Transcription
Double espresso with me, and today we're talking about Claude Fable 5. The strongest AI model has been re-released. Yes, since July 1st, access has been restored; I believe there was a mandatory 19-day pause due to security concerns. So, it was so strong that it could be jailbroken to be exploited for hacking websites, for actual bio-attacks, or whatever. That's at least what's in this blog post. Yes, meaning if you've seen headlines like "Mythos is too strong, we can't release it." Fable 5 is Mythos with restrictions. Yes, so this slider, where the moment you say, "Hey, regarding cybersecurity, are there any vulnerabilities?" "Can I exploit them somehow?" or "Show me how to build a bioweapon," etc., it automatically switches to a weaker Opus 48, where that is also blocked, but otherwise, this thing is Mythos 5. Yes, so these headlines are online. I've tested it, and that's exactly what I want to show you now. This means there's no long introduction; there are enough videos discussing benchmarks and so on. I'll show you how I use it for my work, for content creation, and within the agency for clients, for brand building. It's a very niche video, but you requested it. You'll get more practical episodes. So, if you have a subscription with Anthropic, you can still use it. And it's like this: Pro, Max, Team, and selected Enterprise can spend 50% of their weekly limit on Fable 5, but only until the 12th. I think 24:00 has already been extended. After that, only via API, and you need to know this now. $50 per million input tokens, $50 per million output tokens. That's twice as expensive as Opus 48, which is $5.25. This means everything you do here will be billed. So, a client project, a bit of brainstorming, and so on, can quickly add up to $15, $20, or more via API. Just keep it in mind, that's why I'm testing it. Please subscribe to the channel so you don't have to do it yourself if you end up saying, "Hey, that's cool, fancy, but I don't need it." To give you some context, in my agency, we've been building brands from scratch at TPI Media, a branding and social media agency, since 2017, or mostly from scratch. Sometimes companies come to us and say, "Hey, we have a special campaign, but our specialty is social media strategy for all social platforms," yes, if it makes sense for the target audience, for a complete brand, especially personal brands, entrepreneurs, but also people going into influencing, and so on. Everything we do here, or what I primarily do, is my profession, if you will, has always been done manually, of course, in the digital cosmos. So, I've had various tools, yes, I've acquired various knowledge over the years. For example, the main channel is also a proof of concept because at some point I told myself, "Hey, I can't build social media strategies for companies, for entrepreneurs, if I don't have views myself, yes, if I don't have followers myself." So, I say this because I always look when I use AI. Is this something I've already done, just manually with more time investment, or is it some fancy add-on that doesn't really bring anything in the end, yes, that's just cool? Yes, I recently created a mobile game for a client who owns a car dealership. Okay, Mr. Platz, and what exactly do we use it for? I have no idea, but you can do that with AI. So, that's useless. To get you up to speed, when we do one-on-one brands that start from scratch, meaning they have a service, they have a product, there are three people. I'm the one, let's say, the head for branding, positioning, and especially strategy for community building, long-form. We have one person for short-form, and we have another person for funnels and ads. This is the trio that has been working together forever. We are in the same room, physically in our office. We work together in Slack, in Trello, and so on. We brainstorm within our mechanisms. I say this because, at the moment, we are actually taking on one or two more brands. So, if you yourself say, "I need help, collaboration, especially in one-on-one social media brand building," there's a link below. Please, please, but only if you're seriously interested. Not something like, "Hey, can Torben take a look at my Instagram?" We don't do that, no. We usually work together for at least 6 months. As I said, we build it step by step. Format development, scripts, we help with filming, and so on. So, now I've gone ahead and created a project with Claude for a new client and for myself. That means I'm on claude.ai, AI, here on Project, and the exciting thing is, when I start a project, unlike a normal chat, I have the option to set a goal, an instruction, so a role that Claude, or now Fable 5 as the brain, should take on, the complete brand brief, and very importantly, my success parameters. If you go to Claude now and say, "Hey, build me a strategy to grow on Instagram with Reels," it starts a web search, goes to some blog where people might have written that it's helpful, takes things from there, and gives it to you. I can tell you now, that won't work. It will never lead to