← Escape the Technocracy
Loading summary
Transcript39 lines
- [00:18]
A
Hello and welcome. This is Gabriel Custodiet of Watchman Privacy privacy practitioner, consultant, author and frontline fighter in the push for privacy. I know why you're here. Like the rest of us here in the resistance, you're trying to escape the technocratic apparatuses that you see enveloping you and crushing your freedoms. That's why I created all of this, all without sponsors. I hope you enjoy this show. But then when you're ready to take the next steps to secure your privacy and your future, Visit my website, escapethetechnocracy.com to start the real journey. Your support alone does determines the future of the show. See you there. Welcome everyone. I'm very pleased today to be joined by Daniel Semper Pico. Some of you might know him from his previous handle, BTC Gandalf. And Daniel has been talking about a lot of AI things recently. So we're going to have a nice discussion here about some of the latest tools with AI, some strategies for using it. Just have a more general conversation about AI. And Daniel, welcome to the show. How are you doing today?
- [01:20]
B
I'm doing very well. Thank you very much for having me. I'm excited to do this.
- [01:23]
A
Maybe we could just start by having you introduce yourself a little bit. As we said, a lot of people know you as this character BTC Gandalf years ago talking about bitcoin related things, dressing up as Gandalf. Just tell your story a little bit and to the point where you're now a big proponent of AI and talking about this pretty consistently.
- [01:44]
B
Yeah, man, it's crazy to hear the phrase like that. I, you know, I'm known as Bitcoin Gandalf from years ago because it feels like a very recent part of my life. But it has been like five or six years now. So before, before I got into bitcoin, which is if anyone does know me, that's probably how they know me, as, as BTC Gandalf. Before I got into bitcoin, I was a, I was in the real estate business in Hong Kong. So I was a real estate broker and I owned a real estate business and I did that for about eight or nine years. In 2021, my wife and I decided we wanted to leave Hong Kong. I'd sort of been bit bitten by the online business bug like I wanted to do. I wanted to build a career where I didn't need to be in any given place at any given time. So I could work virtually from anywhere in the world, with anyone in the world. But I had sort of no idea what that might look like specifically. I just knew that I needed to do something on the Internet. And so my wife and I decided to. To leave Hong Kong. And around that time is when I discovered bitcoin. And I just opened the Twitter account that some people may know me for, the BTC Gandalf Twitter account. And I started tweeting about bitcoin and I just happened to hit like, you know, the right topic at the right era of Twitter, that I started getting a lot of followers and started becoming known in the space. And I ended up working for a few bitcoin companies mainly in the content, basically writing words online, if we were to put it simply. And then I always wanted to get back to doing my own thing. So at the end of 2023, I left my main gig, which was with the bitcoin mining company Brains, and went out on my own again and eventually transitioned out of this pseudonymous bitcoin Gandalf character into, you know, going online under my real name. And yeah, of course, like everyone else, I've become enthralled with AI and started writing about it. So it's pretty much the same as like with bitcoin is I just tend to write about whatever I'm interested in. And right now that thing happens to
- [03:54]
A
be AI And I'll say for the listeners who don't want to check the show notes where it will be there. The Twitter. The X profile is Dan Sempere Pico. That's D A N S E M P E R E P I C O. Are there any other places where people can track down what you do?
- [04:11]
B
Right now, Dan, it's mainly Twitter. I'm going to start YouTube soon, but for now, if I do anything else, I normally talk about it on Twitter so they can find out about it from there.
- [04:23]
A
What was that like? You were going by this pseudonymous and then you decided to go by your real name. Do you, do you regret that at all? What was that transition like?
- [04:33]
B
No, I don't regret it at all. I mean, sometimes I miss, you know, I miss the Twitter era of when I was BTC Gandalf, but I don't regret the change. If anything, I regret not doing everything under my own name from the beginning. Although it's hard to say whether I would have had the following. I think being called Bitcoin Gandalf had a lot to do. Like, you know, the mystique and the pseudonymity actually played to my advantage. But if anything, I kind of wish that I'd started tweeting under my own name and didn't focus so much on only bitcoin because in a way, you become a little bit captured by your audience. Like, if you build it, if you only talk about one thing, then everyone who follows you is obviously following you because they're interested in that thing. And like, you know, with anyone, or at least with me, my interests, you know, I get obsessed about something. I might, like, dive really deep into it for a few years, and then eventually, like, something else comes along and I get obsessed about that and I dive deep into that. So if anything, I wish I'd spoken about more things and done it under my real name from the beginning. But the reason. The reason I went pseudonymous was it was kind of twofold. Number one, I was in a. In a situation at the time where I didn't want to be associated with bitcoin. I won't get into the. I can get into the background details of it if you want me to, but it's a. It's a. It's sort of a tangent. But I didn't want to be, you know, I didn't want to have my real name and identity associated with bitcoin at the time. And I also remember from being in, you know, quote unquote, crypto Twitter, lurking on crypto Twitter in like, 2017, I remember that there was a lot of people that I followed that were also pseudonymous. The main one that sort of inspired the name Bitcoin Gandalf was crypto Cobain, who's now just goes by Kobe. That's. That was like the four. That's basically the format I based my name, my pseudonymous name, choosing on. So, yes, I don't regret. I don't regret transitioning out of the BTC Gandalf character.
