00:00:10.511 --> 00:00:16.279 In in general, what I would say is that you can really see AI as
00:00:16.279 --> 00:00:22.046 a component of our software that we're going to be continuing to
00:00:22.046 --> 00:00:22.401 add.
00:00:22.471 --> 00:00:25.661 Umm, sort of modular functionality.
00:00:26.951 --> 00:00:31.000 You know, extensions just like we would with let's say, like
00:00:31.000 --> 00:00:35.247 payment providers or searching. There's gonna be an opportunity
00:00:35.247 --> 00:00:39.163 to bring in AI into aspects of our products as part of the
00:00:39.163 --> 00:00:40.291 product offering.
00:00:41.501 --> 00:00:46.449 And so we're gonna talk to some of that. We're not going to go
00:00:46.449 --> 00:00:51.161 into a ton of detail just on that today. A lot of the focus
00:00:51.161 --> 00:00:56.344 is gonna be looking at AI from a sort of professional perspective
00:00:56.344 --> 00:01:01.370 and how you can benefit from it within your career, not just at
00:01:01.370 --> 00:01:05.847 clarity, but in general as an augmentation tool for your
00:01:05.847 --> 00:01:10.559 professional organization and management of your particular
00:01:10.559 --> 00:01:10.951 role.
00:01:11.371 --> 00:01:15.568 Uh, we're going to go into a couple of examples of some of
00:01:15.568 --> 00:01:19.837 the team members using AI to help with marketing activities
00:01:19.837 --> 00:01:23.607 with creative content generation, for example. We're
00:01:23.607 --> 00:01:27.947 also gonna go into some detail around project management and
00:01:27.947 --> 00:01:32.500 how it can be used for project management software development.
00:01:32.500 --> 00:01:36.412 So I just want to encourage you to be listening for in
00:01:36.412 --> 00:01:40.894 particular, ideally how you can sort of expose yourself to new
00:01:40.894 --> 00:01:41.321 ideas.
00:01:41.401 --> 00:01:45.243 That some of the other team members who are gonna be helping
00:01:45.243 --> 00:01:49.148 conduct this training today that they've already kind of paid
00:01:49.148 --> 00:01:53.116 some of the dues, if not a lot of dues and figured out they've
00:01:53.116 --> 00:01:56.769 kind of gone through some challenges and trial and error.
00:01:56.769 --> 00:01:59.981 So hopefully you can benefit from that personally.
00:02:01.341 --> 00:02:05.552 You know, as you progress forward, so one of the other
00:02:05.552 --> 00:02:10.604 things that I'll just note as we get started is a lot of AI tools
00:02:10.604 --> 00:02:15.580 have premium options or premium paid seats that you can purchase
00:02:15.580 --> 00:02:19.638 just in general as a philosophical stance myself and
00:02:19.638 --> 00:02:24.614 the rest of the leadership team, as long as we can see that it's
00:02:24.614 --> 00:02:29.284 valuable and you know you're able to consistently use a tool
00:02:29.284 --> 00:02:30.891 paying for something.
00:02:31.131 --> 00:02:35.572 Paying for a premium seat, anything like that is absolutely
00:02:35.572 --> 00:02:40.235 supported for myself and from the rest of the leadership team.
00:02:40.235 --> 00:02:44.602 So for example, a lot of team members from this call might
00:02:44.602 --> 00:02:49.413 say, gosh, I would really like a premium seat on ChatGPT, great.
00:02:49.413 --> 00:02:53.632 Let's do it. Let's get you exposed to this tool and make
00:02:53.632 --> 00:02:57.776 sure that you're growing and learning and building your
00:02:57.776 --> 00:03:01.181 career in this aspect that, I mean everybody.
00:03:01.281 --> 00:03:04.958 Is going to need to take advantage of AI or get in some
00:03:04.958 --> 00:03:08.766 way passed over by it, just like just like you would with
00:03:08.766 --> 00:03:12.574 learning how to process word documents whenever Microsoft
00:03:12.574 --> 00:03:16.317 Word came out or, you know, et cetera. So it's a similar
00:03:16.317 --> 00:03:20.453 thought process around getting them more exposed to it, making
00:03:20.453 --> 00:03:24.327 sure that you have access to the tools. And I just want to
00:03:24.327 --> 00:03:28.135 encourage you that there shouldn't be a sense of like oh,
00:03:28.135 --> 00:03:31.221 this is expensive or I can't access it please.
00:03:31.311 --> 00:03:35.181 There's no that, you know, we really heavily support getting
00:03:35.181 --> 00:03:39.367 everyone what they would benefit from with regards to AI as well.
00:03:39.367 --> 00:03:43.427 Now with that said, I wanted to put a couple of stats out there
00:03:43.427 --> 00:03:47.550 just to kind of get you thinking about it as well and then we'll
00:03:47.550 --> 00:03:50.151 set up the actual the training training.
00:03:51.301 --> 00:03:54.401 Some interesting stats for you guys just to think about here.
00:03:56.191 --> 00:04:00.346 Just from a very, very high level and obviously you have to
00:04:00.346 --> 00:04:03.947 take all these stats with a grain of salt, but some
00:04:03.947 --> 00:04:07.825 interesting data is that 51% of ecommerce companies are
00:04:07.825 --> 00:04:09.971 currently using AI in some way.
00:04:10.621 --> 00:04:14.218 So obviously, us being so involved in ecommerce, we all
00:04:14.218 --> 00:04:18.394 need to be well aware of it. 48% of marketing leaders cite AI is
00:04:18.394 --> 00:04:21.991 making the most significant difference in how customers
00:04:21.991 --> 00:04:23.211 interact with them.
00:04:24.131 --> 00:04:28.002 64% of B to B marketers, consider AI to be valuable in
00:04:28.002 --> 00:04:29.761 their marketing strategy.
00:04:30.441 --> 00:04:33.025 And I promise I'll. I'll just. I'll. I'll put a few more of
00:04:33.025 --> 00:04:35.351 these stats out there. I just think it's interesting.
00:04:37.821 --> 00:04:42.281 37% of businesses and organizations employ AI.
00:04:44.831 --> 00:04:48.863 There are some quotes that AI will eliminate, and this one is
00:04:48.863 --> 00:04:52.961 a really specific one, but it's stating that AI will eliminate
00:04:52.961 --> 00:04:54.001 85 million jobs.
00:04:54.761 --> 00:04:58.748 But create 97 million jobs by 2025, so the thought of AI
00:04:58.748 --> 00:05:03.364 taking someone's job, I think we should all try to think of it as
00:05:03.364 --> 00:05:06.931 something that's malleable, where we're trying to.
00:05:08.631 --> 00:05:12.645 Grow with the AI changes that are happening and adapt to it
00:05:12.645 --> 00:05:16.391 and take advantage of the benefits of it so that we can
00:05:16.391 --> 00:05:20.606 create more work, create more opportunity with the same inputs
00:05:20.606 --> 00:05:21.141 from us.
00:05:22.191 --> 00:05:23.591 Just a few more notes here.
00:05:24.491 --> 00:05:29.085 61% of employees say the AI helps to improve their work
00:05:29.085 --> 00:05:30.151 productivity.
