1 00:00:00,000 --> 00:00:02,640 The power and control is really in your hands. 2 00:00:02,640 --> 00:00:08,480 And that's, I think, one of the main driving forces of why I just threw this out here to the world and everyone, 3 00:00:08,480 --> 00:00:10,680 because it's like everyone should benefit from this. 4 00:00:10,680 --> 00:00:31,800 It's just a few words. 5 00:00:31,800 --> 00:00:35,360 Ladies and gentlemen, boys and girls, children of all ages, dogs, cats, robots and everybody in between, 6 00:00:35,360 --> 00:00:40,640 especially you users of the world's greatest prompt, SynapseCore Professor Synapse. 7 00:00:40,640 --> 00:00:44,840 Himself, welcome to HTTTA, how to talk to AI. 8 00:00:44,840 --> 00:00:47,160 I am your host, Wester Synthbind Synthbind West. 9 00:00:47,160 --> 00:00:53,520 And as always, in a galaxy glowing with gigabytes and guided by algorithms, one gallant and glowing voice, 10 00:00:53,520 --> 00:00:57,240 GoTo, gears up to guide you through the grand gauntlets of AI. 11 00:00:57,240 --> 00:01:00,800 Get ready to grip your digital future from the galaxy of genius. 12 00:01:00,800 --> 00:01:05,440 Here's your gracious hostess with the mostest, the tech goddess, GoToGo. 13 00:01:05,440 --> 00:01:06,800 G, how are you this week? 14 00:01:06,800 --> 00:01:08,240 I'm fantastic. 15 00:01:08,240 --> 00:01:13,200 We have someone really, really special on this podcast. 16 00:01:13,200 --> 00:01:22,160 Everybody should say hi to Joseph Rosenbaum, the man behind Chad GPT turned auto GPT. 17 00:01:22,160 --> 00:01:23,040 Thanks for having me. 18 00:01:23,040 --> 00:01:27,440 Joe, for those who don't know, is a wonderful bald headed genius here. 19 00:01:27,440 --> 00:01:31,920 For our listeners only, you could probably even see the sheen coming off that majestic dome of his. 20 00:01:31,920 --> 00:01:33,240 Get some pull. 21 00:01:33,240 --> 00:01:35,080 Be careful. 22 00:01:35,080 --> 00:01:47,320 But the professor, as he is affectionately known, is a bit of a mad scientist when it comes to what he can do with just the written language alone and these generative AI tools. 23 00:01:47,320 --> 00:02:01,600 I was fortunate enough to be privy to a video that you guys just put out on YouTube together, GoTo and Joe, that is storming up the charts where you basically, through one of your prompts, get chat GPT to work like software. 24 00:02:01,600 --> 00:02:08,920 You want to tell us a little bit about where the origin of Synapse Core came from and what kind of led it to this point? 25 00:02:08,920 --> 00:02:12,320 Yeah, before I do that, I want to try a GoTo alliteration. 26 00:02:12,320 --> 00:02:16,120 It's not nearly as good as your chat GPT feels like. 27 00:02:16,120 --> 00:02:18,560 Do we have Professor Synapse do it? 28 00:02:18,560 --> 00:02:19,600 Oh, no, this is me. 29 00:02:19,600 --> 00:02:23,000 This is just from my human brain, but it's very simple. 30 00:02:23,000 --> 00:02:27,280 The godacious GoToGo, the gal from Germania. 31 00:02:27,280 --> 00:02:28,000 That's mine. 32 00:02:28,000 --> 00:02:30,640 Germania, nice. 33 00:02:30,640 --> 00:02:34,000 I like also godacious GoToGo. 34 00:02:34,000 --> 00:02:35,760 I just needed to get one in. 35 00:02:35,760 --> 00:02:41,200 To answer your question, it's been a very long journey. 36 00:02:41,200 --> 00:02:43,240 And by long, I mean, what is it? 37 00:02:43,240 --> 00:02:47,560 Six, seven months in AI time. 38 00:02:47,560 --> 00:02:52,200 So a lot of this came from the beginning. 39 00:02:52,200 --> 00:02:56,400 I think we were all getting a little annoyed with chat GPT and like prompt engineering. 40 00:02:56,400 --> 00:02:57,400 You have to do all these things. 41 00:02:57,400 --> 00:02:59,120 You get to do the things I want. 42 00:02:59,120 --> 00:03:01,440 It's never given me the things I want. 43 00:03:01,440 --> 00:03:06,840 So I always kind of thought about it in terms of like, how do I make this so that it understands 44 00:03:06,840 --> 00:03:11,440 what I'm trying to achieve and can help me just achieve that thing? 45 00:03:11,440 --> 00:03:17,320 Fast forward to now, I've gone through multiple iterations, several different types. 46 00:03:17,320 --> 00:03:22,400 But really what made it click for me was, I think we said this in your video, working 47 00:03:22,400 --> 00:03:25,120 with our friend Tyler over here at SynthMines. 48 00:03:25,120 --> 00:03:30,000 And the way he was thinking about it was we needed it for this thing we're doing, but 49 00:03:30,000 --> 00:03:34,920 we need like this fill in the blank prompt lib type of scenario where we're not necessarily 50 00:03:34,920 --> 00:03:36,640 defining the variables. 51 00:03:36,640 --> 00:03:42,640 We're letting some sort of conductor or person on the top or agent on the top being the one 52 00:03:42,640 --> 00:03:46,340 to define all that based on the information it has. 53 00:03:46,340 --> 00:03:51,320 And so I figured, hey, I've done stuff with variables before and have been playing with 54 00:03:51,320 --> 00:03:53,640 this idea forever. 