 It's like a nice way of getting out of the way. Hello, did I finish the link call for April 12th, 2023? Thanks, Lancian. Keep going. Yeah, so I'm on call for like Google Meet and apps, a workspace in general for second level and also in the evenings for social co-op. Nice. So it's like, yeah, yeah. But then it's nice, I get it out of the way and then I will travel, I will go to the US and so on. If our futures are portfolio careers, how are we all going to manage all of our relationships and responsibilities? How does this work? Well, intriguing. Anybody else? How are you doing? Yeah. I know the answer to that question. Oh, perfect. What is it? We don't? We'll just let go. When I hear the young people these days don't do email, I'm like, well, good, fine. But then I'm like, OK, OK. So they're texting or what's happening or insta-ing. They're using insta not for pictures, but for messages, which happens a whole bunch. So how do you find the thing you promise to do anybody? And how do you remember what channel it was on? Like really, seriously. Or are they just being birthed now that humans have evolved such terrific brains? Are they being born with perfect recall? So this is a non-issue for them and it's merely an old fart issue for me. I mean, I think that it has to do with trying to parse down what's required of you in part through difficulty. Yes, for the other person. Yes. So baking it so that somebody does have to reach out to you again and ask you a second time. Because, yeah, email, it's the spam problem, but it's not just spam. It's work spam. There's so many things that people are asking to do in a portfolio situation in life. And people all have it. It's the top priority for them at that moment in their entire life for everyone when they ask it. But is it their top priority tomorrow? Well, if they remind you, then it moves up the priority. Right. So you can call that a distributed cognition. As in, I'm a procrastinator, I'm in recovery, now I'm trying to leave it behind. And when people ping me on something, which I showed them, I'm like, well, thank you. I mean, we're thinking the way they're now. I know that this is really important. So I guess you get a chance to actually do it. It's been working so far, but it is not. Yeah, as somebody who has followed this technique for a while, I don't know how well it's working. But I don't know if I have any other choice either. It's the problem. Yeah. Yeah, I think Unified Inbox is an interesting side question. Like, will it help? I guess we said spam, but the attention scarcity problem is like, you can only pay attention to so many things. So you could imagine having a Unified Inbox, which used to be the panacea for me, but it's like, well, then suddenly you have like 1,000 things. So I have an idea. Why don't we create a Unified Inbox that has an auction as its gate inward so that everybody who wants your attention has to bid for it, and then you just gift your friends like free passes through the gate. Ross Mitchell and Jeff Lynch had that as a startup. And Esther tried to propose that, too. Isn't that what Open Collective does? Is it? Like, does an Open Collective allow you to set bids on bounties on GitHub issues? I think that was a, maybe it's not a feature there anymore. Maybe it was a feature of something else that worked with Open Collective. Like, I swore, and I saw it there, where it's like, is this important to you? Put a bounty on the GitHub issue, and then somebody will resolve it. Maybe it was for the GitHub managers. Like, so that way you could get more people to be open sourced and involved in open source. I don't know, I think like the unified inbox concept is, I think, fundamentally bad. Like, all of my organizational work is trying to separate my inboxes when people, other people want to combine them. Even my actual email inbox, right? Like, I create an email alias for different things that I engage in that are different contexts, right? So this group is actually under my IndieWeb alias. So if I want to think about IndieWeb at that moment, I'll go to all of the things that are labeled IndieWeb in my inbox, which is automated through coming in through that email. But if I'm talking to someone about Crest Forward work, for example, an open source project I'm involved with, that has its own email address. And I go to that when I'm ready to deal with that. And of course the problem is sometimes there are things that I forget that I need to deal with in one of these inboxes. But that's what my notes are supposed to help me out with in terms of prioritization. But that's, I mean, you could just have one email address and just use different labels, you know, based on various parameters. But I mean, you're relying on everybody to use email. I've got five clients, I've got three clients and they all use teams at the moment, but different teams. I've just lost track of how many different teams I'm now a member of, each with them with dozens of channels. And that's just one platform, teams. Then I've got, you know, then the people trying to contact me by WhatsApp and SMS and Signal and Messenger. And I just, this is why I've really invested a little bit of time in a tool. You said notes, like obsidian, because I just pour everything into that and try to keep track of it. But it's still, where is that message? Because I need to reply to it eventually. I can't reply to it in obsidian. And it's a nice, kind of absolute nightmare, but there's no way that we're going to be able to unify all these things because none of the vendors want to shed questions. So this is where, like, I always go back to the idea of like, we will have to, you know, recede into a user agent in the end. Like, do we need the browser to actually, like give us the tools for, like, actually interconnectivity and, you know, cross-reference? At least that seems like it will cover everything with each web base, which is not soon usually, but, you know, like