Anthropic Watermark Jailbreak

0:00Tyler Reddick here from 2311 Racing. Victory Lane? Yeah, it's even better with Chumba by my side. Race to chumbacasino.com. Let's chumba. No purchase necessary. VTW Group. Void where prohibited by law. CT &C's 21 Plus. Sponsored by Chumba Casino. You're listening to a podcast right now. Driving, working out, walking the dog. If you're into podcasts, chances are you have something to say too. With rss.com, starting your own podcast is free and easy. Upload an episode and we distribute it to Apple Podcasts, Spotify, Amazon Music, and more. Track your listeners, see where they're from, and start earning from ads just like this.

0:36If you've been thinking about starting a podcast, this is your sign. Start your new podcast for free today at rss.com. Today on the podcast, Anthropics' Claude Watermark was broken within four hours of it launching, which is hilarious for a lot of reasons. Amazon has dropped the$19.99 fee for Alexa Plus, and Calendly is launching an AI note taker. There's a lot of competition in the space. Calendly has some interesting features that they make this unique. Marvel has jumped their stock, their price has jumped six percent because Google is going to be buying up to$12 billion for AI chips. Flock has a new AI tool that's going to track drivers by their movement patterns, not just their license plates.

1:24With all the drama around Flock, this is a big deal. And Carpathia's LLM wiki idea, This is something that came out in April. I want to talk a little bit about where this is. He had this idea of personal knowledge bases. We're going to get into where all of this is today. If you want to get access to all of these different stories and the deep dive articles I cover, go check out AIChatDaily.com. This is my own website that is tied to the podcast. And if you click subscribe in the top corner, you actually can get a daily AI news briefing. Put your email in and I'll send you an email every day with the top five stories in AI.

1:58It's everything I cover on the show. but you get way more of a deep dive into the facts and everything happening. And of course, if you go check out the site, you get more data. And I also have a bunch of quick hits, just different headlines that are great. So if you don't have time for the podcast, or if you want a deeper dive on what I'm talking about here, go check out AI chat daily.com, subscribe to the newsletter, and you get a deep dive on all of this. Let's talk about what's going on with Anthropic right now. Because there's a whole bunch of kind of changes that they were making, not just with the text model, but with basically all of the code, everything that was coming out of Claude, they're saying, look, this is all going to be watermarked because of a new regulation that came out of the EU at their new EU kind of bill that the AI bill that passed.

2:37And because of this, they were going to watermark everything that came out of AI. Now, some of the controversy I've heard in the argument here is a lot of people are saying, look, I'm like trying to write an article or I'm trying to write an email or I'm trying to write any sort of document. I'm sitting there and typing out the rough draft. I throw it in there and say, polish this a bit, right? It could be something very small, like, hey, just make sure the grammar is good. When Claude spits it out and it's like, hey, I tweaked the grammar for you or fixed it up a little bit. All of a sudden, the output is going to be watermarked, even if I mean, it's just lightly tweaking it.

3:05You go take this document and you submit it somewhere. And all of a sudden, people are concerned that you're going to get penalized, whether that's at your job or whether that's at school or anywhere else, because you used AI. So there's, there's a lot of that that's going on. But anyways, with this particular jailbreak someone created, so I think a lot of people are excited about this because of all of those reasons. Developer Guillaume Mayer, he published a working watermark override. He did this four hours after Anthropik announced that they were going to be adding the watermark. So it's not like this takes a very long time.

3:35The removal code has been bookmarked more than 20 ,000 times on X, and it has over 100 GitHub contributors. The EU AI Act basically mandates this watermarking for new models. And all this has to happen by August. So the deadline came and they had to have everything done by the end of the month and all of the existing models that currently exist. So the new models, everyone's got to have it by the end of August. But even the older models all need to have this by December. And the fines are not a joke. They're up to 3 % of your annual turnover. So these companies making, you know, Anthropik is at$65 billion in annual run rate.