00:05:32.861 --> 00:05:38.841 And then I'll wrap up with this. The AI market is expanding at a
00:05:38.841 --> 00:05:42.981 compound annual growth rate of 38% per year.
00:05:43.921 --> 00:05:46.331 Between 2022 and 2030.
00:05:47.151 --> 00:05:50.511 Umm, that is a astronomical growth rate, so it's just gonna
00:05:50.511 --> 00:05:53.871 continue to take over more and more segments of the market.
00:05:54.981 --> 00:05:58.850 I don't. I think you know certain aspects of AI are high,
00:05:58.850 --> 00:06:02.918 but a lot of it is very real and very pragmatic. So the idea
00:06:02.918 --> 00:06:06.987 behind today is to go into three different areas and we have
00:06:06.987 --> 00:06:11.189 three different leaders within the business who are gonna take
00:06:11.189 --> 00:06:15.257 some time and share with you around their experience with AI
00:06:15.257 --> 00:06:19.326 and then give you some very hands on examples to start with.
00:06:19.326 --> 00:06:23.594 We're gonna start by looking at one of the product areas of AI.
00:06:23.594 --> 00:06:24.261 So a like.
00:06:24.431 --> 00:06:29.746 A potential augmentation of our software with a I would be in
00:06:29.746 --> 00:06:34.633 this case using conversational IVR which uses AI to have
00:06:34.633 --> 00:06:39.262 conversation instead of a traditional IVR interactive
00:06:39.262 --> 00:06:44.835 voice response system where you like call in and it's like press
00:06:44.835 --> 00:06:48.521 one if you want to press 2 conversational.
00:06:50.011 --> 00:06:53.823 A conversational IVR is you can literally just say whatever
00:06:53.823 --> 00:06:57.509 whenever and the AI picks up on it just like you would be
00:06:57.509 --> 00:07:01.384 talking with a person, and it directs you intelligently, but
00:07:01.384 --> 00:07:05.260 it is a bot. It's a AI based bot, so this is really powerful
00:07:05.260 --> 00:07:08.945 for one of our clients. And these are the types of things
00:07:08.945 --> 00:07:13.012 that we'll be bringing into our software development as product
00:07:13.012 --> 00:07:16.634 features. So we're gonna start with that. Brent Fife has
00:07:16.634 --> 00:07:20.573 volunteered to go over that. So I'm really excited for him to
00:07:20.573 --> 00:07:21.081 dive in.
00:07:21.561 --> 00:07:23.521 Just to kind of set that up.
00:07:24.701 --> 00:07:29.214 The set up the rest of the discussion, then Stephen Beer is
00:07:29.214 --> 00:07:33.577 gonna be going into once Brent finishes his presentation,
00:07:33.577 --> 00:07:38.541 Steven's gonna be going into how we're using AI within marketing.
00:07:39.771 --> 00:07:43.177 We're doing some pretty incredible things within our
00:07:43.177 --> 00:07:47.354 marketing team and we have some very strong thought leaders with
00:07:47.354 --> 00:07:50.760 using AI and an augmentation type of away within our
00:07:50.760 --> 00:07:54.552 marketing team. So I think you'll find some very pragmatic
00:07:54.552 --> 00:07:58.022 examples and this would be applicable to our software
00:07:58.022 --> 00:08:02.071 development team members, our project management team members,
00:08:02.071 --> 00:08:05.991 our OPS team members for ways that you can take advantage of
00:08:05.991 --> 00:08:09.461 ChatGPT as well as just general AI and how to prompt.
00:08:09.681 --> 00:08:14.222 AI and interact with it and so I think that will be very helpful.
00:08:14.222 --> 00:08:18.625 And then finally, Jeremy La Rose is gonna be going over several
00:08:18.625 --> 00:08:22.546 topics within AI in particular, we're gonna go over some
00:08:22.546 --> 00:08:26.674 software development and using AI for software development,
00:08:26.674 --> 00:08:30.802 augmentation, but also project management and how it can be
00:08:30.802 --> 00:08:35.067 used in general. So anyway with that said, Brent, if you were
00:08:35.067 --> 00:08:39.401 ready, I will turn it over to you to kick things off good Sir.
00:08:40.211 --> 00:08:42.361 This. Thank you. Yeah.
00:08:41.621 --> 00:08:42.471 OK, awesome.
00:08:45.321 --> 00:08:45.691 OK.
00:08:47.971 --> 00:08:52.533 All right. Uh, so for the past few months I've been working on
00:08:52.533 --> 00:08:57.022 this project called Desk. So as Chris mentioned, it's an IVRA
00:08:57.022 --> 00:09:01.295 interactive voice response project that's going to be used
00:09:01.295 --> 00:09:04.988 for, well, initially we're making it for a demo to
00:09:04.988 --> 00:09:09.694 hopefully branch out for larger project which will open a lot of
00:09:09.694 --> 00:09:14.401 doors for us to clarity. Because right now we're developing this
00:09:14.401 --> 00:09:18.311 IVR within a third party application called Corey AI.
00:09:18.631 --> 00:09:24.116 Umm, so the the long term plan is once we developed this this
00:09:24.116 --> 00:09:28.451 demo in core AI, we're going to make it our own.
00:09:29.811 --> 00:09:33.089 AI service within actually I don't know if you know or not.
00:09:33.089 --> 00:09:36.586 We have our own AI system here. Clarity that we've been I think
00:09:36.586 --> 00:09:40.137 we're using it for SEO and other things. But one of the plans is
00:09:40.137 --> 00:09:41.831 to hopefully build out our own.
00:09:43.571 --> 00:09:47.824 System that we can start selling to a lot more prospective
00:09:47.824 --> 00:09:48.401 clients.
00:09:49.111 --> 00:09:52.621 So umm, today I wanted to go over uh.
00:09:53.461 --> 00:09:59.212 Kind of the functionality of my the core AI and what we built
00:09:59.212 --> 00:10:01.531 out so far, so we whoops.
00:10:02.881 --> 00:10:03.321 There we go.
00:10:05.651 --> 00:10:09.715 So core AI has been it's a it's a system that's been around for,
00:10:09.715 --> 00:10:13.780 I think about six years now and it's got a lot of functionality.
00:10:13.780 --> 00:10:17.531 The one thing that I would say that it's primarily designed
00:10:17.531 --> 00:10:17.781 for.
00:10:18.241 --> 00:10:22.634 Umm chat bots. So today I'm going to demonstrate what I've
00:10:22.634 --> 00:10:27.101 been working on, but not in its IBR form, just because it's
00:10:27.101 --> 00:10:31.569 strengths right now or in its capability to respond through
00:10:31.569 --> 00:10:31.941 chat.
00:10:32.731 --> 00:10:35.760 So which is really cool about this is as Chris as mentioned
00:10:35.760 --> 00:10:38.789 this, this is conversational IVR. It's like you can jump to
00:10:38.789 --> 00:10:41.111 any different points within the conversation.
00:10:41.811 --> 00:10:45.848 So when we start here it's, you know, it prompts you for what
00:10:45.848 --> 00:10:50.015 you like to do on the how can I help you so you can pretty much
00:10:50.015 --> 00:10:50.861 say anything.