- [06:38]
A
Do your old bitcoin followers follow your AI content, or have you had to rebuild your audience?
- [06:44]
B
Yeah, it's. It's hard because, well, it's. I think the Twitter algo has changed a lot recently. So it's very hard to say whether the. You know, I've definitely had a drop in impressions. And even now when I tweet about bitcoin, I. My average bad bitcoin tweet will get more views and impressions than, like, a great tweet. I think, in my opinion, about something else. There was a moment where my AI stuff was getting a lot of. A lot of views too. There was a moment, like during the Iran. The Iran war. I live in Dubai, so when the US attacked Iran, there was a moment where the tweets about Dubai and the Iran war were doing really well, and I was like, getting millions of impressions in a month. But, yeah, it's really hard to tell whether it's like, the topic switch. It's hard to tell whether it's the name switch and people don't recognize me anymore. It's hard to tell whether it's the Twitter algo. I certainly, like, I've been going to some old profiles of people, you know, that I remember from back in the day on Twitter. And I look at, you know, they might have had tens of thousands of followers, and they're also getting basically no engagement, no likes, no views on their tweets.
- [07:55]
A
Yeah, I did want to ask you about the jurisdiction question. One of your pinned tweets on your profile is that you've lived in Portugal for two years, Brazil six years, Hong Kong, 25 years, UK five years, Isle of Man, two years, Dubai for two years. What has that been like? I mean, some of those jurisdictions, when I see those, it's, you know, bitcoin friendly, crypto friendly, favorable tax regime. What has your experience been like living in all these places?
- [08:23]
B
Yeah, so a few of those are just where I lived as a kid. My. My dad worked for a Spanish manufacturing company, so he moved around quite a bit. So Portugal, Brazil, and initially Hong Kong was all because. We moved because of my dad's job. And then I obviously, we really liked Hong Kong, so I ended up living there even through to being an adult until 2021. I lived there for 25 years. And, yeah, that was incredible. You know, I kind of. I would say, you know, I'm originally from Spain, but I would say Hong Kong is the place that I consider to be my home country in a way. And, yeah, I went to University in the UK and in 2021, my wife and I decided that we wanted to leave Hong Kong. She's from the uk so we went back to the UK and it was literally making the decision to go to the UK between making the decision to go to the UK and actually moving is when I got into bitcoin. And then I realized, like, a place like UK from a tax perspective isn't going to work if I want to be able to use bitcoin as money, because I don't want to be keeping track of every bitcoin transaction that I ever made and whether I made a gain or a loss on it or whatever. So we ended up looking at other jurisdictions that we could live in, and we came across the Isle of Man, so we Moved to the Isle of Man for a couple of years and then ended up deciding that we wanted to, to move from there and came to Dubai. But again, yeah, always, you know, one of the big things has always been like taxes have always been a big consideration because of Bitcoin. Because I do my own thing, I'm an entrepreneur. Like I've sold a business. So those, those kinds of things have, yeah, they've always been a consideration for me as an adult of where I live. I don't think I, you know, my personal opinion is that for most, in most or at least in the majority of places that I'm aware of, I don't think the taxpayer gets a fair deal on their tax dollars.
- [10:21]
A
Now you're in Dubai currently. There were some videos during the Iran war of Dubai being basically evacuated. Nobody's there. What has that been like? What's your experience? Been living in Dubai of late?
- [10:33]
B
Yeah, it's good. I mean we moved here in February 2024. We were here throughout the entire Iran conflict and at least for the first time almost, actually it was, yeah about six weeks. The UAE in general was the most targeted country by Iran. And so you know, there are many days where we had multiple missile alerts on our phones. But you don't, you, you just like being here, living day to day life, you didn't really notice anything. It was if, if they hadn't had the missile alert. You as a resident, unless you happen to be nearby somewhere where a, there was a drone strike or a missile, you wouldn't have known what was going on.