4:12They do not want to spend any percentage of that for fines because they don't have watermarking. So obviously, they're getting forced into this, but people are finding opt-out methods for this. 190 organizations, including OpenAI, Microsoft and Meta have all signed the EU transparency code of practice. They come up with these kind of initiatives. And they're like, Hey, look, do you want to sign the transparency code of practice? And you're gonna look bad if you don't and everyone's going to go sign it. But I think at the end of the day, sometimes this goes overboard as far as the watermarking goes.

4:41So anyways, I kind of love this idea. Personally, the an engineer named Eric Hughes, he also built a quote unquote, competing removal tool, he made this in 15 minutes, he used Claude to do it. But the original tool that everyone that kind of went viral made by Myers, he's a native French speaker. He says he edits all of his writing with Claude and Grammarly. And I mean, if I was trying, I also speak French. I'm not very good. But if I'm if I was if I'm ever sending French to someone, a friend in France, I'm always going to throw it through Chad Chibutier or Claude to go fix the grammar or fix it.

5:11I might even voice to text it, kind of write it out. But like, I know I got the grammar wrong, whatever. So for anyone, the English isn't the first language. This makes perfect sense. And even for me, where English is my first language, I'm still using AI tools to help with my grammar. And so just to have all of that automatically watermarked, right, especially when you think of a tool like Grammarly or any of the others that are implementing an API, because this is the watermark is through the API as well. So anyone that's like, hey, we'll help you with like writing or researching your school will help you with your grammar.

5:38Like any of these AI tools or using an API to cloud, they're all getting watermarked as well. So like everyone in the entire industry is going to get watermarked. anyways, Mayer in particular, if he said that his objection to all of this wasn't like transparency, wasn't saying like, hey, this is, you know, using AI, but he said that the water market as a mechanism, he argues that detection produces only a probability score. And it's not actual proof that something was written by AI and that employers or academic reviews could use false positives to reject candidates or accuse researchers or researchers of undisclosed AI use.

6:08This all goes back to the same argument that happened when Chai Chai PT first came out opening. I had a tool that could detect when something was written by AI. And I mean, that was probably partially watermarked. Maybe they had some algorithm for it. They can discontinue it, number one, because they said there's too many false positives. People are getting penalized at school for using AI. And I think they didn't want the bad PR and they don't want to be held liable to that. Amazon is dropping the$19.99 fee for Alexa Plus. So if you wanted to get their AI to help you with Alexa, you had to pay an extra$20 a month.

6:38And I think the reason for this is not shocking. It all comes down to competition. But for this, it's going to affect every current Fire TV stick, Fire TV Cube and Alexa Plus enabled smart TV in the US is they're going to all get the AI assistant. It's all auto upgraded. You don't have to get an app. You don't have to get a sign on. You don't have to get Prime. You just get this automatically baked in. Amazon said that Alexa Plus users have nearly twice as many conversations on Fire TV as users had with the original Alexa. why are they even announcing that because it is so blatantly obvious i mean let's go with old-fashioned siri versus chat gpt which one is going to get more conversations obviously the one that actually works the old-fashioned alexa which could do some basic conversations or the new alexa that's you know like let's just say it's effectively chat gpt that you could just talk to it about anything is there any shocker that they have twice as many conversations no in fact i'm surprised it's not more except it's amazon it's not like it's you know the bleeding edge ai tool that everyone's using or talking about.

7:37All of this, the reason why I believe they've done this is following Google that just rolled out Gemini on Google TV earlier this year. And Roku is upgrading their voice assistant with AI. They did that last year. So I think Google rolling out with Gemini and a lot of this just kind of coming included in those TVs is making it so Amazon feels like, look, this is something we actually have to focus on. Calendly is launching an AI note taker and Cali assistant to chase post meeting workflows. There's a lot that's going into this and there's a lot of competition. Personally, I'm excited to see Calendly getting into this.

