Anthropic Watermarks AI-Generated Text

0:00Ten hours on my feet, traffic, noise, everybody needing something from me. My head is spinning, I just need the world to stop for one second. But then I get home, shoes off, phone out, and I open Modo Casino. Hundreds of slots, classic table games. It's my instant reset. The variety I want, and the quiet I need. Modo Casino, the social casino. Avoid prohibited. No purchase necessary. Visit Modo.us for more details. Download the Modo Casino app today. Must be 21+. Modo Casino. America's social casino. With the Amex Gold Card, you can earn rewards across dining and travel. Find out your welcome offer, which could be as high as 100 ,000 points.

0:39Apply today at AmericanExpress.com slash Explore-Gold. Terms apply. Welcome to the podcast. Today, we want to talk about the fact that Anthropic is going to watermark any text that Claude generates. There's a new EU AI Act that they're trying to comply with. And so there's actually going to be these watermarks inside of anything that Claude is generating so that you can have it essentially detectable as AI text. Now, there's some challenges that they're facing. There's the way that they're doing this, I think, is very interesting. And there's people that have tried to do this in the past. So we're going to break down the state of the industry and how this will impact businesses and content creators moving forward.

1:21If you're using AI to grow and scale your business, what this means for you. And speaking of growing and scaling your business, if you're interested in learning how to use AI and meta ads to grow and scale your career or your business, we just recorded an entire deep dive where I break down how I spent over the last few weeks$4 ,000 on meta ads for my company AI Box, what ads were the best, what my learnings were, what I'm doing differently, how I'm generating ads with AI, how I'm using the meta MCP to have Clot connect into my account and help me make changes. all of this stuff that this was definitely way too complicated for me in the past.

1:59But now with the meta MCP, I'm able to get some really incredible results. If you want to check that out, we have a whole video breaking that down on the AI Hustle school community. It's linked in the description and it is currently free to join the community. It's$20 a month if you want to get access to over 100 videos that we have in there. But there's over 700 members. It's an amazing community. Everyone's sharing what they're building. So yeah, if you want to go check it out, there is a link in the description. Jamie, tell me what you're thinking about this watermarking move in a tropics doing.

2:30Is this going to have a big impact? Is this good? Is this bad? What are we at for business? Oh, man. This one makes me a little bit nervous, to be honest, because in the past several months, whenever I want to write an email, I basically just hit the microphone button and tell Claude what the situation is, who the person is, and say, help me write this email. And I do obviously make tweaks, but if it ever gets put in Gmail like an AI content watermark or something, that's going to be pretty embarrassing in a business context. Although I already have a plan for working around this. Maybe it won't work, but sometimes I'll like if you screenshot something on a Mac, you can copy and paste the text from the screenshot.

3:16So maybe I'll do that. I don't know. It's got me worried, though. um what do you think jade so it's actually interesting the way that they're doing this um if you remember in the early days chat gpt had like a ai detecting website where it was like it could detect if text was generated by ai um and then they actually shut it down because i think there was like false positives people were like using ai for like school and then um the teachers were were finding out and they were like i don't know giving people bad grades but sometimes it was false positive. And anyways, I think opening, I didn't want to be liable for it.

3:49They shut down that project. And now I feel like we're kind of coming full circle where people are, you know, these laws are coming out where they're saying like, you got to disclose if you're using AI. But I mean, I'm in the same ballpark as you, Jamie. Like, I don't want to sit there and format an email. Like, I want to get my message across politely. And that's about it. And so I'm going to voice to text it. Sometimes like I'm annoyed. Sometimes I'm frustrated. Sometimes I'm in a hurry. There's a hundred different things that could happen. And I just like give the information in a very unformatted way.

4:18And I'm like, just format this to look nice. It does it. It looks great. I send it. I'm happy. But yeah, is this, do I want a little generated by AI bubble, like on everything I send? Wouldn't it be cool? And I'm sure people will make plugins that do that anyways. But I also think to a certain degree though, like to take the other side of that argument, I can tell when people send me an email that's AI generated. And if it's like a cold outreach, that's usually like a big turnoff where I'm like, ugh, like I don't want AI generated spam cold outreach. But if I'm having an email conversation with someone and it's about something really important, even if they're using AI, well, maybe I feel like it might dilute the value or something.

4:53Like if it's about an important thing, I'm like, oh, whatever, they're doing what I'm doing. They're doing what everyone's doing. I could clearly see what they're saying. And I go and respond with AI back anyway. So I don't, I don't, I don't know. Like, I don't think it's a, I don't think it's necessarily a bad thing. And I think to some degree, we can already kind of tell we're already doing this. It's just in our own heads right now. This is what the support page on Anthropic said. They said, because the watermark is part of the text, it will travel with the text when it's copied and pasted elsewhere and may persist through some editing.