most, you know, so, yeah. I don't think that Matrix is a very good platform to build on. It seems to be like as a hub for bridges. So it's like, you know, some platforms are better than others for, you know, in the sense that they support, like, reach your instructions that let you impersonate as well. For a recent group, I tried to install and use Element, which builds on Matrix, and was able to install it and configure it, which was kind of problematic because it was clunkier than most. But then I wasn't able to connect with the other people who were on Element. I couldn't find them in the directory. I don't know what went wrong, and we gave up. We gave up. It was horrible. Yeah, there's some, at least when you push, when you press, like, start a conversation, it doesn't look for people somehow. You have to click again to separate it out. It's, it's, but it's, it's been improved. It was worth. And honestly, it's still the best I know. It doesn't make even with all these robots. We're, so, we're so far down the road on technology and communication and all that kind of stuff. How, how is this not a better solved thing? And how is its complexity seemingly getting worse over time instead of competition? Everyone wants to be the monopoly on your attention. Yes. So we should just all never have progressed beyond Minitel and Prodigy Mail. Or, or just end capitalism. Then, resolve that problem. Very simple. Upgrade. I mean, using upgrading capitalism just because it seems like less confrontational. Well, that's good. As opposed to, as opposed to defund the police, let's upgrade the police. Yes. Well, I think people will, yeah, let's give the police lasers. Well, we couldn't afford the lasers. So we just dropped flashlights to their helmets. I will, I will note that like, I think the big thing that makes it a difference for just making it more distinguishable and usable, which is why I've ended up more heavily in Slack than a lot of the other things is when it's individually addressable. Cause then it gets back to what you're talking about with like the reply stuff, right? If you can link an individually addressable link to the message that this is associated with, then suddenly everything becomes much easier to manage. It's amazing. Yeah. I built this a very simple one for Matrix, which has this. So, okay. So I'm a fan, but. So if you use a weakling or a tag in any Matrix room with this boat, it will both dump the, immediately dump the whole message to the Aura and also link back to the message. That sounds useful. That does. It is very useful. And that's actually one thing, one group reached out and they said like, I would like to have an Aura for that feature. So apparently people are, and it seems so simple to build. So I'm no child platform that I know of, like even in support of this. Thank you. What, is that open sourced? Could be. Yeah. Yeah. It's the simplest thing. Yeah. It's part of our route. I can show you an example. And the other thing, of course, is that all of these different platforms will probably get AI agents in there for you. So, you know, when somebody wants to talk to me, they talk to my agent first and have a go and very, and then the agents have then sent me an email saying this guy wants to talk to you about something. Follow this link and go back to the place. And then I'll be able to handle that because that will bring everything back to email into one place, which I can manage. Maybe not an email. Maybe send something to my personal wiki. I don't know, honestly. You say that, but what's super interesting to me right is you're basically just describing a secretary, right? Yeah. Yeah, I want a secretary. Yeah. But what's interesting to me is like. Don't want to pay. Yeah, what I've seen is that people have found secretaries less and less useful. Like all of the executives at our company have basically like given up on using the secretaries for that type of function. And instead they're more like there for scheduling stuff. I think that's just it's very difficult for another human to figure out your personal prioritization. I despair of an AI ever figuring it out personally. Yeah. I'm not saying it's going to work, but I'm saying I'm sure they're going to try. And I wish they would succeed. I would pay, you know, I would pay an AI for that. Really better than paying secretaries. If it got closer and reduced the choices I had to two or three and I could give this one and then went and did all the footwork to like confirm everything, that'd be pretty awesome just in itself. I have a scheduling, right? I have a scheduling, is it, is it otter? Otter.ai, I think? No, that's a transcriber. Yeah, otter the transcriber. I might have an idea, so it's something else. And Matthew, since you're only with us for a half hour, what would you like us to talk about? We don't have to talk about this. This is just our default qubits topic at the start of the call. Oh, well, I guess top of mind right now is Massive Wiki is coming along quite nicely, as Peter would, you know, would tell you if I wasn't here anyway, I guess. And we would really like to work with one or two people on just the dimensions that we're using. When I say Massive Wiki, I meant the pilot project for, you know, thinking tools, tools for thought map. TFT. TFT map, thank you. It's the end of the day here, I'm a bit exhausted. Had a really crap day actually. Sorry, sorry. Just technical problems, just, I won't bore you with it. And yeah, we've got these 11 dimensions for how you measure a tool for thought. And this is something which I originally boiled down by reading other people's stuff. Bentley did a big document. Two or three other people here posted things which I boiled it all down. We had a big discussion amongst the three of us. And it would be really good to know whether other people agree with these dimensions or not. And so that was the idea of April. In April, we were gonna get people to comment