- [11:15]
A
Well, let's get into some of your bread and butter then. Let's talk a little AI. We can start with some basic questions here. I'm curious your thoughts on how fast AI tools are advancing. They're changing very rapidly. What has your been experienced, let's say go back a couple years to where we are now, how different it is as a user experience. For example, prompt engineering, where you kind of set up the LLM generate a good answer for you, that's kind of gone. What are some of the changes you've seen in the last bit of time? Just from the, the basic user experience of let's say text AI.
- [11:52]
B
Yeah, I mean I think the biggest, the biggest jump has been the fact that it's been agentic AI like text AI was a better version of Google with, you know, the ability to generate some, you know, written, written text that you could maybe pass off as your, try to pass off as your own content as your own words. But you had to be constantly there, prompting the AI, taking the output, you know, pasting the output somewhere else, creating your own file, posting it on social media. And then you know, with, with the explosion. I mean I attribute the explosion of agentic AI in let's, let's say to like, you know, the average person has been. Because of Open Claw. It was really like beginning of this year when people, when OpenCloud became like the most, you know, starred or whatever, most used open source thing or fastest growing open source software ever. Like that was a mind blowing moment. And I think that changed everything for AI because now you don't. You mentioned prompt engineering and stuff like that. Like you used to have to be so specific and careful with the way you steered and prompted your AI to get a good output. And, and now it's almost like the less, at least in my experience, it's like in a lot of scenarios it's almost the less you tell the AI or the, the more like room you give it to pick the best path to execute on what you want, the better it is. So it's like you almost just tell it what the outcome you want and let it pick the way it does it rather than trying to like spend an hour creating this super elaborate, very specific prompt on exact telling it exactly how it should do something. Like. So I think that's the big, the biggest change I would say like has been that number one, it can control your computer and do things for you. Number two is that it can now pick the best way to do something versus before you sort of having to tell it the best way to do something.
- [13:54]
A
That's fair. That has been the big development recently we've done a webinar on OpenClaw and we, you know, we've been talking about AI quite widely, not just as a privacy and sovereignty tool, but for sheer productivity maybe since you brought up the agentic AI, which would be openclaw and other sorts of similar things. Could you just describe for people your setup, like what is, what are these tools doing for you at the moment so people can get a sense of, you know, how far you can go with this.
- [14:23]
B
Yeah. And actually I just want to touch you. So you mentioned like the rate of how fast everything is changing in AI. And I used to try and keep up especially at the beginning of this year, like I was trying to keep up with every Open Claw release and this new model just dropped and this new tool and that new tool. And then I realized like I can spend almost all of my waking hours trying to stay up to date with this stuff. And I still feel FOMO and struggle to do it. And so like people who, for, for whom AI is like just a tool for them to do their job. It's not like they're not following every single, like they have no chance. And you realize like at the end of the day, because the AI is almost like the first software where you don't even have to know how to use it, you can just ask it to tell you how you should use it. And so you don't need to know any like, you know, knowing about, you know, commands and skills and Claude MD stuff. It's like all of that might be relevant for like a couple of weeks right now, but later it won't be like, you'll just be able to tell the AI what to do and the AI will decide if it needs to build itself a skill or add it to the CLAUDMD file. And I guess openclaw was kind of the first one to do this where it was like, you know, just remembering stuff without you telling it to. And it just like felt like it worked. How we would imagine that, you know, this all powerful, all intelligent AI thing would be because remember, like, at some point ChatGPT didn't even have a memory and then it had a memory, it was still really bad and couldn't like, you know, remember stuff across chats and all that. So I think like, trying to keep up with the latest developments is not the greatest use of people's time because things are changing so fast and because we're. The end game is going to be that all that, all that stuff gets abstracted away anyway. All the skills and the prompt engineering and the Claude MDs and the commands and whatever, all that stuff will get abstracted away. We're heading, I think, to a world where you just tell the AI what you want and it does it, and it does, it handles all the technical stuff. It's like, you know, kind of like how you don't need to click around in a UI anymore because you can just get an agent to do it, right? If I, if I want to, like at the moment, if I want to, let's say I need to like send an email to someone with like a bunch of documents. Because, like, let's say I'm doing, I'm applying for health insurance and I need a bunch of documents, right? I don't go clicking around my mess of folders in my computer and trying to find like my passport and this document and that medical record and whatever I Just go into Codex and say, hey, can you draft me an email to this insurance company? Check the email and see what it documents they're