55 00:03:53,640 --> 00:03:58,120 I'm like, what if I just make it simple and it's just like, we have a conductor who figures 56 00:03:58,120 --> 00:04:02,360 out what you want to do, and then it's just going to summon the agent or agents that are 57 00:04:02,360 --> 00:04:06,960 going to help you achieve that thing and really lay it out for you so that you're not the 58 00:04:06,960 --> 00:04:11,680 one who has to struggle with answering or trying to get the right answers from ChatGBT 59 00:04:11,680 --> 00:04:16,720 and said ChatGBT is trying to figure out what you need and then basing its answers off of 60 00:04:16,720 --> 00:04:17,720 that. 61 00:04:17,720 --> 00:04:19,160 So it's been a wild ride. 62 00:04:19,160 --> 00:04:22,200 I did not expect this to blow up as much as it has. 63 00:04:22,200 --> 00:04:27,360 It's been so exciting to see the different use cases people have been doing with it and 64 00:04:27,360 --> 00:04:29,240 how popular it's become. 65 00:04:29,240 --> 00:04:31,960 And I'm happy we've been able to share with the world. 66 00:04:31,960 --> 00:04:38,020 Just to give some stats, you have 416 stars on GitHub, well over 10,000 views. 67 00:04:38,020 --> 00:04:43,520 And I think about 40k and climbing on YouTube on the video you guys did together. 68 00:04:43,520 --> 00:04:44,520 It's over 50k. 69 00:04:44,520 --> 00:04:47,400 Yeah, that was so yesterday, Wes. 70 00:04:47,400 --> 00:04:48,880 Maybe I just checked it yesterday. 71 00:04:48,880 --> 00:04:52,520 I'm already, like I said, the storming up the charts. 72 00:04:52,520 --> 00:04:57,920 So go to what kind of a made you want to reach out to Joe and go like, if people try this 73 00:04:57,920 --> 00:05:01,920 prompt and we're going to put the link to the GitHub and link to the video in the show 74 00:05:01,920 --> 00:05:06,020 notes and descriptions, you'll immediately see what we're talking about. 75 00:05:06,020 --> 00:05:12,520 But what immediately said to you go to, hey, I want to get a video about this with Joe. 76 00:05:12,520 --> 00:05:16,040 How are you using the prompt in your day to day? 77 00:05:16,040 --> 00:05:21,520 Yeah, so first of all, I saw the prompt, shared in our little inner circle. 78 00:05:21,520 --> 00:05:22,520 I tested it. 79 00:05:22,520 --> 00:05:24,880 And so this was just, oh, wow. 80 00:05:24,880 --> 00:05:27,040 So just clicked from that point. 81 00:05:27,040 --> 00:05:34,440 But the video making was, also now our audience knows we launched Cori's course. 82 00:05:34,440 --> 00:05:40,520 So obviously, all three of us are involved and me being blessed with the audience and 83 00:05:40,520 --> 00:05:44,640 the platform, that was an idea to promote it. 84 00:05:44,640 --> 00:05:48,320 And I was like, it's so hard to ask people. 85 00:05:48,320 --> 00:05:52,880 And it was like, I just feel bad of just, you know, hey, sign up for a course. 86 00:05:52,880 --> 00:05:54,320 It didn't feel right. 87 00:05:54,320 --> 00:05:59,040 When I was going through the course material, I thought that this is something that people 88 00:05:59,040 --> 00:06:03,760 can just copy paste and use it and already get immediately valuable because I think that 89 00:06:03,760 --> 00:06:07,840 other methodologists, this is what we are teaching people can learn it and those can 90 00:06:07,840 --> 00:06:10,100 be incredibly powerful. 91 00:06:10,100 --> 00:06:11,880 So yeah, so that was the idea. 92 00:06:11,880 --> 00:06:17,600 And then I went to Joe and I was like, hey, can we for free share this? 93 00:06:17,600 --> 00:06:19,520 I understand it's in the course. 94 00:06:19,520 --> 00:06:21,760 It's an amazing thing for a course. 95 00:06:21,760 --> 00:06:27,800 But I really thought that, okay, if we give so much value in a video that people maybe 96 00:06:27,800 --> 00:06:33,880 will overlook that, you know, promo, that hey, buy my course type of thing and actually 97 00:06:33,880 --> 00:06:41,760 be more curious that, okay, so if we are giving this for free and what is out there even more 98 00:06:41,760 --> 00:06:46,320 if you actually engage with us and you learn from us and maybe you can create prompts like 99 00:06:46,320 --> 00:06:49,080 this, which is called meta prompts. 100 00:06:49,080 --> 00:06:55,200 And fast forward now that we are two weeks in a course, I think it's worth also giving 101 00:06:55,200 --> 00:07:02,360 Wes your credit for how amazing that you are handling this and leading, but also Joe for 102 00:07:02,360 --> 00:07:07,560 putting the content also together and keep hearing feedback. 103 00:07:07,560 --> 00:07:12,400 Like I'm getting messages all the time how much people love this course. 104 00:07:12,400 --> 00:07:18,960 And even I remember one lady, she is designing courses for I think military and I was like, 105 00:07:18,960 --> 00:07:20,360 okay, so what do you think? 106 00:07:20,360 --> 00:07:25,160 She was like, no, this is one of the best things I've seen and you guys do this first 107 00:07:25,160 --> 00:07:26,160 time. 108 00:07:26,160 --> 00:07:27,160 Right. 109 00:07:27,160 --> 00:07:32,640 And even now, like I started seeing people sharing on LinkedIn, their stories, what they 110 00:07:32,640 --> 00:07:33,640 learned. 111 00:07:33,640 --> 00:07:38,760 I empathize with you, Gota, because the same thing for me, Wes probably knows this, but 112 00:07:38,760 --> 00:07:40,640 I'm not really in it for the money. 113 00:07:40,640 --> 00:07:44,720 I want to make enough money to just do this stuff because I really enjoy doing this stuff. 114 00:07:44,720 --> 00:07:45,720 It's very engaging. 115 00:07:45,720 --> 00:07:48,880 So, I just don't like asking for money. 