8:09I'm someone that in the past I paid, you know, for Calendly premium or plus or whatever. And I think at some point Google came out with their own like Calendly competitor built into Gmail that was free. And I've been using that ever since and canceled my$20 a month Calendly subscription because I was like, why am I paying$20 a month just for a nice calendar scheduler? Google also did something in my opinion that was like sort of shady, but maybe there are some good reasons where they made it so if someone booked on your Calendly, instead of automatically adding to your calendar, even though you had like a connection to your calendar, it would send an email to you and it was like, this is from an unknown source.

8:44Do you accept it? And if you accepted it, then it could add to your calendar. I think the pros and cons are like if maybe someone got a hold of your Calendly link and they were not someone that you wanted, they could go book a bunch of slots. Theoretically, in fact, I even heard of someone back in like a couple years ago that ran a bot where he went and found the calendar links to all of the top VCs. And he booked a time on their on all of their calendars. And when they showed up, they're like, who are you again? And he's like, I'm just here, I added myself to your calendar, I want to pitch my product, and he gave them all a pitch.

9:11It's kind of funny. But at the end of the day, I think maybe that's what Google will say they're trying to avoid. The reality is, though, if you use the Google link, the same functionality happens where someone could just book and add it automatically adds to your calendar. So I think it was more just trying to kneecap a competitor versus trying to be more quote unquote safe, because their product is the exact same thing. In any case, Calendly is now jumping back into the game and I'm excited for some of their updates. They have a new AI note taker that is going to record, transcribe and summarize draft and your follow up emails.

9:41And it's going to generate action items from meetings. So I mean, this is basically like Firefly or tons of these other tools where you have an AI bot inside the call. The difference though, is Calendly is trying to copy what Granola is doing. And instead of having like an extra bot inside of a Google Meet, for example, they're going to be listening to your system audio. I think there's a couple reasons why. One of them, which maybe I'm like pessimistic here, but if I was Calendly and I saw Google come in and basically kneecap my company by cloning our feature and then sabotaging all of our signups, they look spammy.

10:14I would be very suspicious about building some sort of integration, which is just like another member bot added to like a meeting. And maybe there's like some way that Google could shut down bots and meetings, or opt in because they only want you to use the Gemini bot in the meeting. Like, you know, I imagine there could be something like this system audio. On the other hand, number one, it just seems way more intuitive. If you opt in on the system wide level, you never have to touch something you never have to say like, this person's meaning taker is in the meeting, you don't know, no one has to accept it, Google can't shut it off.

10:45It's just literally listening to your system wide audio level. So I actually think for a lot of reasons, it's it's a better way to go here. And of course, granola is doing this. So they're just following one of the big competitors. They also are creating a separate assistant called Kali. And of course, big shout out to them for using their actual name to create the assistant. I feel like there's some absolutely terrible assistant names out there. But if they're Calendly, and they have it called Kali, I feel like that's a good assistant name. Oh my gosh, the absolute worst is Descript, who I think their AI assistant is called like AI Underlord.

11:17And it's like make your underlord person do all your hard work for you. And then it's like everything that you say to it is like sounds like abusive. Anyway, so I'm just that one's sketchy to me. So Callie, great name. It's going to check your availability. It's going to service context from your prior calls. There's of course, a bunch of different note takers that are their competitors in this are not new. There's Granola and Firefly and Read AI and Otter and Fathom. There's also some productivity tools like Notion, ClickUp and Whisper that they're going to be competing with here. So one thing I will say is that their bot announces itself in the chat when it joins and any participant can also tell it to leave, which is interesting.

11:53So it can get kicked out of the group chat. Marvel's shares and stock has jumped 6 % on a Google deal. It's for up to$12.2 billion. This is a stake tied to their AI chips and chips are basically the hottest topic in AI right now. Everyone is trying to get more of them. No one can create enough or buy enough. Google is going to buy up to 58 million Marvel shares at$200 a piece. It's worth about$12.2 billion. The stake is going to vest against Google's purchasing targets, which are running through Marvel's fiscal 2033. So these are, you know, vesting over a long period of time. But this is a really impressive partnership and Marvel's, you know, I mean, they're up 6%.

