Anthropic's Fluency Index, Vercept Acquisition, and DoD Standoff

Welcome to the AI Applied Podcast. Today, we're talking all things anthropic. There is an absurd amount of news about this company today.

Specifically, they have a new fluency index that Connor has been researching heavily.

They've also made a recent acquisition into a computer-use company that I'm personally pretty excited about, just based off what I'm using.

That's verse-cept that they acquired, and they also earn a high stake standoff with the Department of Defense right now.

There's a lot to cover. Connor, I'm going to kick this over to you to start it off.

Tell us about their fluency index they've released, because I think this actually is one of those areas that is very applicable for people applying AI to their careers right now.

Well, that's the thing, Jay. On this podcast, as you and I love to do, it's like, really, how does this apply to how people work every day?

And I thought this was such an interesting one. So, yeah, so anthropic, you know, released this AI fluency index.

And they just, like, talk about essentially, kind of like, just to, I don't know, simplify it a little bit.

Maybe it's really like, who are the power users? Like, what are they doing and how are they doing it and stuff like that?

And the thing that I was, a couple of things that just kind of caught my attention right away.

By the way, Jay, the Department of Defense, I cannot wait to get into. Oh, my gosh. That's that. Anyway, we're going to save the tea for a little later.

But, um, but the first thing that I sort of like that, I guess jumped out at me, right, is this behavioral indicator prevalence.

They call it that again. I'm going to say those words again. The behavioral indicator prevalence.

Now, they're talking about behaviors. So why is this so exciting? Jay and you and I have talked about, like, you know, at AI mindset, we just released, you know, our course, which is all behavioral based training.

It's, you know, again, sort of like going out to some of the biggest companies in the world.

But what we do is different because of this, like, it is not a tool in the same traditional way.

This, like, your brain has a hard time with it because it has a hard time treating software like a person.

So why is this relevant to that? Well, with anything else, when you're talking about, like, Microsoft Office 360 or something like that.

I'm not talking about co-pilot. I'm talking about 360.

Uh, in big companies, they measure, um, proficiency, right? So at the business school, it's turned, you know, and when you turn, we measure people's proficiency in certain tech, in certain things like that.

But they're not using the word proficiency. They're using the word behavior because it's not a proficiency.

So what they found was that the number one sort of like essentially almost like indicator of like a power user is like they're defining and iterating.

And Jay and just two more words on this, right? Because that's 85, yeah, 85.7% of people who are like, you know, that's sort of like the number one like standout.

The next one down is like clarifies goals before asking for help. Then it provides examples of a good looks like specifies format and structure needed sets interaction mode.

But even just those four, I'm going to say them again, because this is what power users do everybody. Okay.

It rates and refines clarifies goals before asking for help provides examples of what good looks like and specifies format.

None of that is prompt engineering. None of that is assigned at a role. None of that is it is just what are you here in that? What do you hear? I'll tell you what you hear.

You hear somebody that is good with working with people. Those are the qualifications of a good manager of somebody who creates good work, you iterate and you refine you clarify goals, you provide examples, et cetera.

These are not tool goals. These are not software goals. These are not understanding algorithms and formulas. These are how do you talk to a person. So Jane, to me, it's interesting.

That anthropic is sort of like kind of like really putting this out there being like, listen, if you're good with working people, you're good with working with cloud and other large language models.

100% and I think if you use a lot of AI tools, this is something that I have started realizing recently, specifically for example, like when I'm vibe coding with something like loveable, if you give it a prompt of a tool that you want it to build.

And this is exactly like if you are a manager talking to someone on your team to help you with the project, it's the first thing it's going to do instead of just like jumping in and making a wrap draft and spinning out.

It's going to ask you five follow up questions like, okay, so just to be clear, do we want to do X? Do we want to do Y? Do we want to do Z?

And like for loveable, you'll go step through those five steps and we'll create the thing. That is like and basically the way that it is acting is how you should act in or in like using that tool.

And obviously the team at loveable has custom designed this. So when you give it a prompt, it's got these follow up questions that it will work with.

We've seen this with chat GPT as well in the past, but but basically you doing those different steps when you're using an AI model or when you're trying to build something or trying to use something, that's the most effective way.

And of course, like you mentioned, this isn't just for AI models. This is for humans too. I mean, becoming clear on the steps. So everyone's on the same page.

Like this is that's how you get good results. But it's interesting. It translates directly to these AI models. And I feel like these AI models are so good at this point.

It's not like how do I become better at using AI. That's not the question we should be asking ourselves today. It's how should I be better at managing people in general, managing product projects in general.

Everything we've learned in that in our careers is just directly applicable to AI. So I love that. And that's a great report from anthropic.