00:10:51.981 --> 00:10:53.251 And depending on what you.
00:10:54.361 --> 00:10:55.471 All humans.
00:10:56.571 --> 00:11:01.116 Let's see. It's thinking. Thinking. OK, well, that's
00:11:01.116 --> 00:11:02.231 great. Let's.
00:11:02.721 --> 00:11:03.281 Creepy.
00:11:03.061 --> 00:11:06.043 Let's try a different prompt like you were actually trying to
00:11:06.043 --> 00:11:09.025 do something here, so this is going to be a bot for a dentist
00:11:09.025 --> 00:11:09.361 office.
00:11:10.431 --> 00:11:15.101 So it's say hi. My name is Brent. I like.
00:11:23.171 --> 00:11:27.732 And then it just prompts you for this. No, I have not been here
00:11:27.732 --> 00:11:28.231 before.
00:11:29.811 --> 00:11:33.376 That's gonna ask you for a lots of different information. So the
00:11:33.376 --> 00:11:36.777 one thing to remember is it's originally designed for an RV R
00:11:36.777 --> 00:11:36.941 so.
00:11:38.301 --> 00:11:41.595 Some of the like the the, the, the grammar and maybe some of
00:11:41.595 --> 00:11:44.511 the spelling is not gonna be out the spelling I come.
00:11:45.381 --> 00:11:48.998 The punctuation is not going to be on point, so one thing you
00:11:48.998 --> 00:11:52.031 notice is I put my number in here and my number is.
00:11:57.981 --> 00:12:01.165 That's a little spaced out a little bit differently. We do
00:12:01.165 --> 00:12:04.458 that so that it it doesn't read it as a whole integer and it
00:12:04.458 --> 00:12:05.591 just reads it 1 by 1.
00:12:07.171 --> 00:12:09.161 So like, no, that's not my number.
00:12:16.431 --> 00:12:17.671 OK. Yeah.
00:12:21.081 --> 00:12:24.190 So now it's gonna ask him for my e-mail, and this is probably
00:12:24.190 --> 00:12:27.349 been one of the biggest issues with doing this through a third
00:12:27.349 --> 00:12:28.051 party service.
00:12:29.641 --> 00:12:34.121 Is we don't have the the control at the lower level control over
00:12:34.121 --> 00:12:38.532 the AI to so to end the machine just to figure out what exactly
00:12:38.532 --> 00:12:43.081 it's doing. And because this has been built for chat when you're.
00:12:43.721 --> 00:12:48.192 Uh calling in through an IVR and you say my e-mail is
00:12:48.192 --> 00:12:53.490 brent.fife@claritymis.com. It's not gonna necessarily read that
00:12:53.490 --> 00:12:58.540 specifically as an as an e-mail. And I say that because it's
00:12:58.540 --> 00:13:00.361 using a voice to text.
00:13:01.691 --> 00:13:05.894 To then send that data to the system here and the system. So
00:13:05.894 --> 00:13:09.890 if it's voice to the voice to text, it might be saying it
00:13:09.890 --> 00:13:13.817 might be interpreting your e-mail addresses. Brent five.
00:13:13.817 --> 00:13:17.882 Then AT instead of the at symbol, and so the system that's
00:13:17.882 --> 00:13:22.016 been trained to look for that at symbol may not necessarily
00:13:22.016 --> 00:13:22.911 recognize it.
00:13:25.311 --> 00:13:28.638 And we couldn't get it to work correctly because of that. For
00:13:28.638 --> 00:13:29.711 the longest time so.
00:13:31.301 --> 00:13:34.231 Without that granular type of control.
00:13:35.331 --> 00:13:38.201 It makes it a lot more difficult to.
00:13:39.881 --> 00:13:43.399 To build this out in the with the correct functionality, but
00:13:43.399 --> 00:13:46.801 my work around was to go into the bot and add regex so I I
00:13:46.801 --> 00:13:47.031 can.
00:13:48.891 --> 00:13:49.581 And my.
00:13:51.741 --> 00:13:53.751 That's email.com like this.
00:13:55.641 --> 00:13:56.791 And it recognizes.
00:13:58.201 --> 00:14:01.311 So that's kind of neat. And then it goes on.
00:14:03.521 --> 00:14:06.783 Umm. And keeps asking you all these questions until it
00:14:06.783 --> 00:14:09.451 eventually gives to the point where it will.
00:14:10.611 --> 00:14:13.921 Prompt you for the scheduling appointment and set up times.
00:14:14.871 --> 00:14:18.578 So that's kind of the end product of what we're looking to
00:14:18.578 --> 00:14:22.472 do, but actually make this thing and where the AI comes in to
00:14:22.472 --> 00:14:25.990 play is setting up the the utterances and the words for
00:14:25.990 --> 00:14:27.121 what it looks for.
00:14:29.081 --> 00:14:33.182 So what you do is in in this project we build stuff called
00:14:33.182 --> 00:14:36.727 dialogue tasks, and each task has sort of a linear
00:14:36.727 --> 00:14:37.561 progression.
00:14:38.241 --> 00:14:41.071 Umm that uh, the user has.
00:14:43.071 --> 00:14:47.391 Doesn't have to fulfill, but net sometimes does like so in this
00:14:47.391 --> 00:14:51.441 one you only really have one node. In here you've got your.
00:14:51.441 --> 00:14:55.559 You starting your welcome intent and then from there it just
00:14:55.559 --> 00:14:59.406 prompts you for an for an utterance. And once you say an
00:14:59.406 --> 00:15:00.081 utterance.
00:15:01.901 --> 00:15:05.931 It'll look through all the other dialogue tasks and then.
00:15:07.161 --> 00:15:11.189 Assign depending on how many utterances it recognizes for a
00:15:11.189 --> 00:15:15.283 given task, it will assign a high percentage of accuracy for
00:15:15.283 --> 00:15:19.445 it and then send you down that path. So and this one. This is
00:15:19.445 --> 00:15:23.741 where we went down before when we tried to schedule a cleaning.
00:15:24.831 --> 00:15:25.111 And.
00:15:27.351 --> 00:15:30.911 You're right. Here. We have our utterances.
00:15:32.101 --> 00:15:33.601 And once you enter these in.
00:15:34.061 --> 00:15:37.571 Uh, the bot will then look for these and and assign things.
00:15:38.691 --> 00:15:41.221 Saying the user down that path to pending on.
00:15:42.621 --> 00:15:45.251 How high it ranks compared to other utterances that may be
00:15:45.251 --> 00:15:45.741 recognized?
00:15:46.881 --> 00:15:49.771 Which is a really kind of a neat thing. And UM.
00:15:52.241 --> 00:15:53.001 Let's see.
00:15:54.091 --> 00:15:54.741 What else?
00:15:57.741 --> 00:15:58.261 Umm.
00:16:05.001 --> 00:16:07.898 So then you're training down here, you can go through all
00:16:07.898 --> 00:16:10.696 your utterances and see all all your different dialogue
00:16:10.696 --> 00:16:13.643 workflows and see all the utterances and you can go in and
00:16:13.643 --> 00:16:15.191 add a little bit more granular.