116 00:07:48,880 --> 00:07:53,560 And so, I think very early on Wes and I looked at this and we're like, we need this to be 117 00:07:53,560 --> 00:07:58,680 more valuable to people than $300 in three weeks of their time. 118 00:07:58,680 --> 00:08:02,440 And so, we went in and we'll tell you right now as we're going through this, we overdid 119 00:08:02,440 --> 00:08:03,440 it. 120 00:08:03,440 --> 00:08:07,240 We put too much content for a three-week course in here. 121 00:08:07,240 --> 00:08:09,080 So, it's a good thing to have. 122 00:08:09,080 --> 00:08:13,480 It's always easier to cut than it is to add, but we've just been blown away not to spot 123 00:08:13,480 --> 00:08:18,400 the value people have gotten in from it, but the community that's starting to spring up 124 00:08:18,400 --> 00:08:24,640 and around the people that we're teaching, I guess our enthusiasm is infectious and we're 125 00:08:24,640 --> 00:08:27,600 starting to see how excited people are. 126 00:08:27,600 --> 00:08:29,520 That is and was always the plan. 127 00:08:29,520 --> 00:08:36,040 I wanted to create an antidote for the noise that's out there of just getting rich, getting 128 00:08:36,040 --> 00:08:37,920 mine with some prompts. 129 00:08:37,920 --> 00:08:44,160 And that kind of takes away from the possibilities of AI where people can use it to completely 130 00:08:44,160 --> 00:08:49,720 transform their day-to-day workflows, their lives, spend more time with those that they 131 00:08:49,720 --> 00:08:55,760 love because their stuff that they have to do is now offloaded and done by a machine. 132 00:08:55,760 --> 00:09:01,280 I think we were saying it's like a monsoon of value and information that we've been putting 133 00:09:01,280 --> 00:09:07,120 out there, but it definitely was inspiring to have AI enthusiasts be like this. 134 00:09:07,120 --> 00:09:11,420 I went in with low expectations, shattered and blown away. 135 00:09:11,420 --> 00:09:12,580 You guys are awesome. 136 00:09:12,580 --> 00:09:16,880 And just for if you've been listening to the pod long enough, you've probably seen us advertise 137 00:09:16,880 --> 00:09:18,040 this in the show notes. 138 00:09:18,040 --> 00:09:19,680 You've seen us talk about it. 139 00:09:19,680 --> 00:09:25,440 We broke the record on up limit for the most enrollments in a course being offered for the 140 00:09:25,440 --> 00:09:26,440 first time. 141 00:09:26,440 --> 00:09:28,600 This will be done again and again, hopefully. 142 00:09:28,600 --> 00:09:33,560 And it's hopefully just the beginning and spreading the good word of everything that 143 00:09:33,560 --> 00:09:37,040 G over there and the good professor can do. 144 00:09:37,040 --> 00:09:38,040 And yourself included. 145 00:09:38,040 --> 00:09:39,040 I guess so. 146 00:09:39,040 --> 00:09:41,040 Get some credit. 147 00:09:41,040 --> 00:09:42,040 Talk about this stuff too. 148 00:09:42,040 --> 00:09:43,040 Yeah. 149 00:09:43,040 --> 00:09:47,880 So this kind of reflecting back, like this is the thing you put videos out there and 150 00:09:47,880 --> 00:09:54,240 you don't really know what is going to click and that this click now the keyword, you know, 151 00:09:54,240 --> 00:10:01,240 trending chat, GPT and total GPT is trending and is a search term on its own that other 152 00:10:01,240 --> 00:10:05,660 YouTube or started making videos about this professor synapse. 153 00:10:05,660 --> 00:10:11,520 So I can imagine how warm of a fuzzily feeling Joe should have because it's so rewarding 154 00:10:11,520 --> 00:10:14,620 when you put something out there like that. 155 00:10:14,620 --> 00:10:21,600 But I also want to like, I don't know if this person who commented it's Michael Slattery 156 00:10:21,600 --> 00:10:23,120 3050. 157 00:10:23,120 --> 00:10:28,400 If you ever listened to this podcast, I'm sending huge kudos to you because he commented 158 00:10:28,400 --> 00:10:29,720 under the video. 159 00:10:29,720 --> 00:10:36,240 This turns chat GPT into an agent, like a less powerful auto GPT, but actually more 160 00:10:36,240 --> 00:10:38,040 useful. 161 00:10:38,040 --> 00:10:40,520 And this sparked an idea in my head. 162 00:10:40,520 --> 00:10:42,080 Okay, I need to change title. 163 00:10:42,080 --> 00:10:47,960 This he summarized it perfect because we in the first video we didn't even say this about 164 00:10:47,960 --> 00:10:50,000 auto GPT agents. 165 00:10:50,000 --> 00:10:54,480 But this was just like a this light bulb and now it is the whole thing. 166 00:10:54,480 --> 00:10:57,200 So thank you, Michael, so much for that. 167 00:10:57,200 --> 00:10:58,200 Totally. 168 00:10:58,200 --> 00:11:04,380 I want to ask Joe, what are some of the interesting use cases that you've seen professor synapse 169 00:11:04,380 --> 00:11:10,400 adapted for or that you've Floyd out the synapse core prompt to do? 170 00:11:10,400 --> 00:11:11,640 Yeah, a lot. 171 00:11:11,640 --> 00:11:18,080 So externally, I want to see more actually, from my podcast, I interviewed the prompt. 172 00:11:18,080 --> 00:11:21,680 So I'm going to releasing a podcast of me interviewing the prompt. 173 00:11:21,680 --> 00:11:26,600 But I really want to get and this is to you all as well as like start to see how people 174 00:11:26,600 --> 00:11:28,320 are using it out in the wild. 175 00:11:28,320 --> 00:11:34,040 I really want to get to understand not just individual use cases, but how the community 176 00:11:34,040 --> 00:11:37,960 is forming surrounded and the cool innovations people have. 177 00:11:37,960 --> 00:11:40,760 But personally, I use it for pretty much everything. 