? Like
Pratik: Mm-hmm.
Christopher Long: Okay.
Pratik: I and I have to imagine, I have to imagine, right, back to my point, they're training on your data, right? Like there's no other reason to serve that to you for free. it's not out of the kindness or their hearts. So I have to imagine they want you to adopt, they want you to get an experience that's sticky, like it, feel good about it, use it a lot, and then they train on your data.
Christopher Long: Yes, yeah. Yeah. And what would be upsetting to me about that, and I think this is the case, is I don't recall giving consent in Gmail for that. So I'm just assuming it's probably s part of some larger agreement for just having a Gmail account that you, you know, you forfeit, you know, your rights to keep all this a hundred percent private or whatever it is.
Pratik: Mm-hmm.
Christopher Long: I wonder if there's a setting for all of these things. Like I can just be like, no, I don't want to do this. we'll see. There might be.
Pratik: Yeah. There might be, there might be, but I'm sure you s when you first created your Gmail way back when you were in college, or whenever, right? yeah, you you gave consent for all future data. You gave it all up, right? And so I'm there might be t settings or whatever it is, but I think you gave it all up. This this is a good this is a good segue to let's go back to regulation. Right? I'm gonna read a sentence here.
Christopher Long: Yeah. College sounds right. Gave it up.
Pratik: And and I think this will kick off the segment on on regulation. As of today, this is yesterday, as of today, June thirtieth, the export controls on Fable Fi five and Mythos five have been lifted. So I'm not gonna read the entire article. but you know, the as everyone probably knows, Mythos is this powerful model that does really well on cybersecurity and kind of can bring back vulnerabilities and really expose those. so they actually released a commercial version of this called Fable Five with just a ton of guardrails that they believed would actually work. Amazon went out and said, here's a report that shows that Fable Five did not have enough guardrails and it it did expose some vulnerabilities. The article goes on to say or the the announcement from Anthropic goes on to say that actually it was it was a vulnerability that any model would have exposed, including Kimmy K Kimik and and ChatGPT and and they talked about how what they're doing right now and to to push the bounds of like safeguards is they're they're calling this this concept it's a spectrum, but there it's like a classifier boundary. And the classifier boundary, there's a section of the boundary where it's benign, then that's right next to safety margin, which is right next to ambiguous, which is right next to harm harmful. And on that spectrum, they're saying like we we push the bounds of benign further towards the safe spectrum, sp safe end of the spectrum when we release fables. So there's there's less you can do so that we can get this model out to the market. And pushing the bounds of that, you know, kind of boundary from benign to larger safety margin and getting into ambiguous and harmful things was Amazon saying like we we we jailbroke it.
Christopher Long: Mm-hmm.
Pratik: We we actually got to that ambiguous part of the spectrum on Fable Five. Okay. So what's the point of this? The point of this is the US government saw that report and said, stop the model. Well, one, there's a lot that's interesting. Let's dissect this. The US government didn't say we ran the report. We didn't analyze the model. We didn't do any testing. We didn't do any research. They got their feedback from researchers at Amazon who have a commercial competitive incentive.
Christopher Long: Mm-hmm.
Pratik: To run the research against a competitive model. They waited for that report to come in and said, Other competition, stop your model. That is the current version of regulation that's happening in this country. and yeah. Okay, go ahead.
Christopher Long: Well, let's talk about that, because that's not my understanding. well, let what you s everything that you said may be valid and and true, but what I heard that the government said to Anthropic, and I say heard, like doing research for this podcast, reading articles, all of those things. Nobody told me these things. I went out and read these things. that they came back to Anthropic and said You have to be able to guarantee that only US citizens can use this. And that was the thing where they were like, Well, right now that's impossible. and that's like Anthropoc Okay, we have like we're shutting it down because we can't meet that requirement. is that your like did you see that as well?
Pratik: I did see that. That is not how they explained it in the R in the letter that they released yesterday. In the letter that they released yesterday, yeah. So that that that's what I was saying, right? Like what they what they said yesterday is as simple as Amazon exposed that Fable Fives safeguards, let a penetration happen, what what called jailbreak happen into harmful outcomes. They showed it to the government, and the government said, Stop your model.
Christopher Long: Okay. Talk talk us through that. Yeah.
Pratik: Like that's how they're explaining it, that's how they're framing it. There's no discussion about access countrywise, yeah.
Christopher Long: Okay. Well, so that that's a good thing to understand. Like, maybe that's all they cared like maybe that's all Anthropic cared about was that piece. Yeah.