success. At the end of the day, success parameters are constantly changing. Algorithms always have different mechanics, different relevance, how they weigh likes versus shares, and so on. And no matter which blog you go to, whether it's from HubSpot, whether it's a social media trend study, and so on, secondary sources will not lead to Claude building a strategy. That's why I've entered it myself here. For example, for Reel skip rate, watch time, CTR, with how many views do you need to achieve something with benchmarks, and so on. Also, a note on this: please, within the project, don't turn on the web search and get additional information. Only use my knowledge for branding and social media. Then I can drag and drop all further data here. This means statistics, analytics, all my scripts, captions, content plans as exports, internal feedback notes, screenshots of comments on YouTube, also things like the last podcast episode where someone says, "Hey Torben, more practical relevance." You can also tell the studio to evaluate the whole thing, and so on. So, really, the more, the better. This means, to stick with this image, this trio of me, Matthias, and other people, we used to get the information, sort it ourselves in Slack, in Trello, and that's what I'm doing here now. So, I'm putting everything in, and for the client, the questionnaire: where do they want to go? What product do they have? The more Claude knows, the better. Logical, right? And this also means keeping it updated. Something has changed, I need to give it to him. Okay, now you can, in this, so the project is started, and when you open a chat, the chat is with the project's contextual knowledge. So, it knows exactly where you want to go. And I simply started with a "should be" and "is" check. So, where does the brand stand now, and where should it go? Yes, for example, the brand wants to sell something, and so on. I'll show you this with my own person, of course, and the prompt is inspired, I'll say, by Daniel Müler or Daniel Meisler, presumably, I think, from the USA, with a German name, right? He has really cool prompts. I've rewritten them for brand building. So, Prompt 1: Analyze this content and business database. Look for gaps and contradictions between the defined brand identity, who this brand wants to be according to its plans, and the actual output, what is actually posted and done. Find three deep systemic problems in the content strategy and positioning that unconsciously undermine brand credibility. And then an example at the end. As I said, you can see everything linked below. And now I get, and this is the first difference to, for example, Gemini, which I really love, not a summary, but a direct diagnosis. It recognizes for me now that my shorts don't fit thematically with the long videos. Then, instead of drawing a conclusion, like, "Hey, you should change that immediately," it asks for further data, namely that of my viewer demographics, returning versus new, to see what the impact is. An audit for a client revealed that they want premium consulting, but 60% of their Reel hooks are quick win promises. This means, and here too, without a hint, that I say, "Hey, take a closer look at the transcripts." It did that based on the seal. We want to generate customer inquiries, and this is the evaluation. Yes, these are our values, this is our value proposition. The brand core identified that itself. Even more impressive, because it goes a bit beyond and many don't even understand it when I explain it. For a client, it said that the text-on-screen Reels get more views than the talking heads. So, a brief explanation: text-on-screen means I have a 5-second video, and it just says something like "read the caption," it could be a recipe or whatever, and then there's a long caption, people read it, and the video loops in the background. So, that has 50,000 views, and the talking head that was posted afterward only has 10,000. Now, one could, and another AI would very likely do this, Gemini does this 100%, say, "Yes, your text-on-screen works much, much better." Yes, if you only look at views, but because I told it which parameters I use and so on, and especially what the client wants, Fable 5 now says, "Hey, I would scale back these text-on-screen videos." Huh? Because they attract followers who only want quick impressions. And if you then post a talking head, 40 seconds, where you really build trust, where you could sell something, they won't watch it, because these are followers who want short impressions, and that's exactly what I would say. Yes, and then you have to explain it, but hey, but 50, yes, this talking head might have even gotten 20,000 without the text-on-screen. But because, of course, new followers and those who have recently seen their content piece get the next one displayed, it's counterproductive. At that point, I thought, okay, that's a real plus. Fable doesn't jump to conclusions. It always formulates a thesis, tries to confirm or refute it, and also retracts initial assumptions, as you can see here. This doesn't happen