on it and haven't really managed to do that. So that's something I'd like to throw in the ring for discussion, if that's okay. Apart from that, any ideas on how to hook up a chat GPT or something similar to a personal store of notes and using that profitably? You know, how to interrogate an LLM whilst it's reading a particular corpus of notes is also something I'm looking at right now. So those are a couple of things to talk about. And if either is interesting, I'm cool with either. I'm enjoying the phrase, how to interrogate an LLM and picturing one of them strapped into a chair with electrodes on its head. Yes, you will now not lie or hallucinate and don't be something I didn't already know. And then I'm seeing the famous scene from The Wire and other places where they put a colander on the suspect's head and wire it to the photocopier. And then they have a sheet of paper that says he's lying in the photocopier. Then they press the button and it comes up. He's lying, like, ah! Is this really a thing? I have never watched The Wire. I've heard it's very good. But you must watch The Wire. Some of the best movie slash TV ever made, I think. It is really, really nice, yes. And holds up well. You can also write, like, first watch the one minute Family Guy, I think, clip, which is about how people always are saying that you must watch The Wire. Oh, I've never seen that. Which I can relate to because I do that, yes. Anyway, that's me. What else? Oh, I am sorry. About the map, sorry. Like, should we, you said it on April, so should we do it now? Like, maybe, like, do you have a link? Sorry, I should have it, but I don't. It's fine, mate, it's fine. It's in the chat, yeah, yeah. Go on, go on, go on. The page on the TFT map called How We Measure Thinking Tools. And it sets out a small table with 11 dimensions, a definition in a couple of lines and some guidance on what I think a tool for thought has to do to score highly on that dimension. And, you know, the architecture would allow us, you know, we could have an entire page about each dimension if we need to. There's also some space to put in other proposed dimensions, place to park dimensions we don't use, but it's what it is right now. And so to contribute, I shall take these dimensions and contribute a certain elevation of tools. As it says at the bottom of the page. Yeah, sorry. That's fine. You know, if you have any comments on these dimensions, there's a couple of ways of doing it. There's a commenting function at the very bottom of the page, you can just post a comment. But at all, you actually, like, you know, join in as a massive wiki contributor and then you can do, you know, compose your own notes, you can write blog posts about it, you can just add files to the massive wiki and link them in there so you can see them. I see that Aram has actually, this is your profile, right? And he did a great job going through all the tools he uses. Is Aram, is Aram, you haven't yet, absolutely, do you have an N of greater than one or is Aram the only person who's done this? Well, if you go to the people page, you've got six people. Let's have a look at Francine's. Yeah, I think I left the, I think I might do, yeah, I thought so too. You've done the first part. Okay, okay, yeah, but this is after my question. I was just thinking about it. Hank, Hank, Hank did something interesting. Hank Kuhn, he decided to not just do thinking tools, but also thinking practices. We distinguish between the two. And this is exactly why, you know, why you do pilot projects and you open them up, you know, people come in with ideas, why don't you do that? Oh yeah, it's not a bad idea, we should think about that. So we had a good conversation with him last week, right, Pete? About, so he thinks we should actually have, yeah, these practices as well. I don't know whether Pete's code is gonna be able to cope with that, but, you know, well, that's his problem, you know. So yeah, total number of people, six, but not all of them have done everything. So thank you very much, Aaron, for your, what do you think of the dimensions? I mean, you've used them, you've rated obsidian, press forward, pin board, Instapaper, arena, deep down 750 words by these dimensions. Did they make sense to you? Yeah, yeah, I guess the one thing I think was maybe a little weird was like openness as a rating. I don't know if I got it right. I think I interpreted it as like openness to new users in the sense of like how easy it is to get in. Cause I felt like if openness meant open sourced, that's like a true false, right? I don't know if that's necessarily a rating. I mean, obviously there are open source projects that are more welcoming than others. I guess this rating works for either of those definitions. Well, in the file, you know, how we define thinking tools, the definition of openness is how well does it play with other tools? How interoperable and customizable is it? And now that I look at that, I don't understand why I wrote and customizable, how interoperable it is, is open. Customizable is a completely different thing. So I need to take that word, those customizable term out. It's funny how you see things which, when you talk about it with somebody, that you just don't see when you're writing it. And I wrote this. Oh yeah, that happened with it. Recording together and everything. So right, so interoperable, there may be an opportunity and this is like where I don't know if it's, we'll have to think about the others, but like interoperability is very important and openness in the sense of like free software is very important. And it seems like there may be two dimensions. Ideally, of course you can like probably for most, but like just by fact, because I'm thinking OCDN is very interoperable because it's based on Mardana and all these nice things. Simple standards, but it's actually, for me it will score like less than five just because it's not open source. That seems like it will