The other thing that I was super excited about coming out of anthropic was the fact that they have now acquired a company called Versep.

Now, this is not a company that a lot of people I think have heard about. And the reason for that is that Versep is a computer use company.

So essentially, they're helping, essentially, it's like taking over your screen and doing things. Computer use is like all of these tools that take over your screen, they click around, do things.

It's kind of like this concept of agents that we always had. The thing that I'm excited about is as far as anthropic getting better at this, I currently use anthropics Google Chrome extension. It's a sidebar.

It's amazing. I tell it to do something like this. This might mortified developers. But recently I was vibe coding a tool that I used to help with my podcast production and scheduling.

And I had a limitation in this tool that I've created. It couldn't publish two hour long podcast episodes. Like my back end wasn't built. I mean, when I say my back end, I don't even know what that means.

I have not a developer built it for me, but it was informing me your back end is not capable of doing this. And I'm like, OK, what do I need to do? And I was on loveable. And it was like, oh, well, you need to go set up like a Google Cloud and go into like the control console and like put in this code and do this stuff.

Like terrified, of course, because I have no idea what's going on. I go to Google Cloud Connor. I am not joking. I open anthropics Google Chrome extension is a sidebar. I copy and paste all the instructions that loveable gave me to that. And I'm like, do this on my Google Cloud account.

It goes. It clicks around. It sets the whole thing up. And now my tool can publish to our long podcast episodes. Do I know how? No, is that good? I mean, if it's a tool for myself, I'm not too concerned. I'm not, you know, trying to sell this to the enterprise or anything. It's just a personal tool.

But at the end of the day, the reason why I think I'm excited about this and many other projects is, is these models get better and better, these computer use, even if you are not an expert, even if you don't know anything about a topic, using these tools to control your screen, take over and get stuff done.

Like you just ask chat GBT, what do I need to send to my developer? It gives you the response. You give that to a computer use tool and or and or my financial advisor or my XYZ, like fill in any job role. And it takes over your screen. It doesn't I am beyond excited about this capability.

It's unbelievable. You know, and the really cool thing about it is, Jayden, I love how you brought that back to, you know, how people are using it every day because it's true. Like I do the same thing. I use Claude Claude's extension and Chrome.

And it's just amazing. I'm like, I can't and the most basic stuff. But I'm like, I just can't figure this out. Now what I think it where I think it falls down is in doing things that you could do much faster. Right. So for example, like one of the things that they kind of like illustrate a lot and even in our in our course, I think we kind of like illustrate the fact that it can make.

Find you flights and things like that. Right. The truth is, I find flights way better than Chrome is good because you sort of like see it actually filling this in sort of like slowly. I'm like, dude, just just go. However, on other things like like what you're talking about, like I was on Google Drive. It's funny because I was using Claude and Google Drive. But I was like, I can't figure out this file path. And it just did it. And then it downloaded all the files for me. It did everything I needed to do in two seconds. This is what Ethan Mollick talks about in this terms of jagged frontier. Right. Some things it does well. But the reason why I first up, I think is so interesting is if you look at it, you can see it.

So interesting is if it can speed up or lower the latency, maybe like sort of say, like of it, be able to bounce around your computer, that is something that I think has been probably a little bit of a drag on adoption.

You know, the fact that like it's just like it doesn't happen quickly. We had Logan, Patrick on whenever it was sort of like just recently and he was talking about that too. You asked him, what's the thing you're most excited about? He's like speed. He's like just like speed. I think speed like really, really kind of tears things up.

I'm excited about that.

Jean, come on. Department of defense. Like these stories that you have sort of like making the news right now. It's like this crazy standoff. And the background of all this guys is.

Anthropics is a safety company. Anthropic came online. Anthropic Dario and Daniel, I'm a day left open AI and took somebody people with them because of safety. And now Anthropics getting a little bit crushed on the safety.

But like now they're being asked to sort of like, you know, because they're such a powerful model. Well, they're collaborating with the Department of Defense or amazing now the Department of War. But anyway, point being.

There's like drama that's making like newspapers these days. So talk us through that.

Okay, routers was reporting on it. And I just, this is just the wildest thing ever. But essentially what's happening is US defense secretary Pete Hegsett, I believe you actually summoned Dario Amadeo to come and do like a meeting.

And they basically had this meeting scheduled out where they were talking about removing safeguards. Anthropic, I think put up a bunch of safeguards on their models. So when the Department of Defense is using them.

They don't have certain capabilities or functionality that Dario says is against their kind of terms.

The way has to love that, right? It's like, sorry guys. You can't.