requesting and go find them in my, wherever they are and attach them and it does it for me. So like all the technicalities will get abstracted away. I think all the like pressing buttons in a UI for the most part is going to get abstracted away and you're just going to. Well, you can already do this, which is like you just tell the AI what you want and it doesn't. And sorry, I know I may have gone off on a tangent on the like fast changing stuff and you asked me a different question. So we can go back to the, back to your latest question. Oh man. So the only thing I use OpenClaw for now is I have a health and fitness AI bot. And so I'm trying to get in good shape so that I in there, I just tell it what I eat every day and it tracks my workouts as well. So when I go to the gym it knows, you know, I've given it like on Tuesday this is the workout that I do. On Friday I do this workout. On Saturday I do this workout. So when I'm at the gym, I message my Open Claw bot and I say, give me my Tuesday workout and it tells me what I have to do. And then I log, you know, my workouts in there just with text like, you know, dumbbell curls, this much weight, this many reps. And so, and it keeps track of all of that. So I can ask it like, what's my, you know, PR on bench press? How much stronger have I gotten on my bench press since December last year? And because it has all of that logged, it can tell me. So that's the main thing. I use Open Claw now and mainly because like, because I was using Telegram. Sorry. Because I was using Open Claw through Telegram almost like as the only AI tool for like six or eight weeks between February and March this year for some reason I've just built this habit of opening Telegram and, and going there to tell the stuff. So I'm mainly remotely. So that's why I still use OpenClaw for like my health and fitness. But when I'm sat at my computer, I mostly at the moment use the Codex desktop app. So that's, that's the app that I use for everything else. So I've got, I do some copywriting for clients. So I've got a project in there which has, you know, which is trained on my clients blog posts, Twitter posts, etc. And that's what I'll use to help write copy for them. I like I said, I use it for a lot for admin stuff. I use it a lot for learning as well. Admin stuff is the best because you know who likes going. I don't know about you guys, but like my, my file system is a complete mess. But it can be because the AI can find anything. It doesn't have to be organized. So I love that now it can do all the boring admin stuff where it might take you 20 minutes to get 12 documents together to email to someone for some random admin thing that you need to do. So I use it a lot for that and yeah, I'm enjoying it a lot for learning. So now I read a lot of books about copywriting and the way about writing in general. I just love, I love writing and the way that I apply the lessons from the books that I read now is I will write and then I will take my writing to a chat in Codex or ChatGPT or Claude. You can do this in any, it doesn't have to be agentic either. And I just say can you help me implement some of the lessons from this last book that I just read about writing to make this piece of writing better? And that way like I'm editing my writing using things I've learned from the book I've just read or like reinforcing lessons I've learned from a book I've just read.
- [20:55]
A
Getting into some of the specifics of the different companies out there, the different models out there. How do you evaluate a model? What are the ones that you like and what are you looking for in a new model or a new version of a model?
- [21:09]
B
I mean the main thing I use them for other than Agentix tasks is writing. So I look for it to produce writing that doesn't sound like AI slop or that doesn't insert. Because I'm at least for my own writing, I'm rarely generating text from scratch with the AI. I normally will write something myself and then I will use the AI to help me edit it. So it's the way it writes. So codecs is. Or the GPT models I find are not great at writing. For me, my favorite models for editing my writing are Gemini. I find that Gemini writes the most natural sounding text. Opus 4.8 or generally the CLAUDE models. The anthropic models are pretty close behind but. But Gemini still comes out with, you know, I might go through the same piece of content with all three of them to compare. And I might be trying to like get them to rewrite a sentence or iterate on a sentence over and over and over again. And it gets to a point where you can tell it can't come up with any more variations of that, of saying that, that sentence. And I'm still not happy with how it's, you know, with how it's written. And then Gemini will come up with like a completely new way that none of the other LLMs have come up with. That actually sounds amazing and nails it. Like I've had several moments where I'm like, where Gemini has like floored me with how good it is at the options it can come up with in terms of like editing and rewriting certain sentences and paragraphs in my writing. So I like to use Gemini the most for writing. But yeah, I look at the way they write and then I look at like agentic again. I think Gemini is worse at agentic stuff. Like I've had it. I can't remember at one time I was doing something with Claude Code and with. It was something in Gmail. I think with Claude code and with Gemini CLI and Claude code was able to do something in Gmail faster than Gemini. So the other thing is just like how good is it at Agentix? Agentix stuff. And then very occasionally I might use it to help me design something or create something, put something up on my website and for that front end is better for that. I care about the front end skills and I find Opus is the best at front end. Codex is not a great designer.