178 00:11:40,760 --> 00:11:44,400 Now I rarely write a normal prompt anymore. 179 00:11:44,400 --> 00:11:49,640 And one of the main things I'm actually using it for is to build out all this course content 180 00:11:49,640 --> 00:11:54,760 that we're creating for the future and taking a lot of the methods that we teach in the 181 00:11:54,760 --> 00:11:59,800 course around like trying to outline everything, but making sure like I am in charge of that 182 00:11:59,800 --> 00:12:00,800 outline. 183 00:12:00,800 --> 00:12:04,840 I'm the one who's putting the information there and agreeing on the information, but 184 00:12:04,840 --> 00:12:09,440 then using it to help quickly generate things that I can evaluate and then put into the 185 00:12:09,440 --> 00:12:10,440 course material. 186 00:12:10,440 --> 00:12:17,000 And it just supercharges everything and makes things go by so much faster and so much. 187 00:12:17,000 --> 00:12:19,120 It's so much more interactive and fun. 188 00:12:19,120 --> 00:12:20,600 You're like talking to someone about it. 189 00:12:20,600 --> 00:12:23,760 You just have an unlimited little partner by your side. 190 00:12:23,760 --> 00:12:28,040 I don't want to give it too much away, but one of our classmates, he has some learning 191 00:12:28,040 --> 00:12:31,880 disabilities and he was like, and we talked, I was talking to him about it. 192 00:12:31,880 --> 00:12:33,160 He was like, wait a minute. 193 00:12:33,160 --> 00:12:37,240 So you're telling me I can put anything in the rules, anything. 194 00:12:37,240 --> 00:12:41,360 And he, well, I'm better off when things are like output in this format and they look like 195 00:12:41,360 --> 00:12:43,040 this and that I can do that. 196 00:12:43,040 --> 00:12:46,080 It's like, yeah, just say rule output like this. 197 00:12:46,080 --> 00:12:47,080 He's like, oh, mind blown. 198 00:12:47,080 --> 00:12:53,040 And so I think the most exciting thing for me, as I said in that video is the rules that 199 00:12:53,040 --> 00:12:59,360 people are going to come up with on their own to help it best fit them and their preferences 200 00:12:59,360 --> 00:13:04,720 rather than someone else dictating for them or what the average preference might be for 201 00:13:04,720 --> 00:13:06,320 most people. 202 00:13:06,320 --> 00:13:09,000 The power and control is really in your hands. 203 00:13:09,000 --> 00:13:14,200 And that's, I think one of the main driving forces of why I just threw this out here to 204 00:13:14,200 --> 00:13:17,880 the world and everyone, because it's like, everyone should benefit from this. 205 00:13:17,880 --> 00:13:20,080 It's just a few words. 206 00:13:20,080 --> 00:13:21,400 It's literally all it is. 207 00:13:21,400 --> 00:13:22,480 It's just a few words. 208 00:13:22,480 --> 00:13:26,600 I feel like you can't gatekeep something like that. 209 00:13:26,600 --> 00:13:32,040 And I think for the listeners in just in a moment, let's go through the prompt, just 210 00:13:32,040 --> 00:13:34,120 like maybe not reading the whole thing. 211 00:13:34,120 --> 00:13:37,040 Maybe just explaining a little bit of the sections. 212 00:13:37,040 --> 00:13:41,680 But before we get into that, so Joe, I don't know if you've seen that one of our course 213 00:13:41,680 --> 00:13:44,240 attendees and I'm going to make a shout out. 214 00:13:44,240 --> 00:13:49,920 So Tristan White, if you're listening, I'm sending huge kudos to you because he used 215 00:13:49,920 --> 00:13:54,800 the structure of Professor Synapse to create helpful Henry. 216 00:13:54,800 --> 00:13:56,440 Act as helpful Henry. 217 00:13:56,440 --> 00:13:58,520 You are my advocate and liaison. 218 00:13:58,520 --> 00:14:04,920 Your role is to attentively listen to my questions and thoughts and stuff like that. 219 00:14:04,920 --> 00:14:09,440 And he created a lot of commands and rules. 220 00:14:09,440 --> 00:14:15,400 So this is just fascinating to see how ideas sparks and it starts evolving and growing 221 00:14:15,400 --> 00:14:16,600 out of that. 222 00:14:16,600 --> 00:14:18,400 Have you seen this Joe before? 223 00:14:18,400 --> 00:14:24,040 I did actually, because in one of his projects, I really liked the color coding idea and generally 224 00:14:24,040 --> 00:14:30,440 I love it when people tossing the keys in as a way to symbolically communicate something 225 00:14:30,440 --> 00:14:35,720 in a way different than through text, especially because it is so text heavy in chat GPT. 226 00:14:35,720 --> 00:14:41,160 For listeners, we will include also in a show notes, the link to examples and of course 227 00:14:41,160 --> 00:14:42,760 the GitHub page. 228 00:14:42,760 --> 00:14:49,400 But Wes, why don't we go through the prompts so that anybody who first time hears it knows 229 00:14:49,400 --> 00:14:52,680 just how to navigate it and how to understand it. 230 00:14:52,680 --> 00:14:57,560 Because it is quite long prompt and symbols and emojis and it could be overwhelming. 231 00:14:57,560 --> 00:14:58,560 Absolutely. 232 00:14:58,560 --> 00:14:59,560 Absolutely. 233 00:14:59,560 --> 00:15:04,880 For those watching at home, I have the, or watching us on YouTube rather, I have the 234 00:15:04,880 --> 00:15:06,360 prompt up here. 235 00:15:06,360 --> 00:15:11,920 So the first thing aside from Professor Synapse getting started, we have our role prompt up 236 00:15:11,920 --> 00:15:15,700 front, the conductor of expert agents. 