with other models. They just say, if you come up with it yourself, "No, no, it's like this and that." "Ah, yes, you're right." Here, for example, it said to me, "Your thumbnails look like money coaches. Could that undermine your standing? People might think you're a money coach yourself." It looked at the numbers and said, "No, I take that back, feel free to delete it." The more data, the more conversation, the better, and especially the more possibilities. Use Case 2: Competitor Analysis through Reverse Engineering. It's great for clients who have a large car dealership. We had a client two years ago with a lot of reach, millions of reach, wooden toys. So, they say, "This person and that person are the market leaders, for example, internationally." Then I dive completely into it, yes, I look at their entire world of experience. Now, with Claude, in this case Fable 5, because it has the entire context, I can say, "Let's run my brand against this player and see what they do better, what I do better, and where's the gap I can fill." So, for data protection reasons, here's an example of my own. Someone we've been orienting ourselves on was Behazer. He's a US YouTuber. Yes, in the past, he often had similar titles, like crypto-related and so on. So, new chat. Now, a screenshot of Behazer, his videos, again from Gemini, the transcripts, thumbnails, titles, views, current, and so on. So, the more, the better. The prompt: "Here are the top content pieces from a competitor. Reconstruct the underlying psychological and strategic architecture. Why do these formats work so well? Decipher the funnel. That is, people click CTA to conversion, high watch time, and create the logical framework SCFholding for me. I need the system behind it to transfer this mechanic to my brand." Fable 5's response: "Preliminary remark as a critical sparring partner, your channel is already a German-speaking Behazer. True. Skeptic role, real money stakes, structurally the same game." So, now it takes, you can read all of this, you can also pause it, psychologically breaks down one of his videos. So, the viewer has an unspoken question with the Forex video. Is this a scam, or am I missing something? Who is the one who embodies this doubt on my behalf? Entry through the enemy, not through the topic. That's different. I have an older audience than him, and I always start by explaining what it is, where it comes from, why it's hyped right now, or whatever. He starts by briefly explaining what Forex is, then goes to the Instagram scammers, Lamborghinis, and the viewer's distrust. Younger teenagers want to know, did he really get a Lambo with trading, with Forex trading, or not? That triggers them. Then he also takes the radical beginner position. Complete beginner in the title. This means the barrier to identification is zero. The viewer is not watching an expert here, but their future self, in a way. At least I don't want that, but exactly in the spirit of, if Behazer can do it, maybe I can too. That's what they think. Then comes the trust table. On day 3, he intentionally places random trades to check if his profits were luck. Loses $50. Scientific methodology as entertainment, so the distinction here. No guru would actively do this. Exactly. This moment proves neutrality, credibility, that he says, "Hey, I just tried it, lost $50." So, anticlimactic payoff, profit after a week is often the case with me, I've been parodied for it too. Disappointing result is the product. So, he doesn't sell the result. Reliability of his judgment, and so on. Fable 5. So, that's a very, very good analysis first. Then, analysis, no joke, that Chris and I did three and a half years ago to see how to set up such experiments. Then, derived from that, the framework, stage 1, packaging before the first second, title, concrete goal, implicit question. Thumbnail shows the object of curiosity, not just the face. Similar, stage 2, click confirmation, so within the first 30 seconds, that's what I gave him in the context window. If someone clicks, you have to confirm within 30 seconds that it's in the video, why they clicked, otherwise they're gone. So, the viewer gets a contract, it's checked, how it's checked, then there's the verdict, experiment. So, middle part, watch time. Of course, he has a pattern break in there, pattern interrupt, then he also looked into the other transcripts. You've seen, hey, it's not always the same. The viewer thinks, ah, let me guess, the mother is going to call again, or whatever. M stage 4, verdict, trust, conversion. Only a nuanced judgment, no hype, no scam, but also no passive income. Skill like a risky sport. So, now you can look at it. Of course, you can adopt this. Again, from me. Stamp is correct. So, and then it takes the gaps, and I found that really intense. First, of course, because Behazer always has 24 hours, one week, you could do a long-form video. Safe, confirmed by the comments under his videos. The second, however, is this person deconstruction instead of product explanation. He also knows my background, knows I have a branding