Oh, yeah, you know, this just this bothers them to know. And so that is great.

And that's the other thing with like Hegsett, like, these are not sort of like reasonable people in the Department of War. They are literally called the Department of War.

They are not going to be like, let anybody in who's like telling them that to be like, oh, Silicon Valley, like take start up. You don't want it because of your okay.

So anyways, this is it's hilarious. I'm sure this, the resolution of this is going to be.

If I had to put my money on it, there's probably a polymarket bet. I'm going to bet the Department of Defense probably wins in here.

But it's an interesting case and an interesting argument. So I'll share kind of what's going on. Essentially, I'm not exactly sure at the moment that this was flagged to Anthropic.

I know that the Department of Defense said when they did their Nicholas Maduro raid, they were using Anthropics Claw to conduct the raid, which was crazy.

And I mean, obviously like the raid was a well oiled machine, like it no one died and no Americans died and they achieved their objective.

So like, that's pretty crazy to think Anthropic was helping to power of that.

But what Anthropics specifically said they don't want to do and that like on paper, so I don't know the implementation of anything, but on paper, I agree with Anthropic.

They're like, we don't want mass surveillance of US citizens and we don't want our technology to be used for autonomous drones, right?

And I mean, I think like a lot of people, autonomous drones powered by AI is the most terrifying concept. So like theoretically, I'm like, yeah, that sounds horrible.

And also mass surveillance of US citizens. I mean, I'm just a huge like, I hate the Patriot Act and all that kind of stuff.

So I'm like, you know, no surveillance for me. I'm all on board with that.

How it plays out I think is on and like what they say on paper versus what they're actually the guardrails or what the Department of Defense can do.

The Department of Defense basically said that if they don't, I think they have a deadline of 5 p.m. Friday.

And if they don't basically take off the restrictions, I'll let them use it for whatever they want.

They're going to classify Anthropic as a supply chain risk, which is typically saved a classification save for like foreign adversaries.

So it's it's pretty intense and it's it's a battle going on right now.

It's a, I mean, it's a battle and we can sort of like wrap it up. Jay, no, give me the last word on this, but like listen, Anthropic is a is a phenomenal company, right?

I mean, like in it's and but they really do it's sort of like the Volvo to a certain extent, right?

I mean, like they're always the safety company for a long time.

I don't think people think of them as safety. I think of them as code now, right?

Yeah.

Which by the way, guys, if you have not tried AI box of of late Jaden's company AI, it has blown up.

Jaden was showing me the numbers, which Jaden, I apologize.

I maybe you didn't want me to say this over the airwaves here, but AI box is like absolutely exploding.

And that's because sort of like, first of all, I think you like read the interface.

And, but, but mostly like this idea that you can sort of like use all, you know, trade, you know, do your workflows like inside this thing.

But my favorite thing is like comparing the models, all that kind of stuff.

You have different pricing tiers, everything else.

Absolutely phenomenals would try out AI box.

I the thing about Anthropic is that like right now, they're a phenomenal model, right?

And like, but the idea that they're going to be classified as like as a danger to society is absolutely nuts.

Jaden, I'll give you the last word on this. Like Harry is seeing all this playing out.

So this is what I think the interesting part with this is at the end of the day, I'm sure a lot of people are asking like, why is this even a question?

Why does the US government have to force Anthropic to let them use their tool like in a free market?

Well, they just go find out of their supplier.

This is the interesting thing. Anthropic is the only AI model that is flagged for use by the Department of Defense for like the top secret.

Basically all of their confidential projects.

It's the only one that's been approved so far and you know how like approvals on these projects are pretty insane.

So I think right now the the DOD is negotiating with Google XAI and OpenAI to be other models.

But in the meantime, it feels like they're kind of stuck in a lurch because they just got these like safe guards put on by the company they're using.

I think inevitably what will happen is they will make deals with with Google XAI and OpenAI and probably Anthropic to some degree if their public spat hasn't just like sourd them too much.

But they'll make multiple and if anyone puts any guardrails and any one particular use they'll just switch to another.

So I think that they'll kind of have access to all of them.

I mean, it's sort of a lesson on a on a national scale of not putting all of your eggs in one basket.

And it's also tricky because Anthropic is perhaps the best model for some of these reasoning tasks.

Like you understand why you want to go with the best model, but yeah, the department of defense should probably not put all of their eggs in one basket.

And I think they'll have learned from this lesson for the future.

But anyways, guys, thank you so much for tuning into the podcast. This one's been a ton of fun.

We appreciate all of you and we appreciate all of the reviews you guys have given.

So if you haven't already make sure to write a rating review on the show helps it out a ton and we'll catch you guys all in the next episode.