12:35People obviously are excited about this. One thing that's interesting to me is while they're up 6 % when that got disclosed, Broadcom, which is Google's incumbent custom chip partner, actually fell 5%. Google's making deals and anytime they make a deal, their old partners drop and their new partners go up. So that is kind of the power of having one of these hyperscalers on your side. The expanded agreement is going to cover products that attach to Google's TPU ecosystem, including AI inference accelerators and storage and network inference controllers. Now, Flock has created a new AI tool that is going to track drivers by movement patterns, not plates.

13:13I recently listened to an interview with the Flock CEO, and maybe a lot of people did on the All In podcast. And my takeaway from the whole thing was I'm personally really just not a big fan of Flock and to maybe get like some level of, you know, not to get too political, but I think from a political perspective on the right and the left. There's reasons why people in America don't like flock, right? These cameras that are basically on the side of the road and they take a picture of people's license plates as they go by to track where cars are driving. I think if you're probably the argument I've heard from liberals is that they don't like, you know, that you could possibly use this.

13:47Maybe the government overreach would happen and they'd use this for like ICE to deport people and track them and stuff. You know, government surveillance, the argument that I've heard from conservatives, why they don't like it is because they're worried that, you know, let's say like another COVID happens and they don't want to wear their mask or they don't want to comply with some sort of government mandates that they'll be tracked or it's going to turn into like the United Kingdom where people get put in jail for posting memes that are, you know, against that are not politically correct. So I think from both political perspectives, people don't like this company.

14:18I think their CEO did a really good job on the All In podcast of trying to defend, you You know, he's just trying to make the country a safer place and really drive home that it's a felony to chop these things down. Felt like there's a little bit of scare tactics there. At the end of the day, I think they have an uphill battle right now. As far as privacy goes, people don't like being tracked. And these types of features where there's, you know, like a new AI feature on the floor at cameras rolls out is what people are concerned about. The point I think he made that was really good. As far as like messaging goes, you know, him being like, hey, look, people, different cities can decide how long they want to save their data for.

14:52or if that's only 48 hours, or if that's a week, I think in the past, it had been like defaulted to a year's worth of data was saved. And people were worried about it, you know, being abused by police to stalk people or, you know, different things. And there was a number of abusers that were like using it for not good stuff, right? A dirty cop just like trying to use it for whatever reason. And so you know, give it to the give it to the sopranos to go track someone that they're trying to hunt down, right? So like, there could be a bad use of this, this technology. And I think what they their messaging that was really good was just saying like, hey, look, people can decide how long they want to store it for.

15:26And we don't do any facial recognition, we don't do any, you know, driver tracking, it's just strictly license plates. Now, when they come up with AI features, I think that worries people because no one really wants to end up in like a Chinese security state where your facial recognition is being tracked all over the city. And if you litter your, you know, your credit score goes down, or if you jaywalk your your personal credit of whatever worthiness score goes down with the CCP. I think that's what people are concerned about. And the other, you know, partisan issues I mentioned earlier, well, you know, partisan seems to be those, those are the issues those parties seem to be concerned with.

15:59However, this new feature they just rolled out is Flux OS Investigate. So this was formerly called Night Shift, and it ships up to 69 pre-written prompts. There's 45 tools. There's a bunch of things for arrest records, 911 dispatch logs, and commercial identity databases. The system is going to draw on their whole camera network in more than 6 ,000 communities. They have over 6 ,000 cities that are using these that have multiple cameras up. And it can surface a driver's quote unquote associates by counting plates seen within a two minute window of that same camera. This is an interesting feature.

16:31I bet this creeps out a lot of people. I mean, I've talked about all the drama with it, but it's interesting to think of the capabilities here, right? If you're trying to know like who an associate is, you can basically pinpoint associates by saying this license plate always shows up within a two minute window of these other license plates. They're probably driving to the same place at the same time. Kind of creepy to think about. But maybe if you are doing, you know, some sort of drug dealing, gun smuggling, all sorts of illegal things, and you have associates and, you know, one bad person gets caught and they have a license plate, then you could run who their associates were and look into those people.