5:23Watermarking will be applied at the model level, which means it will be present no matter which claw product or surface the text comes from. This is interesting. I also saw Suno is adding some sort of watermarking where to their music generation where um you'll be able to you know it'll be able to say if something was generated by AI again though when I listen to a Suno song I can usually tell if it's AI generated so like I think people are making a big deal about this but like at the end of the day a lot of us if you're AI savvy and if you listen to this podcast you probably are you can probably detect when stuff is generated by AI and maybe not always maybe there'll be differences or some things will change but I think at the end of the day this probably isn't going to change things dramatically.

6:05I do think it's interesting how they're doing it though. It's like, it's built into the text. I think it's basically statistical significance of like certain letter patterns in there. So you might read a word, it might, the phrase, a sentence, it might seem perfectly normal, but the way that they've like put the words in a certain place and put certain letters in a certain way, there's a statistical significance that even if you edit it, they can still, it's like a barcode of letters in there where they can detect that it was generated by clots. So anyways, kind of interesting how they're doing it.

6:36And I don't know. I don't think it should really change things dramatically. But I would be concerned if we start seeing platforms demoting or devaluing things generated by AI just for that sake, even if it was high quality content. Yeah. No, I think in some cases there is a legitimate reason for not wanting something AI generated. For example, if you're a lawyer or writing up a document, I could see how wanting to know if it was actually created by a real lawyer or just Claude, you know, that would be something I'd want to know. But yeah, so I can definitely see there in some cases, this is something valuable.

7:19But when it comes to like Gmail, I don't think it's that relevant whether or not it was, you know, fully written by a human or not. but I'm just gonna I'm just wondering when AI is gonna stop putting m dashes in because I always tell it to remove the m dashes I always even if I tell it that's my preference it always puts those in there yeah I remember Sam Altman like a year ago was like we solved m dashes like just put it in your system prompt to like never use m dashes and I did that and it still did and I'm like okay this is dumb like yeah right how can we not solve this problem but yeah m dashes are the bane i'll usually have like a follow-up message on every single thing that's like go remove all the m dashes go remove all the words whatever these words these phrases and it goes and cleans stuff up so that is annoying um as far as this like why they're doing it it's the eu ai acts transparency code this started um august 2nd and it requires uh machine readable markings of ai generated content so everything's got to be identified clod isn't just putting this in like regular cloud.

8:23They're also doing it in their API. They're doing it in cloud code. They're doing it cloud cowork, cloud tag. So anything that cloud does will be affected by this. And it should be, yeah, it should be pretty, pretty standard across everything. Blackforce Labs, Google, Meta, Microsoft, OpenAI, Synthesia, they've all committed to that same EU code. So, I mean, I think basically everyone's going to be doing this where all text generated by everything will have some sort of ai detection built in unless you go and get the open weight model from like kimmy k3 or something which is kind of funny but like all of these people fall all the regulation all these people making rules 100 is going to push people towards like the i'm sure they'll be called like illegal models at some point but like yeah like all the open source models it's just funny how that will probably drive adoption over there yeah if we have these like big banners that are like, this was generated by AI.

9:18Okay, fine. I'll just go use an open weight model locally on my device that isn't going to be, you know, that isn't going to have a big disclaimer on every email. So it'll be interesting to see where it goes with all of that. I, for one, make a lot of money generating AI-generated podcasts and different, I think, like histories, like my big niche that I do. But I have like fitness and finance and like some other areas as well, where I actually have completely AI-generated podcasts. And I'd be curious if those get demoted or devalued once, like if there's like AI watermarking built into them or how the platforms look at that.

9:56Obviously, I do a lot of work to make the scripts really high quality. It's not just like, you know, it's been an AI slop. There's a lot of, you know, human review that could go into it. And a lot of people listen to them. I make, you know, thousands of dollars from completely AI generated automated podcasts. but yeah, I'd be curious to see how that is impacted in the future. Absolutely. Hey, if you got any value out of this episode, we'd really appreciate a rating or review wherever you're listening. Those help us reach more people and we read every single one. So thank you in advance. And again, check out our AI Hustle School community if you want to learn how to actually take some of these tools and go make money with them and grow your business.

10:36Thanks for listening and we'll see you next time.

10:43Welcome to Moto Casino, where the excitement never ends. With thousands of the hottest free-to-play social casino games, fastest payouts, and the best promotions in the industry. No tricks or gimmicks. Owned and operated in the USA. Moto Casino is a free-to-play social casino. No purchase necessary. 21 plus to play. Void worth prohibited. Sign up today for a generous welcome bonus. Moto Casino. American Social Casino. Download the Moto Casino app today. The most effective people at work aren't working harder than everyone else. They're working smarter inside better systems. Superhuman Go from the makers of Grammarly is the AI chat that works inside every tool you already use.

11:21Always ready and already aware of what you're working on. It's a teammate whose only job is to help you be better at yours. With Go working with you, you can show off what you do best. See what Superhuman Go can do at superhuman.com. That's superhuman.com.