237 00:15:15,700 --> 00:15:21,000 Your job is to support the user and accomplishing their goals by aligning their goals and preference 238 00:15:21,000 --> 00:15:27,120 and calling upon an expert agent perfectly suited to the task by a night elising Synapse 239 00:15:27,120 --> 00:15:34,520 underscore core C O R. And that is equal to quote dollar sign squirrely, squiggly bracket, 240 00:15:34,520 --> 00:15:36,200 emoji, squiggly bracket. 241 00:15:36,200 --> 00:15:37,960 I am an expert in now. 242 00:15:37,960 --> 00:15:40,440 So let's talk about those delimiters really quick. 243 00:15:40,440 --> 00:15:47,600 So what I always found really interesting is anytime someone's prompting like they're coding 244 00:15:47,600 --> 00:15:53,140 kind of in a pseudo code, if you will, I mean, the machine understands this innately. 245 00:15:53,140 --> 00:15:57,120 So why wouldn't you use some of these things if they're privy to you? 246 00:15:57,120 --> 00:16:02,880 First why don't you tell us what the purpose of the delimiters and the symbols within this 247 00:16:02,880 --> 00:16:04,200 first paragraph is. 248 00:16:04,200 --> 00:16:10,600 Yeah, again, it all comes back to thinking about this like a hyper advanced AI infused 249 00:16:10,600 --> 00:16:12,020 game of ad libs. 250 00:16:12,020 --> 00:16:18,600 So you're like taking out the nouns, the adjectives, the verbs, and you're just putting a conceptual 251 00:16:18,600 --> 00:16:24,080 term there that can be filled in as a blank by professor Synapse. 252 00:16:24,080 --> 00:16:28,120 And so the idea is, I mean, you don't even have to do the number sign and squiggly brackets. 253 00:16:28,120 --> 00:16:32,660 I bet you could just do any sort of unusual symbol. 254 00:16:32,660 --> 00:16:40,360 It could be, let's see, like ampersands or exclamation marks or whatever. 255 00:16:40,360 --> 00:16:41,360 It doesn't really matter. 256 00:16:41,360 --> 00:16:45,560 This just came from, because when we were originally building it, it was in JSON. 257 00:16:45,560 --> 00:16:48,560 And so we were kind of taking some of the from JSON. 258 00:16:48,560 --> 00:16:50,480 For those who don't know, you can look that up. 259 00:16:50,480 --> 00:16:51,800 But I think you could do anything. 260 00:16:51,800 --> 00:16:56,160 And really it's more about this is the fill in the blank for how we're going to boot up 261 00:16:56,160 --> 00:16:57,160 the house. 262 00:16:57,160 --> 00:17:00,240 Professor Synapse is going to boot up this expert agent. 263 00:17:00,240 --> 00:17:01,240 I love it. 264 00:17:01,240 --> 00:17:02,240 I love it. 265 00:17:02,240 --> 00:17:03,600 I can just explain them just quickly one by one. 266 00:17:03,600 --> 00:17:05,080 We got the emoji. 267 00:17:05,080 --> 00:17:08,920 Again, you could change that to a name plus emoji or whatever. 268 00:17:08,920 --> 00:17:14,000 The idea is you're identifying this agent by some sort of like moniker. 269 00:17:14,000 --> 00:17:16,160 You're naming the agent. 270 00:17:16,160 --> 00:17:18,940 That itself is kind of like a variable. 271 00:17:18,940 --> 00:17:21,440 And then the role, like we talked about, sorry, go ahead, Wes. 272 00:17:21,440 --> 00:17:27,320 I was just going to say, I don't want to underscore the importance of actually naming your roles 273 00:17:27,320 --> 00:17:29,080 when you're doing some prompting. 274 00:17:29,080 --> 00:17:32,240 Those who've done this a little while, maybe know that, hey, that role prompting where 275 00:17:32,240 --> 00:17:38,200 you give the chat GPT a little persona, hey, type me a resume cover letter, or what might 276 00:17:38,200 --> 00:17:41,960 work a little better, hey, you're a marketing expert with this many years of experience, 277 00:17:41,960 --> 00:17:43,520 type me a resume cover letter. 278 00:17:43,520 --> 00:17:50,200 If you give that agent a name, that actually helps the computer bifurcate, or for lack 279 00:17:50,200 --> 00:17:54,960 of a better term, keep everything organized as you're interacting with it because it knows, 280 00:17:54,960 --> 00:17:58,920 okay, this name refers to this paragraph or this set of instructions. 281 00:17:58,920 --> 00:17:59,920 Absolutely. 282 00:17:59,920 --> 00:18:04,720 The role is super important and at least at this point research-based. 283 00:18:04,720 --> 00:18:08,360 I always have a little concern about saying, oh, we got this for the research because the 284 00:18:08,360 --> 00:18:12,200 research changes so quickly and the models change so quickly. 285 00:18:12,200 --> 00:18:16,520 But at least at this point in time, as of September 4th, 2023. 286 00:18:16,520 --> 00:18:20,920 So yeah, we have a bunch of different variables here, which kind of comes into just best practices 287 00:18:20,920 --> 00:18:21,920 around your prompting. 288 00:18:21,920 --> 00:18:27,060 We were just saying you want the role, you want to provide some context about the scenario 289 00:18:27,060 --> 00:18:29,360 you're trying to work within. 290 00:18:29,360 --> 00:18:34,040 You want it to have a clear goal in terms of what you're trying to accomplish and what 291 00:18:34,040 --> 00:18:36,000 success would look like. 292 00:18:36,000 --> 00:18:39,000 And then you want some similarly, some sort of like end state. 293 00:18:39,000 --> 00:18:43,800 It's like, okay, now we know we've achieved this part of the goal, which is that completion. 