muddy the waters to put it somewhere. Yeah, this is it. This is whatever happens when you have a conversation with people who know or feel reasonably well tools for thought, you take one dimension and say, I think we should split this into two. And before long, you've gone from 11 to 59 dimensions. Right, right. And it's a lot of work for people to score 59 times on each tool. And it's also the audience for this are people who are new to thinking tools. And what they would like is, which is the best one? Which is the one that scores the highest out of 10? Right, you want one number because you have like, so maybe one suggestion to not increase the number of dimensions will be maybe data sovereignty and interoperability seem closer to error to me potentially than open source, net free software and like interoperability. So maybe that would be easier. Just listen to this here from Jerry first because you put your hand up. Sorry. Thank you. So a question that's starting to come up a little bit more for me in talking generally about the kind of stuff we're talking about here is how do we help muggles find their way to the thinking tool that's their best fit? So and the map is an attempt to offer a tool for that process, I think. And also for IT architects or knowledge management experts or whoever else to go like shopping or comparing or even to promote the union of different the interoperability of tools that aren't yet interoperable because they would complement each other so well who knows how that plays out. But I think that there's a really interesting thing we might be able to make practical maybe even just as a simple decision tree or as a, hey, go test these four things and then answer these three questions and that will bring you back into these are very likely the tools you would like something like that. But I think there's, I don't know I haven't given any sort of formal time to this question but of somehow it may not be simplifiable to what I'm hoping for but a really simple muggle friendly front end that we could just aim people toward and say, hey, if you find any of these tools for thinking conversation cool and would like to try it yourself here's a way to find your way to a tool or a set of tools that you'd like and there's a complimentary thought to this which is even closer to the TFT map which is once you have like more than 20 things in a map like this and there's a lot of data collected it turns into a hairball really quickly. So if there's a room for the couple of curated points of view like, oh, Pete has a suite it's a little bit like build a second brain in Tiago Forte or linking your thinking and Nick Milo they have chosen a particular set of tools which they train you up in and then they have a series of intellectual frameworks for how they do what they do and so forth that's interesting because that is a curated view on the hairball and if you can get if you can curate a half dozen curators to have interesting and different ways of thinking about and ways of using the tools that's easier to come into because then you're basically finding a recommendation that works for you and then you install three tools and you don't have to worry if they're mixed in match because the person recommending is like hey, these three things work really well together so there's that too. Okay, I'll respond to the second thing more than the first although to be honest, I have the decision to it but we should talk about that too because that's a lovely idea but what you just described is actually inherent in the architecture if you go into somebody's personal profile so our rooms or mine or whatever in the template with people and the people provide the information they don't just provide here the scores of the tools they also, there's a section in there called my current tools and practices, right? So mine says I use inbox curation to identify high priority sources of content auto label, high priority newsletters I'm maintain the Twitter list I use GTD, I use progressive summarization via Zedal Cast and Overviews I'm skipping over it but in five or six bullet points I mentioned all the tools I use and most of the important practices in a, it's like a story almost so there's a little bit above that about me who I am and then here is how I use a particular set of tools and practices together this is how I combine them and then below that you see the scores I give the tools that I know about and to me that's probably I included that in the architecture because I thought that a newcomer is more likely to say, well I'll find somebody in this list of people who seems to have a similar approach than that I would have but is further down the path and I'll be inspired by their collection of tools, right? And then when I click on a link to go to a particular tool from that personal profile I find other people who use the tool I'll go and check out how they've integrated it into a different set of tools and practices but there's one tool in common so it's not just a collection of data about ratings of tools it's also a collection of stories about how people use them and I think those stories will be really good raw material for your decision tree. Love that. And it could be that one person's answer is Oh, I took the Build a Second Brain course and I do that and that's just a reference over to BASB. Absolutely. But then some people have developed their own flows and were really interested, you know? I wanna know how Marshall Kirkpatrick manages his info I wanna know how Maggie Appleton curates her gardens for digital garden. Yeah, in an ideal world we'll get this to the point where Maggie and Michael and everybody else will put in their stuff, you know? You can also blog you can write blog posts in here and then repost them and do permanent versions and do all sorts of other nice things and if it becomes a useful in the universe we'll get that sort of content in there. And then I imagine