- [23:40]
A
I wanted to ask you another practical question here, which is the cost of these tools. So that's always been a little bit of an issue for some people. And when you start using AI as an agentic tool, as we've been discussing, where it's essentially running on its own device 247 in some cases that builds up a lot of, you know, a lot of credits that can, that can certainly if you don't set certain limits, if you're not careful about which model you're using, it can rack up a significant bill. I'm wondering if you have, you know, any tips or, or thoughts on how that financial aspect can be managed.
- [24:20]
B
Yeah, I mean it really, it really depends on the sort of constraints you put on yourself in terms of like what models you'll use and why you'll use them. And if you hate a sudden, you know, some people are like, I will, I refuse to use OpenAI or I refuse to use this company. I, for me they're for agentic use. The OpenAI models are good enough and they let you use the subscription, the Chat GPT Pro subscription on third party tools. So those are heavily, heavily subsidized tokens which means like the companies are selling you those tokens at a discount. Like if you paid for the equivalent use of a ChatGPT Pro subscription using the API, you'd be paying a lot more than the hundred bucks or whatever it is that the GPT Pro subscription costs. And same with the Claude Max but sadly Claude Anthropic don't let you use their stuff on, on third party applications. So for third party apps right now I just run them on through my Chat GPT Pro subscription because I think I, I think I have the a hundred dollar one. I've never hit the limit ever. So I mean I'm not doing like, you know, I'm not coding applications or like using it for anything that's like 24, 7 running all of the time. I have a couple of, of scheduled tasks that are content related but nothing that's like you know, churning through tokens all the time. So for that, for, yeah, for the third party apps I use Chat GPT Pro and then I'll occasionally, occasionally sub in Gemini which I just have the Gemini that comes with Google Workspace. I use Google Workspace for my work email and that comes with some level of Gemini which I again have never hit the limit on. And then I think I also pay for the Claude Max $100 subscription which again I, you know, because I have all three of them and I use, you know, I will use them for different things. I never hit the, the limits on any of them. So let's call it, you know, my Google workspace is like 20 bucks a month and, and then I'm spending a hundred on Codex and 100 on, on Claude. So 200 bucks a month it you can go, you know, depending on your usage. You could just go with like the, the, the 20 buck subscriptions. I was on the $20 Claude subscription for, until January this year, you know, so I was paying like 20 bucks for a charge. 20 bucks for Claude, 40 bucks. It might sound expensive like compared to other software but if you compare it to the fact that it's like it can replace you hiring an assistant or it can like the amount of, the amount more work that you can get done with those tools is so massive that it's. Yeah, I'd probably pay $1,000 a month if I had to for you know,
- [27:10]
A
one of these subscriptions yeah, that, that totally makes sense. I think a lot of people will see the hundred dollar a month price point for some of these and roll their eyes. But if you are really leaning into this stuff, that's probably something you're going to have to do and that you're going to want to do for sure. So yeah, just kind of change the way you think about this now. We're, you know, I'll have some questions about your concerns about privacy related things in a minute. But, but assuming we're just pushing things to the max for the moment, I'm curious how if you've implemented other AI tools into your life. For example, there's a device called the Plaud, the P L A U D. It's like a credit card shaped computer. It has a microphone, it's listening at all times and then it processes everything you said during the day and you can ask it like, hey, I had a consultation, could you, could you write an email based on that consultation I had with this person? It's basically listening at all times. Curious if you're familiar with the Plod and otherwise if you're just really pushing all the AI stuff to the max.
- [28:09]
B
Yeah, so I had last year, at some point, I think it was last summer, I bought the Limitless pin. So it's a pin you wear and it can listen to everything you say. I think it has a. But it has a button which you press to have it listen and then if you haven't pressed that button it is supposedly not listening. But who knows really if that's true or not. That got purchased by Meta and then got shut down. And then actually just recently I got Applaud before that because I was using OpenClaw more because I was using OpenClaw almost exclusively. I was able to just like voice note stuff to it. Um, I didn't have it listening to like, you know, meetings and stuff. So I have the plot on the back of my phone. I don't really have a lot of meetings with people. So it is just there for. Whenever I have an idea I just press the button on the Plot and speak into it and then that gets saved into the Plot app. And then in terms of other tools, let me just think. I use Perplexity for search. I just again, it's like it's probably not necessary but it's just a habit that when I have a question about stuff, I like to use Perplexity. I like the fact that it just defaults to searching the Internet and I find it's just Like a better Google. So I use Perplexity, I use the Plaude. Think that's pretty much it. Notebook lm. I use that sometimes as well. But yeah, that's pretty much it. In terms, I mean in terms of privacy concerns and stuff. I used to, yeah, I went through a stage where I was like trying to Google my life and go, you know, use Linux and use graphene OS on my phone and stuff like that. But for me personally, it just took, took so much time to do certain things that I was like, it's not worth it. I don't think I'm that important that, yeah, I took the view of like, I'm not so important that like, you know, if anyone finds out some random thing about me that I put into my computer, like I don't care anymore. I'd rather just have the convenience.