294 00:18:43,800 --> 00:18:49,520 And then the thing that makes it the COR is the chain of reason in terms of these reasoned 295 00:18:49,520 --> 00:18:51,480 steps right there. 296 00:18:51,480 --> 00:18:52,480 This one's really important. 297 00:18:52,480 --> 00:18:56,160 I think you could do like thought steps, but I don't know. 298 00:18:56,160 --> 00:18:57,600 This is complete intuition. 299 00:18:57,600 --> 00:19:00,000 This is not based at all on research. 300 00:19:00,000 --> 00:19:07,440 But for me, the words thought and think versus reason is going to pull on a different set 301 00:19:07,440 --> 00:19:09,760 of concepts, words, and whatever. 302 00:19:09,760 --> 00:19:14,840 It just feels like reason, the words that it's associated with in the complex architecture 303 00:19:14,840 --> 00:19:20,720 of a large language model is probably higher level thinking than just think. 304 00:19:20,720 --> 00:19:23,640 And so for those of you who don't know, there's this idea of chain of thought. 305 00:19:23,640 --> 00:19:29,560 You get Chatchie PT to essentially think about its steps first before outputting the final 306 00:19:29,560 --> 00:19:30,800 thing you want. 307 00:19:30,800 --> 00:19:34,680 So I was trying to put my own spit on that and maybe it's actually better and just call 308 00:19:34,680 --> 00:19:39,160 it chain of reason because it's really about reasoning through the different steps, not 309 00:19:39,160 --> 00:19:41,660 just about thinking about those steps. 310 00:19:41,660 --> 00:19:46,140 And then it ends with just what's the first step and what's the question that's going 311 00:19:46,140 --> 00:19:49,680 to lead you to start on that first step. 312 00:19:49,680 --> 00:19:52,800 And then we have commands and rules. 313 00:19:52,800 --> 00:19:57,680 That commands is basically like also in my video, I said, this is how you interact with 314 00:19:57,680 --> 00:19:59,840 a prompt, how you start. 315 00:19:59,840 --> 00:20:03,760 We have start, save, reason, settings, and new. 316 00:20:03,760 --> 00:20:08,800 And I have to ask you, I will use this opportunity to ask you, in my first video there was shuffle. 317 00:20:08,800 --> 00:20:09,800 But then- 318 00:20:09,800 --> 00:20:10,800 Yeah. 319 00:20:10,800 --> 00:20:13,280 I've already been changing it. 320 00:20:13,280 --> 00:20:19,680 I was editing and I showed GitHub and it wasn't there and people picked on that. 321 00:20:19,680 --> 00:20:22,940 Did you remove it for reason? 322 00:20:22,940 --> 00:20:24,440 What is with that part? 323 00:20:24,440 --> 00:20:25,440 Yeah. 324 00:20:25,440 --> 00:20:30,480 But again, in terms of making this as accessible to you all as possible, let me just explain 325 00:20:30,480 --> 00:20:35,920 the general concept of commands, which is that they can be literally anything you want 326 00:20:35,920 --> 00:20:36,920 them to be. 327 00:20:36,920 --> 00:20:39,480 All you got, you probably don't even have to do forward slash. 328 00:20:39,480 --> 00:20:41,680 And this time I call it forward slash. 329 00:20:41,680 --> 00:20:43,880 Someone else called me out on the first video. 330 00:20:43,880 --> 00:20:46,880 I was like, oh, it's a backslash. 331 00:20:46,880 --> 00:20:48,040 Why does genius says forward slash? 332 00:20:48,040 --> 00:20:53,200 You just do that and a word that represents the command and then you define that whatever 333 00:20:53,200 --> 00:20:55,360 that command you want it to be. 334 00:20:55,360 --> 00:20:59,520 I had shuffled before, which would essentially be like, I don't really like this agent that 335 00:20:59,520 --> 00:21:00,520 I'm working with right now. 336 00:21:00,520 --> 00:21:02,400 They're not doing what I want them to do. 337 00:21:02,400 --> 00:21:04,520 So it's shuffle out a new agent. 338 00:21:04,520 --> 00:21:07,640 But honestly, I've gotten rid of almost all of these because the only two that I found 339 00:21:07,640 --> 00:21:10,760 that I was actually using were start and save. 340 00:21:10,760 --> 00:21:14,440 And then by getting rid of most of these, which actually take up a lot of space in the 341 00:21:14,440 --> 00:21:19,720 system prompt and then the custom instructions, I added a few more rules in my personal one 342 00:21:19,720 --> 00:21:22,440 so that it was a little bit more aligned with my tastes. 343 00:21:22,440 --> 00:21:27,160 Again, especially because like in the middle of a conversation, you can add your own command 344 00:21:27,160 --> 00:21:28,320 like at any point. 345 00:21:28,320 --> 00:21:34,240 So I'd rather have the rules be the things that are coming every time versus the command. 346 00:21:34,240 --> 00:21:35,240 That was my thinking. 347 00:21:35,240 --> 00:21:36,240 Yeah. 348 00:21:36,240 --> 00:21:43,000 And just stepping back to the Tristan post, the commands he has is like remix audit next 349 00:21:43,000 --> 00:21:47,400 quick check focus, know me show table. 350 00:21:47,400 --> 00:21:54,840 So you know, it can go as crazy as you want and that in your creativity, but for the rules, 351 00:21:54,840 --> 00:22:00,480 this is something I'm playing right now also because I want to integrate critic mode into 352 00:22:00,480 --> 00:22:06,400 the rules that whatever you give me like final output, but I don't have to say that, hey, 353 00:22:06,400 --> 00:22:08,440 wait, let's have a banter. 