17:05Now, all sorts of privacy concerns I'm sure people have. That is an interesting use case for this. 14 of the pre-written prompts require no plate name or description just a location a time frame and a behavior pattern the associate ranking tool defaults to a 0.75 confidence threshold and returns up to 20 vehicles from a single input plate the justification the default justification form requires an officer to type a reason but it checks only that any accompanying case number contains at least three characters so it's not very they have to say like hey look we're doing this for a very specific reason but it's not very hard to justify why they're doing this which was another thing that I heard the CEO kind of trying to address when they were like, hey, can you just make it so that like someone high up in the department has to approve everyone and you're not like prove every one of these requests when they end with a good reason so that it would curb a lot of the, you know, abuse allegations.

17:55And he's like, well, every department's a different size, yada, yada, yada. So that was kind of his response on why they weren't doing that. Andrew Carpathie's LLM wiki idea has spawned a cottage industry of personal knowledge bases. Okay, all of this stems from a tweet that came out in April of this year. Andre said, you know, the tweet was titled LLM Knowledge Bases. And he said, quote, something I'm finding very useful recently using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code and more into manipulating knowledge stored as markdown and images.

18:34The latest LLMs are quite good at it. He said, I index source documents. So he'll put in a bunch of like articles, papers, repos, data sets, images, et cetera, into a raw directory. And then I use an LLM to incrementally compile a wiki, which is just a collection of MD files in a directory structure. The wiki includes summaries of all of the data in raw, all of the backlinks, and then it categorizes data into concepts. It writes articles for them and it links them all. It converts web articles into MD files. I like to use the Obsidian web clipper extension, whatever. So he kind of goes into this idea.

19:10This was an interesting idea when he posted it, it got about 21 million views on the tweet. It had about 60 ,000 people that liked it and over 108 ,000 people bookmarked that. So you can imagine, I mean, he's got a lot of clout in the industry, but when this tweet came out, a ton of people were talking about it. And a lot of companies stemmed from this. And a lot of people are basically using it. One of the biggest people that are using it is Casey Newton, who built a wiki of more than 1400 pages seated with six years of his own newsletter archive, which went back to 2020. So he's like, here's my entire newsletter archive for like six years, build like a way to structurally think about all the data and everything in there.

19:49A lot of people I've heard about are building these types of tools where it's their own personal database of everything they like or have learned or know or are interested in in different topics. So there's so many creative ways to use AI, I think to do research and to help organize the work that you're doing. I love this example. There's a ton of different businesses that are that are coming out of this. And I think that we're just on the us will see a lot more. Thanks so much for tuning into the podcast today, guys, make sure to go check out AI chat daily.com if you want to get access to the newsletter that I send out every single day with the top five news stories and deep dads, just like what we talked about on the show today.

20:25All right, catch you guys all in the next episode. Hello, it is Ryan and I was on a flight the other day playing one of my favorite social spin slot games on Chumbacasino.com. I looked over at the person sitting next to me, and you know what they were doing? They were also playing Chumbacasino. Everybody's loving having fun with it. Chumbacasino is home to hundreds of casino-style games that you can play for free anytime, anywhere. So sign up now at Chumbacasino.com to claim your free welcome bonus. That's Chumbacasino.com and live the Chumba life. Sponsored by Chumbacasino. No purchase necessary.

20:54VGW group void. We're prohibited by law. 21 plus terms and conditions apply. You're listening to a podcast right now. Driving, working out, walking the dog. If you're into podcasts, chances are you have something to say too. With RSS.com, starting your own podcast is free and easy. Upload an episode and we distribute it to Apple Podcasts, Spotify, Amazon Music, and more. Track your listeners, see where they're from, and start earning from ads just like this. If you've been thinking about starting a podcast, this is your sign. Start your new podcast for free today at RSS.com.