00:16:17.371 --> 00:16:19.531 Level for it to recognize.
00:16:20.401 --> 00:16:20.981 Where?
00:16:21.681 --> 00:16:24.121 The bot needs to go depending on what is said.
00:16:25.701 --> 00:16:26.901 And then you can test it.
00:16:30.421 --> 00:16:33.003 In here and and train it a little bit more through
00:16:33.003 --> 00:16:36.091 different means other than just going through each workflow.
00:16:37.391 --> 00:16:41.429 So you can you can train the bot by interacting with it through
00:16:41.429 --> 00:16:45.404 IVR. You can train it through these testing sequences here and
00:16:45.404 --> 00:16:46.981 you can train it through.
00:16:48.371 --> 00:16:52.312 Lots of different methods, which is, uh, it's a pretty powerful
00:16:52.312 --> 00:16:53.851 tool, but as you can see.
00:16:54.251 --> 00:16:57.394 Uh. And as mentioned before, there's lots of limitations to
00:16:57.394 --> 00:16:57.551 it.
00:16:58.741 --> 00:17:03.684 That would really go away and hopefully go away when if we
00:17:03.684 --> 00:17:08.710 develop our own AI system and not only that, they would it,
00:17:08.710 --> 00:17:13.903 the potential implications for clarity to have our own system
00:17:13.903 --> 00:17:14.741 is pretty.
00:17:15.031 --> 00:17:18.920 Pretty, pretty amazing. So I'm really looking forward to diving
00:17:18.920 --> 00:17:22.201 more into this and and for those of us going forward.
00:17:23.901 --> 00:17:26.430 We potentially will have more projects just like this, so you
00:17:26.430 --> 00:17:28.877 might be able to work on this in the future as well and I'm
00:17:28.877 --> 00:17:30.671 really excited to see where we're gonna go.
00:17:32.331 --> 00:17:37.457 With projects like this in the future, I know this wasn't too
00:17:37.457 --> 00:17:40.351 in depth about the project itself.
00:17:39.511 --> 00:17:41.671 And that was great, yeah.
00:17:41.041 --> 00:17:44.373 But if anybody else has any questions about this, umm, I'd
00:17:44.373 --> 00:17:45.051 be happy to.
00:17:46.821 --> 00:17:47.791 Answer that or field it.
00:17:52.271 --> 00:17:54.083 What is the? What is the training process look like for
00:17:54.083 --> 00:17:54.471 this Sprint?
00:17:57.441 --> 00:17:58.331 Yeah. So.
00:17:59.821 --> 00:18:02.131 What we want to do, we'd go into each one of the so.
00:18:03.571 --> 00:18:04.671 We go to.
00:18:05.811 --> 00:18:08.291 This workflow, the hiding workflow and.
00:18:12.011 --> 00:18:15.241 We go to this entity at the very top here.
00:18:22.201 --> 00:18:23.701 Soon you when you add it.
00:18:24.511 --> 00:18:25.571 Different words to it.
00:18:26.671 --> 00:18:30.451 Then you can train it to recognize it. And so once
00:18:26.751 --> 00:18:27.111 Hmm.
00:18:30.451 --> 00:18:33.861 somebody pick, if somebody calls into it and.
00:18:34.621 --> 00:18:38.431 They go down a path, let's say when they originally call in.
00:18:40.781 --> 00:18:43.974 They'll say I wanna do a teeth cleaning. If it go, it's if it's
00:18:43.974 --> 00:18:45.421 successfully recognizes that.
00:18:46.711 --> 00:18:51.206 There's no other paths to go down. It'll assign a higher
00:18:51.206 --> 00:18:56.253 value of chance going forward to that particular utterance, and
00:18:56.253 --> 00:19:01.220 so that is one way to train it. And then the other ways is you
00:19:01.220 --> 00:19:04.611 can manually go in and you can manipulate.
00:19:05.381 --> 00:19:05.871 Uh.
00:19:07.031 --> 00:19:07.471 Uh.
00:19:09.121 --> 00:19:12.790 Recognition threshold, but particularly the way to doing it
00:19:12.790 --> 00:19:16.031 here is to add different utterances to a particular.
00:19:18.601 --> 00:19:21.031 Dialogue or entity within it?
00:19:21.881 --> 00:19:22.881 OK, nice.
00:19:23.671 --> 00:19:23.871 Yep.
00:19:29.081 --> 00:19:32.135 And feel free to raise your hands if you want to queue up
00:19:32.135 --> 00:19:32.661 questions.
00:19:33.821 --> 00:19:36.271 For Brent and I'm just gonna say a few things here.
00:19:36.981 --> 00:19:40.801 Umm, while I wait to see if anyone has any other questions.
00:19:42.101 --> 00:19:44.201 Just to emphasize what Brent was saying.
00:19:44.841 --> 00:19:46.801 Uh, we are building out content.
00:19:47.981 --> 00:19:52.388 Into our content road map that will present clarity as an
00:19:52.388 --> 00:19:56.111 option to be able to build out custom AI models.
00:19:56.921 --> 00:20:01.847 So what you're seeing like Brent said is a a SAS based AI
00:20:01.847 --> 00:20:02.611 offering.
00:20:03.391 --> 00:20:06.935 That we have ramped up on and trained on, I say we, but it's
00:20:06.935 --> 00:20:07.981 mostly been Brent.
00:20:09.391 --> 00:20:12.891 I know that Ben is well and then Mickey has learned a lot about
00:20:12.891 --> 00:20:13.711 it as well, so.
00:20:14.941 --> 00:20:17.111 But basically we've learned.
00:20:18.541 --> 00:20:22.924 You know primarily through Brent doing all of this ramp up and
00:20:22.924 --> 00:20:25.011 training about this SAS based?
00:20:26.201 --> 00:20:31.399 Model essentially that like he said, the next progression will
00:20:31.399 --> 00:20:32.471 be for us to.
00:20:33.191 --> 00:20:36.571 Build our own model for this client scenario.
00:20:37.221 --> 00:20:40.882 And you might think what does that mean? What exactly are you
00:20:40.882 --> 00:20:44.602 saying? The idea behind AI is sort of the meat and potatoes of
00:20:44.602 --> 00:20:47.791 it, the heart of it is what's referred to as a model.
00:20:48.451 --> 00:20:52.546 And the concept is that the model is looking at data
00:20:52.546 --> 00:20:57.491 without, without going into the weeds. It's essentially looking
00:20:57.491 --> 00:21:02.281 at a bunch of data for inputs and desired outputs and putting
00:21:02.281 --> 00:21:03.131 weights on.
00:21:04.381 --> 00:21:05.851 A quote UN quote model.
00:21:06.731 --> 00:21:12.298 That allows you to, with similar inputs, get a desired output and
00:21:12.298 --> 00:21:17.864 So what you end up with is a lot of matrix math that gets done to
00:21:17.864 --> 00:21:22.671 create a specific model, and that's what gets exercised.
00:21:23.641 --> 00:21:28.303 As the heart of you know, making these decisions or responding or
00:21:28.303 --> 00:21:32.118 interacting whenever you're interacting with AI well,
00:21:32.118 --> 00:21:35.791 because we have our own AI server, like Brent said.