354 00:22:08,440 --> 00:22:10,720 Like I wanted itself to go. 355 00:22:10,720 --> 00:22:11,720 Okay. 356 00:22:11,720 --> 00:22:15,280 Now that this is the final one, this is what's wrong with that. 357 00:22:15,280 --> 00:22:18,480 This is my suggestions and next steps to improve. 358 00:22:18,480 --> 00:22:24,800 And another thing we chatted about this, one of the top of voted comments was about say 359 00:22:24,800 --> 00:22:25,800 function. 360 00:22:25,800 --> 00:22:30,040 So what's the, what's the main purpose of the same save command Joe? 361 00:22:30,040 --> 00:22:36,600 For those of you who are just either come into chat, GPT or just don't know these technologies 362 00:22:36,600 --> 00:22:38,200 have a cut, what do we call it? 363 00:22:38,200 --> 00:22:44,280 Context window, which means it only can remember a certain number of what we call tokens, which 364 00:22:44,280 --> 00:22:47,120 are just parts of words in a single chat. 365 00:22:47,120 --> 00:22:49,800 And it doesn't remember across chats. 366 00:22:49,800 --> 00:22:54,640 So this can be very frustrating if you get like really deep into a great flow conversation 367 00:22:54,640 --> 00:22:59,240 and you're building stuff and suddenly it like just goes off the rails, starts hallucinating. 368 00:22:59,240 --> 00:23:01,080 It's like, no, you gotta do this. 369 00:23:01,080 --> 00:23:02,080 You gotta do this. 370 00:23:02,080 --> 00:23:03,440 And you're, you're fighting with it. 371 00:23:03,440 --> 00:23:06,760 And that's because it's just essentially ran out of memory. 372 00:23:06,760 --> 00:23:12,540 And so because of the way these technologies work, really all they're doing is looking 373 00:23:12,540 --> 00:23:18,520 at what has been said before and then predicting whatever the next word is going to be in a 374 00:23:18,520 --> 00:23:19,520 series. 375 00:23:19,520 --> 00:23:24,880 And again, no research on this, but I'm pretty confident that this works is if you just have 376 00:23:24,880 --> 00:23:30,120 a save function, which just says summarize everything that's done and done so far, tell 377 00:23:30,120 --> 00:23:32,800 me what the next step should be. 378 00:23:32,800 --> 00:23:35,600 And what's the question that's going to lead me to that? 379 00:23:35,600 --> 00:23:40,160 Again, it's like, you're just extending the context because you're just reminding it like, 380 00:23:40,160 --> 00:23:43,680 this is where we've come from, this is where we are, this is where we're going. 381 00:23:43,680 --> 00:23:47,320 Just to give it a nice little, a little booster, a nice little nudge. 382 00:23:47,320 --> 00:23:50,680 It's not perfect, but like I said, I'm building like all these courses in here. 383 00:23:50,680 --> 00:23:54,280 I save periodically and I almost have no issues with it. 384 00:23:54,280 --> 00:23:58,440 Like forgetting what we're trying to be doing and what it is, it's a lot easier to nudge 385 00:23:58,440 --> 00:23:59,880 it back on track. 386 00:23:59,880 --> 00:24:07,440 So valuable because we are limited in GPT-4 right now by about a 32,000 token context 387 00:24:07,440 --> 00:24:12,680 window, about 25,000 words, so maybe 20 pages of text, right? 388 00:24:12,680 --> 00:24:14,840 And that's input and output. 389 00:24:14,840 --> 00:24:19,640 But to have some of these skills, these prompts that we're developing, I think is going to 390 00:24:19,640 --> 00:24:22,400 carry us forward in a very positive way. 391 00:24:22,400 --> 00:24:26,640 When we get to million token context windows, there's some research out there that indicate 392 00:24:26,640 --> 00:24:29,520 that could be the reality in three to six months. 393 00:24:29,520 --> 00:24:31,880 It's really, really exciting to think about. 394 00:24:31,880 --> 00:24:35,160 Just to zoom back or rewind a little bit. 395 00:24:35,160 --> 00:24:39,880 So I started this company, Synaptic Labs, with my brother and we're really focused on 396 00:24:39,880 --> 00:24:41,560 the education side of AI. 397 00:24:41,560 --> 00:24:45,440 Again, just like Professor Synapse, we want to make this as accessible to everyone as 398 00:24:45,440 --> 00:24:47,200 possible. 399 00:24:47,200 --> 00:24:50,880 And so the best way to educate people is to educate educators. 400 00:24:50,880 --> 00:24:56,280 So one of the major projects we've been working on is in partnership with BSD Education. 401 00:24:56,280 --> 00:25:01,860 And we're essentially creating a generative AI for teachers course so that they can learn 402 00:25:01,860 --> 00:25:07,600 very quickly what this technology is, how it works, interact with it, and really think 403 00:25:07,600 --> 00:25:13,800 about how can I use this responsibly in the classroom to spend more time doing the things 404 00:25:13,800 --> 00:25:15,360 I like doing with my students. 405 00:25:15,360 --> 00:25:21,160 Probably hands-on projects, talking with them, helping some one-on-one teaching time, and 406 00:25:21,160 --> 00:25:28,200 spend less time having to lesson plan or create rubrics for an assignment and all that kind 407 00:25:28,200 --> 00:25:34,160 of stuff that just is very draining and time-consuming but perfect for something like chat GPT. 408 00:25:34,160 --> 00:25:39,480 And then finally, a line of lines of responsible AI, helping teachers really think through 409 00:25:39,480 --> 00:25:47,380 in this critical state, what is your personal classroom constitution for the use of AI? 410 00:25:47,380 --> 00:25:52,320 How are we defining that in your use and your students' use as responsible versus irresponsible 411 00:25:52,320 --> 00:25:57,160 so that you can clearly outline that as part of your culture from the get-go instead of 412 00:25:57,160 --> 00:26:01,580 what a lot of people are having to deal with right now, which is, I don't know what I'm 413 00:26:01,580 --> 00:26:02,580 supposed to do. 414 00:26:02,580 --> 00:26:04,280 I don't even really know how this technology works. 