when you got that huge, that collection of content building a decision tree on top of that could be quite a lot of fun. I'm not quite sure how to do it but that's Pete's problem. Excellent, perfect. And it could be, tell me if this is too difficult to prompt but if we had a series of dimensions which is part of this conversation could we, and if we pointed to somebody like Marshall Kirkpatrick or a bunch of others who have written a lot about these topics online and if we were using an LLM that's up to date not stuck in 2021, could we ask it to populate the dimensions for that person and infer from their writings what that is and therefore not have to ask them to fill it out and then we could email the resulting thing and say, hey, our smart brain thinks that you would answer it this way could you correct it? Wow. Oh, the fan since he's put his hand up. Go for it. On that topic. Yeah, I mean, I think the there's, I keep going back to the, what I believe could be the killer app although that seems violent for like, you know, for some of these which is like, you know, the actual sharing for example, like, you know, as soon as you have M people wanted to contribute their notes or pool to a commons, massive wiki and I would, et cetera. Then you know the sources, you know where the repository where their data is. So just with that, you have like, I would start to figure out how they're actually editing that because different tools will have different fingerprints to put it some way. In the case of the Aguara, when you add a repository, you can say which tool you use mainly. It's just a free form field. I've forgotten about Staxure. Thanks for pointing to it. Pete, you're very quiet. What do you think about the idea of having an LLM pre-populate or draft populate, you know, personal profiles or other content and then letting it loose? I don't know. It sounds like a bad idea. Although I have to say part of the reason I was so quiet is I took about two-thirds of Jerry's list of thinking frameworks and said, hey, GPT, what are some more? Wow. So here you go. Oh! Oh! Oh, wow! Nice! Oh! Bam! Ignis me. Hey, nice work. And half and a third of those I've got. Another one. Yeah, third of those I've got. So, very cool. Wow. The real question is how many of these are entirely fantasy? Yeah. Yeah, exactly. Yeah, but here in a weak environment is like the ones we don't exist, maybe we'll create very easily. Yeah, actually there's a weird thing where sometimes GPT will hallucinate something that ought to exist. Yeah, yeah, yeah. No, exactly. It's basically projecting into the space things that sometimes are eminently reasonable that we just haven't done yet. Like I'm willing to bet the double diamond framework is completely invented and we need to go figure out what it is. I'm just kidding. I'll just ask chat GPT what it is and then figure out if it's useful. Right, right. You know? Exactly. Pete, I haven't finished your novel, your chat GPT novel, Hot Air. But I mean, how long did that take you to do? Wow. You know, the funny thing is it took probably about 30 or 40 minutes cumulative time and then probably two hours to actually publish it. Have you actually asked chat GPT to write the code for the TFT map that you need to do for the TFT map platform? I haven't because it's a little involved. It's a little bit quicker to just write it myself. But sometimes it's worth coding with. I've only tried it once and it was okay. So thanks for mentioning it Matthew. This was a story I kind of accidentally started with chat GPT and one of the things I didn't write in the write-up is it's like, okay, so it's getting to the point where instead of choosing a book to read, a pre-written book to read, you just like help chat GP tell you the story of the book that you want to read. So it reminds me of Flancy and a couple meetings ago said, how long until we have a gen tube where it's just like, I wanna watch something and it just starts making a movie for you. I also experimented a bit with a fictional force. I don't know if you have this thing. I saw, I would go to hacker news often. I guess we all have all this hacker news, you know this, yeah, but like we all have our feeds, I guess. And I saw this article that had this very evocative subject or a title. And I was like, well, maybe what I want is to ask Judgement D to write, you know, this article just because it seems so simulating as a prompt and read that first before I actually read the actual thing. So, you know, you can imagine an extension that gives you like the same as opening new tab, you know, opening Judgement D or in like a generative context. So you could be forking into like fictional reality, you know, like more easily. It wouldn't be interesting to have a chat, TPT bot that exists in your browser like hypothesis where it'll tell you that it's got some comments or, you know, it knows about this or it could summarize this for you or whatever. That'd be a fun example. Oh, they're like clippy. We could totally resupplicate clippy. We could have clippy with like a bandage over his forehead and the end of his clip in a sling and he could show up with like, you know, a missing tooth and talk to us. I mean, we could do that because already now that you notice our baby clips or the runaway, the runaway paperclip function. Right. Yeah, I mean, this is it. If it's, I think it comes back to the thing that Aaron mentioned, you know, how many of these 58 frameworks actually exist or not? I, you have to be on your guard so much if you have a little assistant popping up and saying, I know something about this, you know, and then it says something completely wrong, completely untrue, but absolutely believable because, you know, it's mining the entirety of everything that's ever been written. So it looks really right. And it's customized to your interest. It's watching what you're doing and you're reading all your emails and looking at everything you're looking