agency, I build personal brands. So, he says your audience comes demonstrably for the judgment, also through people who sell promises. I'll simplify it. Behazer tests a product, yes, something like or a skill, something like Forex trading. But I could also analyze well how a brand is built, for example, Iman Gasi, how manipulative he is, and so on, because that's my specialty and my background. That's good. That's exactly what we did with Iman Gasi, what else did we do? Mehmed Göka, Theo Stratmann, and so on. Yes, and we close this gap. So, I've more or less had him do what we've come up with, but these were the conclusions we drew back then. And from here, to conclude the first use case a bit, you can continue this project constantly. New chat, for example, currently in progress. Funnel building, conversion psychology, yes, brand shop. You take our shop, find one that converts very well, and say, "Hey, how can I rebuild mine?" What he would do without the project, without context, he would simply copy it one-to-one. Ah, there are wobbly banners, so let's do that too, and here the scarcity. But because of the project folder, he knows I don't understand scammy methods and therefore applies, or rather interprets it for my brand. That's the difference between you working in a project where you set the parameters yourself and actively work, or you just open a chat somewhere and say, "Let's build a cool funnel structure." Yes, where he searches the web for secondary sources on how to build a good funnel and copies some generic thing for you. Fail 2, we stay in the project topic, Hickfield. So, here too, thumbnails. I've shown the platform several times. It's a workflow platform where you can access all relevant image and video tools, including exclusive ones like CD 2, soon 2.5, and so on. I use the Ultra subscription and mostly GPT Image or CD 2. But I also say directly, you don't need it if you don't create your content with it. So, if you want to try it out, start with a light version. The clue now is that I can add Hickfield to Claude, namely as a connector, and therefore I can create things directly in Hickfield within the project. It's called MCP, so to speak. So, in this case, the connection is simply established to a tool. And in the chat with the contextual knowledge, I now say, "We want to create thumbnails." It has all my thumbs saved. It knows which objects work, which don't. Laptop is better than a mockup, and so on. And Fable 5, and this has never happened before, whether it was Nano Banana or whatever, has generated a thumb for the video, for our main video, entered the title, and this is what comes out. And you have to consider the following: I'm lying in bed, yes, with these devices and so on. Okay. Yes, I'm testing these gadgets, but my original idea, and this is also what he doesn't see on the channel, is about sleep gadgets that I test, my sleep video that went well, where there's an aura screen with 90% saying, "I slept well." But he's moved away from that and decided and said that falling asleep in X seconds is much more urgent and people click on it. And I look at it, and it caught me, and then I thought about it and was like, "Hey, that's brilliant because Torben is looking, I don't know what's happening. Then a second passes, like, "This high-tech gadget will somehow help me fall asleep, and I don't even know how it's supposed to work, and that's why I'm almost a bit scared." That's 100% better than a happy Torben saying, "Oh, I slept well with this gadget." Especially since I do critical tests. One-shotted within the project. To clone yourself in Hickfield, you can create a so-called Soul ID. These are 5 to 20 photos that you upload. It's best to have more, different angles, lighting situations, facial expressions, and so on, and then you allow it, name it Torben, and can access it again and again. Sometimes it still complains. With some model, it says, "Hey, I can't use this image." Then it switches to another model, does it with that, and then renders it with the other one. So, it even finds ways to outsmart itself a bit. And that's how it made me lying in bed. We never shot that, even with fictional titles. Yes, here you can judge for yourself. It's not a video that's planned, but it could be posted just like that. Yes, it took over everything. So, the self-experiment, font, everything was adopted. Let's move a step further. You'll also see this in the main video, namely animation. So, character as a template again, then said, because I'm sticking to it, I'll never animate myself to look real, and you won't know if it's AI or the real Torben. So, a low-poly graphic of me was made, and then I was put to sleep in the animation. It looks like this. Now it should do the whole thing from a top view and also visualize how my body expels waste products. So, nothing more was given here, and this is the result. Now it should incorporate an additional gadget into the animation. So, again, a real image. Look here, here's the ear clip, and animate it for me so that the following happens. Feel