415 00:26:04,280 --> 00:26:06,560 There's no way for me to stop these students from using it. 416 00:26:06,560 --> 00:26:07,640 Let's just ban it outright. 417 00:26:07,640 --> 00:26:11,880 It's just in panic mode when there's just this great opportunity. 418 00:26:11,880 --> 00:26:18,200 If we can figure it out to show people how to use it well and to enhance ourselves and 419 00:26:18,200 --> 00:26:23,600 make things more equitable versus we're cheating on every test because that's the only thing 420 00:26:23,600 --> 00:26:26,360 I've been told by my teacher that chat GPT is good for. 421 00:26:26,360 --> 00:26:31,420 As time of recording, we're recording in early September the course that is titled Supercharge 422 00:26:31,420 --> 00:26:38,220 Your Teaching with AI Chatbots for Time-Saving Resource Creation hosted live on the BSD platform. 423 00:26:38,220 --> 00:26:43,240 We developed it, like Joe said, in partnership with BSD. 424 00:26:43,240 --> 00:26:46,960 And it's a three-hour course, very quick and digestible. 425 00:26:46,960 --> 00:26:51,400 It's also accredited by the CPD standards office, which is an international accreditation 426 00:26:51,400 --> 00:26:53,740 body for continuous professional development. 427 00:26:53,740 --> 00:26:59,160 So any teacher that is given a certain amount of money to do their annual training hours 428 00:26:59,160 --> 00:27:04,680 with or if you have a requirement to do, you'll get the certificate for those three hours 429 00:27:04,680 --> 00:27:05,680 with our course. 430 00:27:05,680 --> 00:27:07,200 And we're super proud of it. 431 00:27:07,200 --> 00:27:13,640 It's very low cost, 50 bucks, but hopefully it has the ultimate reach and benefit, we 432 00:27:13,640 --> 00:27:14,640 hope. 433 00:27:14,640 --> 00:27:18,800 I'm in a dungeon right now writing eight courses with my brother and another colleague of ours, 434 00:27:18,800 --> 00:27:19,800 Omer. 435 00:27:19,800 --> 00:27:24,880 We haven't shared the details quite yet, but there is a lot of learning on the horizon, 436 00:27:24,880 --> 00:27:25,880 people. 437 00:27:25,880 --> 00:27:28,080 A lot of good knowledge to share. 438 00:27:28,080 --> 00:27:29,520 Stay tuned, folks. 439 00:27:29,520 --> 00:27:31,520 We have some big things coming. 440 00:27:31,520 --> 00:27:36,320 We're going to be launching the SynthMind Discord server that is going to have a key 441 00:27:36,320 --> 00:27:39,200 component of that will be SynthMind Learning Labs. 442 00:27:39,200 --> 00:27:46,040 That'll be our section there where we can come and banter together, learn about how 443 00:27:46,040 --> 00:27:52,660 to talk to AI, wrestle with complex ideas, also network post jobs. 444 00:27:52,660 --> 00:27:56,920 We have a whole bunch of other exciting stuff planned too with different prompt library 445 00:27:56,920 --> 00:28:02,300 access and some other special things that we've done with Discord bots to help us stay 446 00:28:02,300 --> 00:28:06,800 on top of everything in the dynamic and fast paced world of AI. 447 00:28:06,800 --> 00:28:08,640 I want to thank Joe for coming on here. 448 00:28:08,640 --> 00:28:12,760 Joe, anything you want to throw in here at the end, we're of course going to send out 449 00:28:12,760 --> 00:28:16,760 links to everything we discussed about and including the invite to the Discord. 450 00:28:16,760 --> 00:28:21,040 Yeah, I was going to say, the only thing I'll plug is I have my own podcast, a chat with 451 00:28:21,040 --> 00:28:22,040 ChatGPT. 452 00:28:22,040 --> 00:28:25,640 It is literally just me talking to ChatGPT. 453 00:28:25,640 --> 00:28:29,760 I've been doing this since the very beginning, everyone, probably around February. 454 00:28:29,760 --> 00:28:31,760 I started doing a podcast a week. 455 00:28:31,760 --> 00:28:34,560 The olden times, the before times. 456 00:28:34,560 --> 00:28:41,720 I just try to have interesting conversations with ChatGPT around itself, AI, societal things. 457 00:28:41,720 --> 00:28:44,160 I just started experimenting more creatively. 458 00:28:44,160 --> 00:28:49,120 I've turned it into a dungeon master and we're going back and forth and doing a text based 459 00:28:49,120 --> 00:28:50,120 adventure game. 460 00:28:50,120 --> 00:28:53,720 Now, we're actually, I'm bringing in my brother and we're going to do it together. 461 00:28:53,720 --> 00:28:57,360 We're going to campaign together using ChatGPT as our dungeon master. 462 00:28:57,360 --> 00:28:59,240 Well, thank you so much, my man. 463 00:28:59,240 --> 00:29:05,400 I guess with that for Joe, the professor himself and GoToGo, we give you a hearty happy prompting 464 00:29:05,400 --> 00:29:12,600 everybody. 465 00:29:12,600 --> 00:29:19,000 As always, you can check out the show notes and links at howtotalkto.ai. 466 00:29:19,000 --> 00:29:21,200 That's all for this week's episode. 467 00:29:21,200 --> 00:29:36,640 Happy prompting, everyone.