- [30:21]
A
I've seen a lot of people moving away from. I've had people I know who basically told me, yeah, privacy is not as important as it was when they, when they see the productivity of AI. Now that's not me. But let me ask you this though. So besides the, the privacy stuff, giving a, an AI access to an entire computer, it brings up security issues. Of course, we're also dealing with these big tech companies that in a certain sense we've been fighting for a while. Their vision of the world, their political correctness. What are your thoughts on those issues as we bring in these big tech companies basically into more and more of knowing our lives and what we're up to?
- [31:01]
B
Yeah, it's getting crazy because we are literally training the AI. Unless you're using Gemini via Google Workspace account, which supposedly they don't train their AIs. It's just like, you know, Google Workspace, they don't look at your Google Drive stuff and they don't look at your emails if it's Google Workspace. Because that's like an enterprise client and there's obviously privacy. You know, companies have security and privacy things that they have to take into account with client information and stuff. But if you're using like the consumer Gmail, Google's reading your Google Drive and Google's reading your emails. And if you use WhatsApp, Meta is probably reading can access and read your WhatsApp and it's probably using that to train their models. So it's like, at least I have found that it's like the trade between privacy and convenience seems to be like a direct trade off. And most people are either not aware that they're making that Trade off or they are choosing, you know, they are aware, but they're choosing the convenience. And I think there's like a minority of people that are choosing the privacy side. For me, the whole like the reason why I would go for a fully local intelligence, it wouldn't be because of privacy. It would be because I don't want to have the intelligence gate kept from me by third parties. So with the whole Fable 5 thing, like we've seen that if the government doesn't want you to have access to Fable 5 or Anthropic decides that they only want to release it to a certain number of companies first because it's a cybersecurity risk or whatever. Like I don't really want, it's like kind of like being shut off from social media or if like, you know, Google banned you from using Google services. Like they've become such an integral, they're such an important part of being able to be like a productive, functioning economic member of society that companies being able to gate your access to those tools for me is a bigger deal than any privacy concerns. And so the reason I would go local in terms of like running local models and I am going to, as soon as the new Mac studios come out, I'm going to get the beefiest Mac studio I can and I'm going to run a local model on there. It would be just to not be shut out of access to that intelligence. In terms of the privacy, I think I totally understand from a values, from a values perspective, like an ethics and values perspective that people have the stance that they don't want to share that data, even if it's not like sensitive with big tech, big corporations or whatever. My, my feeling on it is like I already use Google Drive, I already type, you know, everything into Google. I have a Mac, you know, I do, I'm, I'm not good enough at coding and whatever to know if my Mac, if Apple can read everything that I've ever said on my Mac, you know, I, we use telephone companies like God knows if they can, you know, listen to your conversations. The US suddenly has spied on citizens talking on phones, you know, as like Edward Snowden whistleblue. And so yeah, the privacy thing is not something that hugely concerns me anymore. It's more the, yeah, the access to
- [34:32]
A
the tools on that topic. So the local hosting AI, basically we are taking our own devices, our own computers and running these models, mostly the open source ones. So we're not necessarily going to be using honestly some of the most impressive AI models out there, but we can run some open source models on our own devices now. In the last few months there has been a huge increase in the cost of ram, graphics cards, things of this sort, to the point where we have trouble recommending local AI a lot more than we did six months ago. Do you think that is an existential threats to local hosted AI is the sheer cost of getting a device that can even run a decent version of these tools compared to as you say, just using the cloud accounts and instantly getting access to these super impressive models?