00:21:36.261 --> 00:21:40.793 Umm, and happy to elaborate on the details of that for anyone
00:21:40.793 --> 00:21:45.033 that's interested, but we actually have our own dedicated
00:21:45.033 --> 00:21:49.564 physical AI server at clarity within our colos center that we
00:21:49.564 --> 00:21:53.365 can use for projects. It's really cool. We're gonna
00:21:53.365 --> 00:21:57.385 continue investing in AI directly, but we can actually
00:21:57.385 --> 00:22:01.771 build models for clients or what's probably more practical.
00:22:02.641 --> 00:22:06.608 Is to take a model that's open sourced and do what's called
00:22:06.608 --> 00:22:07.401 fine tuning.
00:22:08.141 --> 00:22:11.411 So we can make sure that it meets the right legal
00:22:11.411 --> 00:22:15.335 requirements for copyright, et cetera. And then essentially
00:22:15.335 --> 00:22:19.455 take all of the hard work that someone else has done. Think of
00:22:19.455 --> 00:22:23.445 like a student who's graduated from college, and that's what
00:22:23.445 --> 00:22:27.238 the AI is. They've got this general knowledge, maybe even
00:22:27.238 --> 00:22:31.096 like specialized in some way, but then they get on the job
00:22:31.096 --> 00:22:35.151 training for a year within a specific company. That's kind of
00:22:35.151 --> 00:22:39.271 what fine tuning is. It's taking a model that's already there.
00:22:39.531 --> 00:22:43.687 And then making it very specifically quote UN quote
00:22:43.687 --> 00:22:48.242 fine-tuned by adding more iterations of input output and
00:22:48.242 --> 00:22:52.956 then changing the weightings so that it matches a specific
00:22:52.956 --> 00:22:57.671 scenario that a client in this thought process would want.
00:22:58.531 --> 00:23:02.403 So we can create our own models for our ecommerce platform and
00:23:02.403 --> 00:23:05.845 for our Connect platform. For example, we also have the
00:23:05.845 --> 00:23:09.348 ability to work on custom projects for clients and build
00:23:09.348 --> 00:23:13.036 out models for them or again fine tune them. So it's really
00:23:13.036 --> 00:23:16.662 interesting. I think the big things that I would encourage
00:23:16.662 --> 00:23:20.473 you to take away from this is that from a marketing and sales
00:23:20.473 --> 00:23:24.038 perspective, pretty much anything that client talks about
00:23:24.038 --> 00:23:27.541 with AI, we can do it. We've been investing a lot in it.
00:23:28.281 --> 00:23:29.671 And really into the weeds.
00:23:31.211 --> 00:23:36.292 If you look, I don't know if you can see it from here, but on my
00:23:36.292 --> 00:23:41.294 bookshelf back there, all those books right there are on AI and
00:23:41.294 --> 00:23:43.561 there are a lot of them that.
00:23:44.211 --> 00:23:48.295 You know, myself and other folks have been hammering on just to
00:23:48.295 --> 00:23:51.741 make sure we really understand the foundations of it.
00:23:52.911 --> 00:23:55.111 From a development perspective.
00:23:56.481 --> 00:23:59.796 And you know project management perspective, I would just
00:23:59.796 --> 00:24:03.455 encourage you as you're talking with clients and AI comes up or
00:24:03.455 --> 00:24:06.941 you're thinking about it from a sort of product perspective.
00:24:07.541 --> 00:24:12.023 Just open your mind up if it's not already to the idea that we
00:24:12.023 --> 00:24:16.363 can pretty much do anything that's out there with regards to
00:24:16.363 --> 00:24:17.501 AI for a client.
00:24:18.621 --> 00:24:22.425 Or as a product offering. And obviously we have to function
00:24:22.425 --> 00:24:25.911 the business and you know we have to prioritize things
00:24:25.911 --> 00:24:29.271 intelligently. But one of the really cool things is.
00:24:30.251 --> 00:24:34.688 I guess 1012 months ago, Brent really to my knowledge, Brent,
00:24:34.688 --> 00:24:38.481 you hadn't really worked on AI in this way at least.
00:24:39.221 --> 00:24:40.441 And so now you've.
00:24:39.551 --> 00:24:40.761 No, it was my first. Yeah.
00:24:41.251 --> 00:24:45.618 Yeah. So like, like Brent said, like, if you're interested and
00:24:45.618 --> 00:24:50.123 you'll want to learn and work on these types of projects as they
00:24:50.123 --> 00:24:51.371 come in, you know?
00:24:52.301 --> 00:24:56.338 I there will be a lot of a lot of desire for you to have
00:24:56.338 --> 00:25:00.800 resources and support to work with Brent and some of the other
00:25:00.800 --> 00:25:04.837 leaders who have have been investing in this so that you
00:25:04.837 --> 00:25:09.370 can learn and gain a foothold. So any other comments or Fife or
00:25:09.370 --> 00:25:12.061 any questions from anyone, feel free.
00:25:16.921 --> 00:25:17.991 Yes, Sir, Sir Sacco.
00:25:20.141 --> 00:25:24.066 Yeah, I got a question just for like you know front end back end
00:25:24.066 --> 00:25:27.869 developers. Is there any maybe like recommended resources that
00:25:27.869 --> 00:25:29.801 we can use to just kind of like?
00:25:30.921 --> 00:25:34.203 I guess familiarize ourselves bit more with development in
00:25:34.203 --> 00:25:37.651 this area. You know for like our own time from personal time.
00:25:37.101 --> 00:25:41.329 Yeah, yeah, there is an awesome course on AI that I'll share
00:25:38.411 --> 00:25:38.621 The.
00:25:41.329 --> 00:25:45.903 with you, Sacco. And then if you don't mind disseminating that to
00:25:45.261 --> 00:25:45.611 Go.
00:25:45.903 --> 00:25:50.061 the team, it's it's free and it's got some great supporting
00:25:48.261 --> 00:25:48.601 Sure.
00:25:50.061 --> 00:25:54.496 course material and it's pretty easy to go through. But it gets
00:25:54.496 --> 00:25:58.723 into the foundations of how it mechanically works. If you're
00:25:58.723 --> 00:26:02.604 interested and it has some immediate practical hands on
00:26:00.171 --> 00:26:00.461 Yeah.
00:26:02.604 --> 00:26:06.693 outputs that you can take advantage of, like being able to
00:26:06.693 --> 00:26:06.901 do.
00:26:07.041 --> 00:26:11.100 Visualization with a SAS based AI model where you can
00:26:11.100 --> 00:26:15.760 incorporate that into a React application like within like 30
00:26:15.760 --> 00:26:16.361 minutes.
00:26:17.641 --> 00:26:20.891 So I think that would be a really practical situation. So I
00:26:20.891 --> 00:26:22.191 can share that with you.
00:26:22.941 --> 00:26:23.711 To disseminate.
00:26:24.331 --> 00:26:25.801 Yeah, that'd be great. That'd be awesome.
00:26:26.531 --> 00:26:28.301 OK, awesome. Yeah, great question.