at. It would be almost impossible to, well, I mean, if you don't want to be misled, you'd have to turn it off. I mean, it would be, you spend more time checking whether it was true or not. I'm not sure that the cost benefit is there if you're using it for that sort of thing. That's why I wanted to bring up Pete's hot air thing because that seems to be something that chatGPT is more made to do, isn't it? To write for you, not necessarily things which are true, but he's written basically, you've written basically the treatment for a TV series, I think the way it does what it is. You know, each thing is an episode in these people and you find out what happens in each episode without actually watching the whole thing, yeah. Brief aside, I went to a meetup here in Portland yesterday morning that was all about chatGPT and so forth and there were three panelists and it was actually quite good. And just for fun, there was Q&A and I asked one of the questions and I introduced myself as Trent Crim, the Independent. And that got a really nice laugh. But it was very interesting how many people were there who were pretty deep in. The advice was there's kind of a land grab right now. You start anything you want, the cost is low, but don't expect it to be the same in 18 months. This stuff is moving really, really quickly and so your current ideas will probably be outdated pretty soon, but kind of go for it. It was really interesting. And I met a couple of local entrepreneurs. I just haven't been in the community, in the local tech community enough, so that was nice. And it was nice to be with humans. Humans are good. Yeah, what's that like? Yeah, I was rusty. So I asked chatGPT to add Wikipedia links to that list. And it's funny, it got about two thirds of the way and it quit, it met out of memory or something. But anyway, since I said add Wikipedia links, all of them had to be Wikipedia links. So there's the innovation ambition matrix. I'm like, that sounds like a BS one. So I click it and yeah, it's BS. And then I did a search for it, innovation ambition matrix is a real thing. It just doesn't have a Wikipedia page. Yeah. So it provided a Wikipedia link, but the link doesn't, is 404? Yeah. But it should be, it's a real thing. So somebody should go write the Wikipedia article. Okay. So here, I guess, I don't know if you were experimented with like using Wikipedia for writing plans, because plans are at the intersection between fact and fiction, right? Oh, that's good. I tried, I asked it, well, for a few plans, and it seems to do quite well. Also it has this like, you know, pretty structure, you know, like, you know, 10 steps and so on. Yeah. So I guess I'm, and this brings me back as usual to like, I guess pattern languages and so on, right? Because like, in the sense of like cool exploration of like problem space yield solutions, just by, you know, simplifying and breaking down into steps and so on. Which are things which seem like, you know, within its, you know, skill set. Cool. I don't think I'm making a decision. You know, I'll bet the chat GPT would be really good at converting text-ish stuff, like descriptions into pattern languages. I don't think it'd be perfect, but that would be a good way to start, you know, flesh out a pattern language of something that is kind of deep and rich enough. It would be the raw material that it would process. Like a framework, you know, dead people. Yeah, I like that. Documentation as well, for procedures. Like, yeah, like a, like pattern language for innovation frameworks or something like that. It's funny, like, I don't know about you all, but trying to get used to the chat GPT world has me like thinking more crazy ideas than before. It's like, it feels like there's more possible where before I just, I saw a lot of stumbling blocks and I'm like, I don't know how to overcome that. So that path stops. Now it's like, you could at least give it a try. And there's a, you know, even odds that something good would come of it. And if not, in the doing of that, you might actually find your way to some other idea or around the sand dune. And that's really cool. There's like a required flexibility of thinking that we need to figure out. By the way, I think there's a really, certainly like Evo Haining already has a Promptcraft book out. Apparently Promptcraft engineers are like commending like $400,000 salaries or something crazy ass like that. So A, if you all want to change tracks right now and go do that, I recommend it. It may not last a year, but hey, making nearly half a million in a year is not a bad thing. But also there may be a really interesting kind of bootcamp offered to do somehow here. And I was actually seriously thinking, because I'm trying to do the cyborg thing. I was like, and to my mind, Promptcraft is an aspect of being a good cyborg. And we're going to need to learn how to integrate ourselves with software in ways that are as comfortable and flexible and generative as we are describing right now. And so should we create a channel on the matter most called Bootcamp or ChatGPT Bootcamp or something? We already have a ChatGPT and I channel. Should we use that to try to do some of this? But if people wanted to learn up, like where would they go? Medium, where they'll find blog posts written by ChatGPT based on other blog posts. No, I found one because I curate stuff into my hub and I was reading this blog, I was reading this blog post which I cued earlier. I think I've read this before. I'm writing notes on it, I've read it before. So I do a search in my hub for the same time, no. So I did a search for, this is how I would have tagged this new article. And I found another article which I had like a week or two ago. And I put the two side by side and one was basically a plagiarization of the other. And it wasn't a human plagiarization, it was almost certainly a chat. So ChatGPT used to plagiarize articles about ChatB2GPT to write new articles