- [35:27]
B
Yeah, I think for most people like they don't need to run Opus 4.8 or Fable 5 or anything even remotely close to that. I think most people are just quite happy if like chat GPT can help them reword their email so that it's you know, know slightly better and more succinct and like help them search for stuff better. I think 99% of people aren't the, in the camp where they're really making full use of the, of the like the best models. It is true though that yeah if you want to run something that's close to like an Opus 4.8 on local hardware, it's going to cost, you know, how much is a Mac studio? How much is the Mac studio with one terabyte of RAM going to cost? Probably like 10 or $15,000. So for the people that are tinkerers or they're using it for work or from a values perspective they want to be self sovereign enough to be able to run that type of intelligence no matter what without it being gatekept then for them maybe it's worth it for the majority of people, I mean, excuse me, I haven't played around with them with Google's Gemma 4 but I heard that that model can be run on quite reasonable hardware and I bet for most people that are just like using it to rewrite work emails and stuff like that and maybe do some research, maybe a model like that is good enough and I'm sure that we'll get to a place where like the models will keep improving that you will be able to. For most people they will be able to run a good enough model locally on you know, a reasonable PC that they might use at home for other stuff like it's funny, YouTube's been, YouTube, YouTube on my TV for some reason has been serving me all these like old videos of like tech like basically from the BBC in the UK which is like the local, the, the national broadcaster like videos about the computer revolution in the 80s and used or the, and the computers in the 60s. And you see, it's like, you know, computer was the size of a room and it's like it can write three lines of text. And obviously now, you know, we have computers that can sit on our lap that can do infinitely more work than a massive computer could back then. So I'm sure that within a decade we'll be running Opus 4.8 type intelligence on our phones or something like that.
- [37:51]
A
What are your thoughts on image AI, video AI, some of the other tools out there. Do you find those as useful?
- [37:59]
B
I personally haven't used video AI much. I've used image a bit. Sometimes if I'm doing some copywriting work I might want to try and like I might be mocking up a graphic where that my copy is going into. And so it's useful for like prototyping and trying to, to, yeah, trying to prototype something in your head. But as a producer of like something final that you will ship from a graphical, like a graphics perspective from. At least for my level of understanding of how to use it, it's not quite there yet. For websites I think it's good enough. Although I think, you know, if you're a web designer, just like if you're a good writer, you can identify, you know, AI. Well, I think most of us can identify AI written slop anyway. But I think, you know, designers will say, oh, this website was obviously generated by AI, but to the average person they can't tell. It just will look like a, you know, decent enough designed website. It is pretty crazy. Like I think a lot of people are already starting to fall for, for AI generated video online. Like I have. My mom tells me about older family friends of ours who are constantly sending her what are clearly AI generated videos, sharing them with her as if they're real. So yeah, I think that that's going to be pretty crazy.
- [39:28]
C
Yeah, sure. As someone who used to be an artist, specifically in video game, I absolutely love the art. Doing art with AI, I find it very interesting. I use it like you described to create concepts. Maybe I want to paint something by hand, but then I need to quickly get a lot of reference and image and idea to see if something I like or sometime what I do is I will do the entire draft by hand because I want to be very precise and then I will use an AI to modify the elements in the image to speed up time. Personally I used midjourney a lot, but I didn't like how it was integrated weirdly in discord and then their own stuff it was just I didn't like very much. So now use actually the ChatGPT for image. I don't remember the name now, but it's the one you get Dall E. Maybe they change. Yeah, I don't remember what's the name, but I found the newer version are quite good. And Nano Banana Pro is also I found especially with different style. Like if you want an infographic, it is very, very powerful.
- [40:40]
A
How do you maintain voice consistency across AI generated content? Do you have a style guide? Is the AI learning from samples? How do you make sure to have this kind of consistency?