Pratik: Fr from their lens, probably, right? Because they are a global company. They want other countries to use their their model. At the end of the day, whether it was don't let this other country use that jailbreak or don't let anyone use that jailbreak, I think from Anthropic's perspective, they don't want anyone jailbreaking it, right? That's the whole point of the safeguards. so that's how they're framing it, and I'm going entirely off their letter here to the masses. But I s I still think the point remains, regardless, is that the report was prepared by a competitive company, given to the US government. The US government took it and said, stop your model. Right? Like, yes, there are regulations coming. You know, again, I mentioned a few Colorado AI Act. Trump is working on a a kind of a whole act with the NSA NSA the Great American AI Act through Congress. New York is working on stuff, California is working on stuff, EU is always working on stuff. There are regulations coming, but the reflection of the regulations that come out of the government, forget about AI for a second. Let's talk about digital. They were always in the digital world six to nine months behind. Right. And then one of the moments that comes to mind, I remember Mark Zuckerberg getting grilled by Congress, and a congresswoman said, something along the lines, or woman,
Christopher Long: Mm-hmm.
Pratik: You know, t tell us about those cookies. Tell us about those cookies. Right? And you're laughing, Chris, because cookies at that even at that point, you don't even know year what year was, but you know that it was Mark Zuckerberg. So you know the fact that they were asking about cookies just so archaic. Like you you put that together in your head. But for our audience, like at the time of Mark Zuckerberg being in front of Congress, cookies were so ten years ago, right? The amount of technology in terms of privacy and all that stuff that was happening at that time.
Christopher Long: Right.
Pratik: Cookies was irrelevant. So the point remains of Congress was so far behind on their even their questions they were asking. Right. And and for all intents and purposes, I know that Congressperson did their preparation and their team did their preparation, but that's where that's the gap between where technology was at that time and where regulation was at that time. AI, I can't even keep up with AI. Every 20 minutes they're doing something new. We get we gave up, right? We gave up. So how how does REG
Christopher Long: Mm-hmm. No, we gave up trying to stay current. Yeah, I mean to the moment.
Pratik: Yeah, so how does regulation in a world where laws and policies take so long to get through, how does regulation keep up? I don't know, but is the answer competition doing research on competition and then getting a stop order?
Christopher Long: It's it it's philosophical. It's a philosophical debate because like in the lack of regulation, do you or do you not take action when there's evidence presented that points to perhaps action action being needed. So that's philosophical, right? Like it's okay. All right, so let's let's say And and people look at AI with the potential to be Earth shattering. Like there's a potential for it to just cause so much chaos and disrupt so much that it it could be dangerous. There's also the other side of that spectrum, super helpful, curing diseases, like all of those things. But in the lack of regulation, if you were to look at it as potentially harmful.
Pratik: Mm-hmm.
Christopher Long: Then there's a philosophical debate about okay, if there's some kind of moral imperative where ba it doesn't matter that this research came from a competitor, like if there's the chance this thing could get out and cause harm, do we or do we not act? And like we could have that debate for yeah.
Pratik: In in a world In a world where so much of what's right and wrong, good or bad is gray. Who determines morality? And
Christopher Long: That's the that's the philosophical debate, my friend.
Pratik: Right. And and so so like I I We're currently debating who is regulating. But then let's say we settle who is regulating. There's a further question on who determines what's right or wrong, who determines what's good or bad, who determines morality. and and the problem with I think the lack of regulation is that you're gonna have all these random players pop up. And and in a world where those players are incentivized for dollars and cents, morality gets tied to those dollars and cents. In some loose fashion, maybe it's direct, maybe it's indirect, but there's there's I think there's a tie. I don't think I want a tie between morality and dollars and cents as it relates to regulation. I know we need to wrap up here. So Maybe we stopped there. I I don't know. We didn't answer any questions.
Christopher Long: we can't end on that bummer of a note.
Pratik: One thing I want to add on is yeah, there's a lot of doom and gloom in the conversation we just had. one inspira inspiring read, the Pope. Why am I listening to the Pope? Well, one, he is the voice for billions of America sorry, billions of humans around the world. so I'd love to hear what they're hearing. but two, he happens to be now Commenting on AI, and I think that's interesting. And so if you have not written or read the letter that the Pope wrote, do so. And maybe we talk about it later on a different episode. But it is inspiring. He talks about the safeguarding the human person in the time of artificial intelligence. And he he kind of forces us to philosophically think about what is innovation for? Innovation in history and innovation as it relates to AI. why why do we innovate and what what is our mandate to innovate and and for for whom so it is an inspiring read it definitely put a little bit of a smile on my myself as as I was thinking through what he was saying and I I recommend it for anyone that hasn't read it
Christopher Long: Very good. So you're offering that that what would you call it? Is it a thesis? Is it just an essay?
Pratik: It is a it is a letter, that I think is intended to be a letter to the people as a conduit to the corporations.
Christopher Long: Interesting. Okay. All right. Very good. Okay. So man, we covered a lot of ground today. We kind of went yeah, we kind of went everywhere today. So yes, if you're listening, please let us know any comments or questions, anything you'd like us to pick up again or dr or dig into again. as always, we're available on YouTube.
Pratik: We did. We did. In a very short amount of time.
Christopher Long: Spotify and Apple Music. So if you're listening to this and you want to see the video version of this, check out YouTube. we really appreciate your time as always and just wanna say thank you. Have a great week. we will talk to you as soon as we can.
Pratik: Thank you.