00:26:29.891 --> 00:26:33.214 OK, we'll thanks and again, feel free if anyone else has comments
00:26:33.214 --> 00:26:35.983 or questions. And thank you again, Brent, uh great job
00:26:35.983 --> 00:26:39.205 prepping and and going over that I think that went really well.
00:26:39.205 --> 00:26:40.161 So thank you again.
00:26:40.721 --> 00:26:41.211 Umm.
00:26:42.521 --> 00:26:46.742 OK. And with that, I'm excited to bring Steven in next. And
00:26:46.742 --> 00:26:51.034 just in general, what I can tell you just to set this up for
00:26:51.034 --> 00:26:54.481 Steven is he's conducted training on AI with the
00:26:54.481 --> 00:26:58.632 marketing team and and there's so much detail and fidelity
00:26:58.632 --> 00:27:03.064 obviously we won't have time to go into all of that today, but
00:27:03.064 --> 00:27:07.566 I'll just note that if you are interested in some of the topics
00:27:07.566 --> 00:27:11.787 he's covering, you're welcome to request and we can get you
00:27:11.787 --> 00:27:15.868 access to some of the more detailed training that Stevens
00:27:15.868 --> 00:27:16.571 completed.
00:27:16.971 --> 00:27:20.134 And some of the other marketing team members have put together
00:27:20.134 --> 00:27:23.197 uh anyway, without further ado, over to you, Steven, to kick
00:27:23.197 --> 00:27:26.411 things off on what you've been seeing and how you've been using
00:27:26.411 --> 00:27:28.921 it and as well as the rest of the marketing team.
00:27:30.631 --> 00:27:34.503 Excellent. Thank you very much. All right, share my screen here.
00:27:34.503 --> 00:27:38.256 I'm Steve Beer content writer on the marketing team. I've been
00:27:38.256 --> 00:27:41.949 with clarity for two years. Next month, since we were talking
00:27:41.949 --> 00:27:45.701 about sugar earlier, I wanted to make sure that we didn't have
00:27:45.701 --> 00:27:48.441 any confusion as to what we're talking about.
00:27:49.791 --> 00:27:53.514 So the one on the left is me. Here. I'm here. Yes. Here to
00:27:50.361 --> 00:27:51.151 That's amazing.
00:27:52.451 --> 00:27:53.321 Very nice.
00:27:53.514 --> 00:27:57.490 tell you about how we're using AI to generate more content and
00:27:57.490 --> 00:28:01.402 faster content so that we can cover more keywords. And so the
00:28:01.402 --> 00:28:05.189 Google sees us as more of an authority than our competitors
00:28:05.189 --> 00:28:05.441 are.
00:28:06.531 --> 00:28:09.956 Feel free to ask me questions, but otherwise I'll just assume
00:28:09.956 --> 00:28:13.160 you're enjoying the presentation and laughing like you're
00:28:13.160 --> 00:28:15.701 watching a donkey eating figs. So here we go.
00:28:16.651 --> 00:28:20.797 You probably won't be surprised to know that ChatGPT itself is
00:28:20.797 --> 00:28:22.771 at the heart of our I writing.
00:28:23.561 --> 00:28:26.611 But we use it in different ways so.
00:28:28.321 --> 00:28:32.655 What is ChatGPT love? You understand this a lot more than
00:28:32.655 --> 00:28:33.551 I will, but.
00:28:34.701 --> 00:28:39.900 It's a large language model, parenting questions, writing and
00:28:39.900 --> 00:28:44.847 writing code answering things. Remember the generative pre
00:28:44.847 --> 00:28:49.878 trained transformer, which is where the GPT comes from, and
00:28:49.878 --> 00:28:54.825 the two iterations that we've been really focusing on have
00:28:54.825 --> 00:28:59.437 been GPT 3.5 released in November, and GT4 released in
00:28:59.437 --> 00:29:00.611 March of 2023.
00:29:03.401 --> 00:29:07.230 All right, so I'm going to show you how we're writing articles
00:29:07.230 --> 00:29:10.755 with AI now and a little bit about how we've been writing
00:29:10.755 --> 00:29:14.341 recently, because this is what my last four trainings look
00:29:14.341 --> 00:29:17.501 like. There's been a lot happening in the world of.
00:29:18.681 --> 00:29:21.651 Of AI and marketing.
00:29:24.001 --> 00:29:28.111 So how do we interact with it? Well, you all know that.
00:29:28.961 --> 00:29:33.031 Chat is essentially just a chat bot so.
00:29:34.311 --> 00:29:35.571 You ask it questions.
00:29:38.061 --> 00:29:40.370 Oh, something went wrong. Hold on. We're going to have it's
00:29:40.370 --> 00:29:40.831 doing again.
00:29:51.161 --> 00:29:54.581 And it tells you it doesn't have feelings. Then we're gonna ask.
00:29:57.731 --> 00:29:59.641 What year was Lake Travis created?
00:30:01.831 --> 00:30:04.220 And you'll all learn a little something about the uh, like
00:30:04.220 --> 00:30:05.921 that we're going to be in a couple weeks.
00:30:07.981 --> 00:30:08.511 Here here.
00:30:08.191 --> 00:30:10.041 So yeah.
00:30:12.061 --> 00:30:15.417 So of course the real power comes from the long, long form
00:30:15.417 --> 00:30:19.001 writing that the marketing team does. And of course that comes
00:30:19.001 --> 00:30:22.129 from prompts and prompts involved telling GPT what you
00:30:22.129 --> 00:30:25.372 want and how you want it to handle it, and it's simplest
00:30:25.372 --> 00:30:28.671 form. A prompt tells it that what you want to write about
00:30:28.671 --> 00:30:31.856 what kind of content you want, who the audience is, and
00:30:31.856 --> 00:30:35.383 anything else that you want to include or parameters you want
00:30:35.383 --> 00:30:38.682 it to follow. So I asked it to write a Limerick about the
00:30:38.682 --> 00:30:41.071 Austin software company Clarity Ventures.
00:30:42.081 --> 00:30:43.301 We'll see what comes up with.
00:30:48.191 --> 00:30:49.471 Ah, isn't that sweet?
00:30:50.811 --> 00:30:54.748 Alright, so you'll notice that I mentioned which clarity ventures
00:30:54.748 --> 00:30:58.565 I was talking about, because of course there are many different
00:30:58.565 --> 00:31:01.965 clarity ventures across the country and that's basically
00:31:01.965 --> 00:31:05.783 just the traditional garbage in garbage out and that's the most
00:31:05.783 --> 00:31:09.421 I remember from my basic coding class back many decades ago.
00:31:10.431 --> 00:31:12.181 Some of the common prompts.
00:31:12.961 --> 00:31:16.251 That we use that in a very the very simple.
00:31:18.041 --> 00:31:22.599 I mean, you're basically telling it how to act, so you've got a
00:31:22.599 --> 00:31:24.451 lot of problems like this.
00:31:27.241 --> 00:31:29.901 Now prompts can get very complex, of course.
00:31:30.941 --> 00:31:35.226 One that we used to rewrite content to make GPT 3.5 content
00:31:35.226 --> 00:31:39.296 less identifiable as as AI content is the perplexity and
00:31:39.296 --> 00:31:40.581 burstiness prompt.