- [40:52]
B
Oh man. So for my own writing, I've tried in lots of different ways to get it to write like me. The only, the only one that has ever worked is I created a skill in my open claw that I called post yap, which when I get ideas for a tweet, for example, I will just voice note to the AI and then it has, it helps me using it. So I fed it all of my, I think I downloaded a file with 63,000 of my tweets at the time. And so it has that and it has the raw tweet draft, let's say from a voice note that I give it and then it's able to generate a succinctly written, concise piece of content in my voice. Generally it's like it works for tweets, like very short. Other than that, the only way that I've managed to like let's say quote unquote, write with AI is if I write first and then I just use it as an editor, right? So I might say like, suggest ways I can improve this, this blog or I might go through and I might identify like I want a better way or a better way to say this sentence. And it's always like just trying to get the AI to interject in either the smallest pieces of writing possible. So like sentence by sentence rather than like rewrite this whole blog or rewrite this blog for me from scratch, right? It's like giving it the raw materials and then going through and try and using it to iterate on the smallest amount of text at a time possible. You can be like, I don't like this word in this sentence. Can you suggest like 10 other ways that I could say or 10 other words I could use here, right? Or I might feed it a piece of writing and say, are there any like logic gaps in my writing? Like something, you know, maybe where I. I jumped from one thought to another without joining them together very well, or is There any, you know, anything that you can point out that I can improve about the flow of this, of this piece of writing and there, you know, I take the feedback and then I apply it myself. And that's how I'm able to quote, unquote, keep my voice. But if I give it a style guide and you know, examples of my writing for the other than this like post jap magical skill on openclaw for short writing tweets, it if I've tried to like train it to write like me and it, it never does it in a way that I think sounds like I would. The thing, the, the thing I always think about with writing is like when I, I have an idea for a piece of, for like a piece of content, I might not always know like how long it's gonna be in the writing process itself. I might decide the, this is stupid or actually there's more to it than I thought. This is more than a. Sometimes I'll start writing it like a three sentence tweet and it will turn into 400 words. Right? And so the very act of writing changes what you write and what the end result is. And so there is no way if, unless you know exactly what a piece of writing would look like at the end before you start, you can't really steer an AI to create that for you because it would be like asking it to predict the future. It's like, hey, I want to write about this. You know, write it like I would. But you don't even know how you would write it, so how can you expect it to write it the way you would?
- [44:45]
A
No, I think those are good thoughts. You're certainly describing a way that we've also been thinking about and using AI for, for text related things. So as we wind down here, Daniel, let's get in one more question, which is we talked at the beginning about knowing some of the particularities about prompts and some of these things. Less important now, how should people think about high level AI use versus the tactics of AI use?
- [45:16]
B
So I've actually been doing some AI consulting recently and you know, with people who have corporate jobs and their companies don't let them use AI, but they want to be able to use them use it themselves. Right. And they're just not used to working with AI as a tool. Like, you know, we're all used to using Google or using a web browser or using, using email. Right. But a lot of people just haven't built that habit of like incorporating AI into everything they do. So and they feel Overwhelmed. They're like, well, I need to learn skills and I need to learn this, I need to learn that. And what I always say to them is like, forget about trying to, worrying about the tactical stuff, the how to. It's like, as long as you understand that if you can speak English, if you can tell another human being what you'd like them to do, then you can tell an AI what to do and you can use AI, Right? Using AI is essentially just the skill of being able to describe an outcome that you would like to happen. And so the biggest thing that people can do is like, don't get too bogged down in the details and focus more on every single thing that you're doing day to day in your job. Think, how can I use this? How can I do this with AI, right? And so, you know, that might mean you open Claude or ChatGPT and you say like, I need to do this for my job. You know, my boss asked me to write this report on this thing, but I have no idea how to get started. What, you know, give me some tips and it will like tell you. Or you might, you know, it might be compiling a bunch of research or you know, putting together a bunch of, organizing a bunch of documents that you already have access to. It's like, well then you use like Codex and get the, get an AI agent to organize that for you. So it's really just like how can you incorporate and do more of your work or have AI do more of your work rather than worrying on the like, oh, I don't, I can't use it because I don't know how to do a skill and I need to be able to set up a Claude MD file and I need to be able to write, you know, how to do prompt engineering. I need to study how to write a good prompt first. It's like, no, just open one of the tools, speak to it like you would to any other human being and go from there and incorporate it into everything that you do. I heard that that tip came to me from Dharmesh, who is the founder, he's the co founder of HubSpot, which is like a CRM, a publicly listed CRM company. And he was saying like the biggest thing you can do is just try to use AI every time you need to do anything on a computer. And so that's really what got me using it for almost everything that I do.
- [47:56]
A
So for the listeners, if you're new to AI, you want to get started, maybe you're worried about some of the privacy sovereignty concerns. We do have a course called the AI Resistance course where we will show you how to use AI for productivity while pointing out a lot of these concerns along the way. And then we also have of course this has been Daniel Sempere Pico and Daniel, give you the final thoughts here where people can follow your work and how they can interact with you.
- [48:21]
B
Yeah, so like I said, they can check me out dansemperepico on Twitter and if I ever have anything else going on, I share it on there first. And yeah, thank you very much for listening to us.
- [48:35]
A
Hey, thanks for listening. I could really use your help. Real quick if you could share this episode with someone, engage with me, leave a review anywhere you this helps me to break the technocratic shadow banning that is happening with my brand. And of course if you really want to escape the technocracy, go to escapethetechnocracy.com privacy tutorial, series, books, newsletters, consulting and of course you can leave a donation. Thank you very much.
- [49:07]
B
Sam. It.