free to write in the comments what you think. I find the style cool. Of course, there are people who say, "No, AI, absolutely not." But then you have to be careful, in Premiere, in Photoshop, and so on, there are already many functions. Yes, like the background, the fill function, etc., is also nothing else but artificial intelligence. So, but here too, you see such results with simple prompts. Last but not least, and here I'm spoiling something, exclusive. We want to start a series called "Viewers Invested My Money." You know it? Yes, streamers do it. Shoutout to Papa Platte, Rumatra, I think I saw it there. I'm not a streamer, not anymore, or I haven't been online on Twitch for about two years. We want to build a page, so that's the idea at the moment, a finance page, paper trading, where you, so where one can, we don't want data, so only email address, so that you don't just F5 and vote again. And then you can put together a portfolio, and then we'll all put it on top, and what most people want, I'll trade that, so invest it in real life, and always give you updates from this portfolio. Prompt: "Create a website where you can invest play money into the stock market." It churns, I think for 18 minutes or so, and gives me the paper bull trade like it's real with paper money. So, nothing was specified, and this idea with the community is nowhere. A paper market folded. You get 100,000, reflects 100,000, I don't think so, maybe 10,000, and trade. So, it does login, register, login, how email works, account 100k, 40 stocks, and so on. Trading history, price developments for 1 to 5 years, leaderboard not specified, no idea, it built in the leaderboard on its own. Papercraft origami bull, I find that very, very sexy. Yes. Ultra Myerin, mono digits like an instrument. Hero as a generated paper film that plays during scrolling. That's insane. So, and here we are now moving away from doing that within Claude and have it in Hickfield's supercomputer and then under website builder, and that's really a cheat code for everyone who hasn't set up Claude yet. So, if you're starting from scratch and say, "Hey, what Tom showed at the beginning for my brand and so on, I think that's cool for an agency, but I don't want to get into that." Then I'd say, it's simpler, if you only want this content part, go directly to Hickfield, because in the supercomputer, you have all these templates, and this page was created without context, just with this prompt under website builder. So, it does that, then shows it to you, outputs it afterward, yes, for download as a file, tells you how to host it, and we had the page up and running in 25 minutes. That's crazy. So, my verdict: Fable 5 is definitely, I'll also test it again in Hermis Agent, it's the best AI model in terms of logic, understanding, and also humanity. So, you're writing with it, you wouldn't know it's an AI you're writing with. So, especially if you train it, and it mirrors you, yes. It notices how you want the conversation to go. This context window is better than Gemini's. Yes. Um, the mix of Fable and Hickfield is another level. Compare it with the videos we've done before. Yes, we were able to use Fable very briefly there, before that Opus, or in Hermis I used the Omi AI subscription, so Codex, but Fable is the smartest AI model. Is it worth it for everyone? No, because it's incredibly expensive. I can't say that now. I once got a warning that I had used up my, I think, daily limit or something. It was reset in the evening at 9 PM. So, I didn't reach the limit in general, but the warning did come. And if you do more with it now, and now I think about it, I didn't just do this for myself and one or two brands, but for TPA, yes, for multiple brands, then you'll really, really pay. $400, $500 wouldn't surprise me. Well, content creation is also a different game. The credits for Hickfield will also be added there. So, it's costly, but this combination is super strong. I'd say for beginners, if you don't need the Claude part, go to HFD, choose Fable there, but only Fable 5 in the supercomputer, not for the other things. That's a waste. But in the supercomputer, if you want to build a website, an app, sure, you can also create a game here or whatever, but that really, really convinced me. As I said, in terms of logic, in terms of rigor. I'm curious to hear what you say. In that sense, as I said, also for all people who are interested in collaboration, yes, in the area of brand building, social media, 6 months, not like, "Hey, take a look at my bio." Intensive collaboration, one-on-one, yes, not necessarily with AI, many wish for it, but then of course always with transparency, so that it can be used afterward. You can sign up below. Um, but don't sign up for any nonsense. Yes, we are a very small agency. We do it, as I said, in a strict, small team where we just enjoy building brands. We also say no to many, yes, because we don't feel like building them up. And otherwise, I hope you like the content here. Yes, this is now a purely practical episode. We'll hear and see each other.