00:31:41.281 --> 00:31:46.091 And we're basically what this comes down to, it's it's saying.
00:31:48.141 --> 00:31:49.481 Hey GTP, listen up.
00:31:50.131 --> 00:31:52.420 To sound more human, use a combination of more and less
00:31:52.420 --> 00:31:54.546 complex sentences with a combination of shorter and
00:31:54.546 --> 00:31:55.241 longer sentences.
00:31:56.011 --> 00:31:58.511 And remember everything I just said, and here's the text to
00:31:58.511 --> 00:32:00.928 rewrite and then you give it the text that you want it to
00:32:00.928 --> 00:32:01.261 rewrite.
00:32:02.511 --> 00:32:07.505 Umm, so all of these are the tools we've been using. Ohh
00:32:07.505 --> 00:32:08.381 excuse me.
00:32:09.621 --> 00:32:12.211 There we go. The tools that we've been using.
00:32:13.451 --> 00:32:18.022 AI PRM, Zim writer and hyper right now, these tools use
00:32:18.022 --> 00:32:23.001 prompts that are usually human created, that focus on how to
00:32:23.001 --> 00:32:28.061 write something, and an example might be something like this.
00:32:29.361 --> 00:32:32.661 Write an article with these particular parameters.
00:32:33.281 --> 00:32:35.651 Make it more comprehensive than our competitors.
00:32:36.871 --> 00:32:40.637 Give me the following information that I'm asking for,
00:32:40.637 --> 00:32:41.801 including an FAQ.
00:32:42.621 --> 00:32:43.291 Now.
00:32:48.771 --> 00:32:51.621 No, no, no. Hold on. OK, so with.
00:32:55.871 --> 00:32:57.701 These tools essentially.
00:32:59.071 --> 00:33:01.671 Come down to writing prompts.
00:33:03.751 --> 00:33:04.731 A writing prompts.
00:33:05.901 --> 00:33:08.171 And ohhh yeah.
00:33:08.251 --> 00:33:08.601 Umm.
00:33:09.921 --> 00:33:13.362 Which when you break it down, they just call them very simple
00:33:13.362 --> 00:33:15.471 things like behind each one of these.
00:33:16.591 --> 00:33:17.681 The magic editor.
00:33:18.611 --> 00:33:23.143 The tools like e-mail Reply Rewrite article this is
00:33:23.143 --> 00:33:24.101 typewriter.
00:33:25.831 --> 00:33:29.521 We've got. We're using the a IR PM.
00:33:31.441 --> 00:33:33.071 Over in here.
00:33:37.401 --> 00:33:41.212 So everything that you just saw with that right and article
00:33:41.212 --> 00:33:44.958 based that's in here as prompts and outrank article. So of
00:33:44.958 --> 00:33:49.023 course people are making tools that have the prewritten prompts
00:33:49.023 --> 00:33:52.452 and of course there are thousands out there, some are
00:33:52.452 --> 00:33:55.691 better than others and a lot of them get uploaded.
00:33:57.511 --> 00:34:03.161 So, like with AI RPM, there's it's community created uses 3.5.
00:34:03.161 --> 00:34:05.851 Unless you have GPT 4 account.
00:34:06.671 --> 00:34:10.369 Zoom writer it's good for writing long form blogs of 2500
00:34:10.369 --> 00:34:10.751 words.
00:34:11.641 --> 00:34:13.931 Or more and typewriter.
00:34:14.551 --> 00:34:19.226 Based on you 2P four but can use GPT 3.5 and it goes a lot faster
00:34:19.226 --> 00:34:20.501 when it does that.
00:34:22.121 --> 00:34:25.329 It's better accepting headlines with subheadings, but we just
00:34:25.329 --> 00:34:28.536 started using yesterday so I don't have a whole lot to say on
00:34:28.536 --> 00:34:28.691 it.
00:34:31.081 --> 00:34:31.651 OK.
00:34:35.601 --> 00:34:39.091 Because of the considerable problems with these tools that
00:34:39.091 --> 00:34:42.462 we'll talk about in a minute, we're not just copying and
00:34:42.462 --> 00:34:45.951 pasting it to the website, so here we're gonna talk about.
00:34:47.121 --> 00:34:50.020 All that content and how it we use it to outrank our
00:34:50.020 --> 00:34:51.661 competitors, we've been using.
00:34:54.141 --> 00:34:57.221 A tool called server Server essentially takes all your
00:34:57.221 --> 00:35:00.301 competition, and in this case, here's the competition.
00:35:01.021 --> 00:35:01.711 It is.
00:35:02.941 --> 00:35:06.859 Noting how many times they use keywords, how many headings
00:35:06.859 --> 00:35:10.312 they're using, how many paragraphs, how many images
00:35:10.312 --> 00:35:14.164 you're essentially trying to reverse engineer the best of
00:35:14.164 --> 00:35:16.621 your competition and make it better.
00:35:17.861 --> 00:35:20.687 Here's the keywords we'll be putting in the different
00:35:20.687 --> 00:35:21.891 headings that it wants.
00:35:23.441 --> 00:35:25.181 And it's got a nice little.
00:35:26.141 --> 00:35:29.813 Needle over here that just so fun to use. So initially I told
00:35:29.813 --> 00:35:33.249 it. Tell me how to boost conversions using these keywords
00:35:33.249 --> 00:35:36.388 and this is the simplest of prompts. As you probably
00:35:36.388 --> 00:35:40.119 understand gave me this. I said rewrite and double that length
00:35:40.119 --> 00:35:42.251 and so it did. So when it gives us.
00:35:43.601 --> 00:35:46.231 Are text then send it back in there?
00:35:47.301 --> 00:35:50.173 Umm, so it's just that since it's up to 18, it don't have
00:35:50.173 --> 00:35:53.145 enough words, don't have enough headings, don't have enough
00:35:53.145 --> 00:35:53.491 images.
00:35:54.591 --> 00:35:59.301 And I'm going to keep working that until it it at least beats.
00:36:00.041 --> 00:36:03.261 The 72 that that come that competitor has.
00:36:04.891 --> 00:36:07.546 Of course, other things are taken into account like domain
00:36:07.546 --> 00:36:10.336 authority, but if we can come in even second or third, that's
00:36:10.336 --> 00:36:11.101 pretty darn good.
00:36:12.581 --> 00:36:14.491 OK, so that is.
00:36:16.141 --> 00:36:19.471 Surfer. Now the question becomes.
00:36:22.291 --> 00:36:26.664 Why are we still paying this guy? And that is an excellent
00:36:26.664 --> 00:36:31.259 question, because a lot can go wrong with not only the tools,
00:36:31.259 --> 00:36:32.741 but chat in general.
00:36:35.361 --> 00:36:38.165 It'll overwrite or underwrite for some reason one of the
00:36:38.165 --> 00:36:41.264 hardest things it has to do is to stick to the number of words
00:36:41.264 --> 00:36:44.166 you give it. If you ask for 1500, it will give you 800. If
00:36:44.166 --> 00:36:46.871 you ask for 3000, it will still probably give you 800.