about ChatGPT basically. So we are already being absorbed into the board. Go ahead. It's even worse than that considering how many medium articles that are written by humans are still basically made up bullshit. Yeah, exactly. It really is. I mean, my medium feed is just, it's just prompt engineering guides. But most of them are just rehashes of each other's, they just rehash each other's work, you know? I mean, now that it's got a paywall in there for the last couple of years, it's one way of making money. At the panel yesterday morning, there was a guy sitting in the front row right in front of me who was actually working ChatGPT as a fourth panelist. And they started with it asking the first question, which was not a terrible question. But they had ChatGPT also generate funny introductions for the three panelists with a particular sense of humor that worked really well, everybody. They got a good chuckle. And was he getting paid $400,000 a year for? Not as far as I know. He would not have been in that room. Unless prompt engineering is a part time kind of job because it's so efficient. So the question is whether it's here to stay, I guess. I think you were up, someone right up here in one of the latest calls, or it's just like sort of like a transition period when, so while we have this window where like, you know, with prompt engineering, you can reach some like results, but in GPD 4 or 5 or GPD 4, it's already here. GPD 5, that will not be necessary anymore because it will be built in. So I guess like, you know, is it a level of frameworks in programming where like, you know, you can already the framework independently, but you usually, you always have a framework that's value or is it like, well, I guess I'm struggling for, we're looking for like an alternative to that. So I guess I sort of believe it's more like a framework. It's a chat ABT to me looks a lot like a general purpose computer, which is fascinating for people, you know, the folks on this call and utterly boring for anybody who actually wants to get any work done. It's like, okay, what do I do with it? It's like, well, prompt engineering, it's so much fun. It's like, no, it's not fun, dude. One of the other comments to get done, one of the panelists answered to one of the questions was interesting. He was like, the difference between 3.5 and 4 is very significant. Trying to feel a commercial service that serves muggles or clients with 3.5, really hard, very dicey, like we're backing off, 4.0 actually sort of gets there and might round that bend out. It was really interesting. I think that's true. It's got a lot more, it's more safe. It's safer. I still think it's too general to use. So I like Plancy now like your idea of frameworks. You won't be writing bare Python code or bare PHP code or bare Node code. It's a lot more like, go use a framework for biological research or legal document stuff or something like that. I think another way to think of it, it's like, ChatGPT is kind of like AWS. It's just bulk services. And to really use it, somebody who installs a server that verticalizes the backend of that. So you'll have health insurance vertical or life insurance vertical or gun control legislation vertical. And you'll buy that. And whoever's done the vertical solution will have a bunch of stuff that adds prompts to actually structures the things that you wanna do. Do you wanna start a legal document? Do you wanna edit a legal document? Do you wanna compare two legal documents? Do you wanna mash them up together? And then, so it'll give you a framework for things that you can do with the AI. And then it'll do value add, okay, they asked me to merge these two documents. Here's the prompts that would do that. And the user won't see that. It's built into the infrastructure of the tool. That's where it's going. What I call a task in the draft blog post, I sent you very recently, but you haven't had a chance to look at it yet. But yeah, the idea is that you've got an interface there and there's some, there's a sort of task you can choose like a dropdown. And when you choose a particular task, it pulls up a number of things that you can do. And depending on the task and depending on what you put in it, it then goes and does some things. Packages all that together, sends it off to chat GPC. So it's like each option in that dropdown is a prompt that really works really well. And then it might ask you for a bit more information about how would you like to me to write that? Would you like me to write it like a high school teacher or a university professor or whatever? And you write a few things down. And then it packages everything up and goes to its own store of content as well as the wider internet. Yeah, this is exactly, exactly what I wanna do when we meet my developer on Monday. So, yeah, I just have to spend a lot of time reading terrible blog posts about prompt engineering. Because at one point I'm actually gonna have to get myself immersed in that and it's gonna suck, it's gonna be awful. To go a little bit further with that packaging thing, there's at least a couple. There's auto GPT I think and baby AGI where they'll take a task and the supervisor will ask chat GPT how would you plan to execute this task? Break this up into things that chat GPT would do. So some people are starting to build kind of autonomous planning engines that use chat GPT to set up a set of tasks for chat GPT to burn through. So using chat GPT to use chat GPT? Yeah. Oh, God. Yeah. And then you can run the whole thing, right? You've got a little supervisor that just does the feedback loop and keeps it going. Which is why you're like, you can imagine a DVD file saying, okay, when a user sends me a prompt, mercies to legal documents or then does that straight and does that with like the prompt, which generated the prompt and so on. I think there's similar things happening in image generation, you know, when like you can use like civil diffusion as on straight, I don't know exactly how the maze work there, or you can like, I saw some experience of people expanding prompts with GPTJ. So the prompt will be richer. So it has things like, you know, photo realistic or you know, it adds detail to the prompt which has the consequence of improving the end result in the future. So yeah, maybe API, but I have to say, like probably the most exciting, I really like how you put it, Jerry, where like you said something like, you have been thinking of crazier or out more other ideas because of GBT, this like, that felt like, you know, open up a problem space, which is clearly, you know, what's happening, you know, because of all these ideas, you know, started doing it otherwise, but you know, imagine like, you know, millions, billions of people maybe, just like thinking new thoughts because you can dare to go there. Because you know, you have these, even the fact that, just a bit may as we push its limits and you know, and the limits of future models, it may still remain like unskillful or flaky and so on. But even the fact that, you know, when you write something, you are like, well, maybe we'll nail it or maybe it will be crap. That actually may be more addictive and fun to some extent as an experience than having to be nailed it all the time. Because when he nails it, you are like, oh, see, I spotted something that worked. I did the right prompt. I mean, you have some participation there. And you know, this experiment that say that, you know, when something gives a reward only part of the time, it is more fun and addictive than when it gives a return all the time. Intermittent reward schedules or something like that? Right, right. So, you know, when you are like pushing the limits in this sense, I guess, you know, all the pieces are there to like make it, you know, what we're seeing, I guess, which is a phenomenon. I don't know. I think you all are a little overly optimistic. Well, that's our job. One hand, sure, people could be using it to think, do thoughts. But in general, when you go out in the wild and see how people are actually using chat and GPT, they're mostly using it to not think. They're delegating their thinking to chat GPT. What's some rather negative results? Are they always negative though? I mean, even if you used to think, I mean, you generate and you're like, do I believe this? No, do I believe this? And eventually you are like, yeah, this sounds about right. So it could be more like a coprocessor, I guess. Maybe I'm back to optimistic. I was, when Jerry was saying earlier that, you know, how it's sort of very exciting and liberating, I was thinking back to the late 90s when then there was this sense that anything could happen. But then anyone in their garage could do something and it could become big. But now it's not like that. Now it's anything could happen, but it's gonna be done by Google or Microsoft because you have to have the sheer mass of power. So building money and compute power, no? Building and training in LLM takes a lot of compute power, but once it's canned and packaged, if you can then use it, any little schmo can go use the thing. So yes, they're limited to the models on offer, but just like HTML and web hosting, like anybody can go do stuff. Right, the easiest way to think about these models is like, think about them like using WordPress, right? The model is the platform for a particular thing you wanna do that it's particularly well-suited for, and then you can feed your inputs and your outputs into it to make it better and improve on it from the baseline. But like I do think starting with the basic LLMs, yeah, and it's not for everyone. Some LLMs don't work for some people, some don't. But like... Just electricity, thinking about electricity, right? Yeah, that might be a little too much, but like it doesn't actually take that much time and money to train an LLM to do something if you come in with like an existing model that's pretty basic that won't deform what you're trying to do, which is out there. It's just then you have to use it for something, right? The training that needs to go into something like ChatGPT, that's exhaustive, that's not with the normal sort of user capabilities. But there's lots of things you can, there are lots of models out there that you can use to train and do stuff, but then like, okay, you've trained a model with all of your text. What thing is this useful for now? And right now, there are not that many answers. I would argue that, we've already had this discussion, so I won't go into depth, but I would argue that ChatGPT is not even itself a useful thing really, right? The people are doing cool things with it in general, but useful things less so. But putting that aside, like, yeah, I have a project where I'm training, building out a pipeline to train an LLM to automatically tag articles for me. Did something very similar for work, tried to do it myself with a different set of sources, but like, that's useful, but like, for me personally, doing what I'm doing, I'm not sure how broadly useful it is. Okay, I've got a weird question. So I, I don't know. So imagine the hot air experiment, right? I ended up with, I don't know, 40 responses from ChatGPT and had to write kind of a preface to the whole thing. And it's this big hairball kind of thing. And it's kind of interesting, I did make a page where you can read the whole thing all at once and it's unreadable because it's just like, there's too much text. It's like an ebook has pages for a reason, it turns out. So I really liked the format I came up with, the massive wiki to publish, you know, little snippets. So imagine something else that size, an experiment. I've got another experiment, which was less well-publicized, even less well-publicized. I've put some links to it in the biocouplex dispatch. It's two versions of kind of a tech summarization of a really long email thread in the OGM mailing list, 87.