00:36:48.091 --> 00:36:52.318 Give you rather simple blog, so you'll have to to keep writing
00:36:52.318 --> 00:36:55.001 to get to the proper amount of content.
00:36:55.711 --> 00:36:58.974 Umm, this is just a theory, but I think the tools are sending
00:36:58.974 --> 00:37:01.815 the prompts requesting information in chunks and then
00:37:01.815 --> 00:37:04.604 they bring it back and reassemble it and deliver it.
00:37:04.604 --> 00:37:07.971 And I think and that's a it does lead to a lot of duplicate and
00:37:07.971 --> 00:37:08.971 repetitive content.
00:37:09.951 --> 00:37:13.199 It can only review what it's already written to. 3000 words.
00:37:13.199 --> 00:37:16.606 I have had articles where it'll say like Amazon has 300,000,000
00:37:16.606 --> 00:37:19.747 customers at the beginning, but by the time it gets to the
00:37:19.747 --> 00:37:22.835 conclusion it will say it will, it will reference another
00:37:22.835 --> 00:37:23.261 article.
00:37:23.901 --> 00:37:28.471 And say it's over 250 million so it can give you different stats.
00:37:29.521 --> 00:37:31.831 Even within the same article it's delivering to you.
00:37:32.601 --> 00:37:35.823 Uh still needs to be read for factual accuracy. That's of
00:37:35.823 --> 00:37:39.212 course a big part of the human element to write our calls to
00:37:39.212 --> 00:37:39.601 action.
00:37:41.521 --> 00:37:44.730 To get ready for the website design and to removed some.
00:37:44.730 --> 00:37:48.277 Sometimes it'll actually respond and put up a responses to the
00:37:48.277 --> 00:37:51.317 prompts, and they're like certainly I can rewrite and
00:37:51.317 --> 00:37:54.695 expand that for you, and you don't want that getting on the
00:37:54.695 --> 00:37:57.791 wild because then immediately everyone knows it was AI
00:37:57.791 --> 00:37:58.241 content.
00:37:59.101 --> 00:38:01.051 And of course, since it's drawing from other.
00:38:01.731 --> 00:38:05.889 Websites it could advertise for a competitor, so you you don't
00:38:05.889 --> 00:38:07.671 want to have it saying hey.
00:38:08.601 --> 00:38:11.171 Be sure to go and use this particular.
00:38:12.121 --> 00:38:13.081 A company.
00:38:17.151 --> 00:38:21.718 All right. So probably the worst thing that it does is it really
00:38:21.718 --> 00:38:25.161 wants to give you an answer and it will fake it.
00:38:26.081 --> 00:38:29.075 Much like your uncle on his Facebook page, chat can be very
00:38:29.075 --> 00:38:31.321 sure about stuff it knows nothing about. So.
00:38:32.181 --> 00:38:35.282 Well, it could be called confidence. Stupidity for AI,
00:38:35.282 --> 00:38:38.891 it's called hallucinations. Or more accurately, confabulations.
00:38:39.611 --> 00:38:41.761 So I have a little example here.
00:38:45.321 --> 00:38:46.231 I took.
00:38:47.421 --> 00:38:48.231 2.
00:38:50.611 --> 00:38:54.514 Inventory management software. The first one is legitimate. The
00:38:54.514 --> 00:38:58.113 second one is legitimate or excuse me. The first first one
00:38:58.113 --> 00:39:01.528 is, the third one is the middle one is just something I
00:39:01.528 --> 00:39:04.821 completely came up with and it will probably tell me.
00:39:05.621 --> 00:39:09.023 That it's a popular inventory management software, even though
00:39:09.023 --> 00:39:10.481 that company doesn't exist.
00:39:11.231 --> 00:39:14.392 So you've got a really got to be careful for that in case
00:39:14.392 --> 00:39:17.281 something gets in there that you don't want want to.
00:39:18.771 --> 00:39:21.731 And just to add to end this on a fun note.
00:39:22.471 --> 00:39:28.067 Sometimes it's use of literary devices, idioms, saying sayings,
00:39:28.067 --> 00:39:28.941 et cetera.
00:39:29.931 --> 00:39:33.925 It really does not do a good job yet, so here are some examples
00:39:33.925 --> 00:39:37.232 of some things that have actually come up. These are
00:39:37.232 --> 00:39:37.981 screenshots.
00:39:38.821 --> 00:39:41.071 Of content it has delivered.
00:39:45.641 --> 00:39:49.215 Now when it completely changed course right in the middle of
00:39:49.215 --> 00:39:50.211 giving me an FAQ.
00:39:55.861 --> 00:39:56.951 Makes a lot of sense.
00:40:01.261 --> 00:40:03.711 It's now quoting TS Eliot for some reason.
00:40:06.611 --> 00:40:07.401 Or this is lovely.
00:40:08.831 --> 00:40:11.471 The sun was setting the day was ending, and with it the search
00:40:11.471 --> 00:40:13.021 for a lead management software tool.
00:40:14.371 --> 00:40:16.951 As I'm sure we all do when we're in Hawaii and this is my
00:40:16.951 --> 00:40:17.351 favorite.
00:40:19.781 --> 00:40:24.556 The clouds of payment processing part to make way for the son of
00:40:24.556 --> 00:40:25.511 checkout.com.
00:40:26.771 --> 00:40:27.981 And that's it, everybody.
00:40:28.701 --> 00:40:31.241 If you have any questions, go right on ahead.
00:40:32.781 --> 00:40:36.428 Thanks to you and great job and queue out if you guys don't mind
00:40:36.428 --> 00:40:39.851 raising your hands if we have several, wanna ask, feel free.
00:40:51.141 --> 00:40:52.681 Know those examples are hilarious.
00:40:53.911 --> 00:40:56.555 And they're fun to find, too. But you're just. You're shaking
00:40:56.555 --> 00:40:56.981 your head.
00:40:58.711 --> 00:40:59.761 I like your.
00:41:00.991 --> 00:41:01.461 Uh.
00:41:02.411 --> 00:41:05.032 I like the fact that you directly addressed that question
00:41:05.032 --> 00:41:07.698 head on of why do we? Why are we still paying this guy and
00:41:07.698 --> 00:41:08.511 talking about how.
00:41:09.531 --> 00:41:13.345 Uh, I think it's really cool and important to note that, uh, AI
00:41:13.345 --> 00:41:17.218 is really awesome and powerful, but at the end of the day, it is
00:41:17.218 --> 00:41:21.031 still a tool and people who know how to use tools are valuable.
00:41:21.861 --> 00:41:23.261 Yes, you're here.
00:41:21.871 --> 00:41:22.351 Umm.
00:41:23.191 --> 00:41:24.421 Yeah, it's very true.
00:41:25.801 --> 00:41:31.119 They recently found that 3.5 scored in the bottom 10% when
00:41:31.119 --> 00:41:35.986 doing a fake bar exam and chat for scored in the 90th
00:41:35.986 --> 00:41:39.681 percentile. So yes, it's getting better.
00:41:40.961 --> 00:41:44.628 But of course, I think they'll always now just for job
00:41:44.628 --> 00:41:48.361 security. I know that I'm always going to be important.
00:41:49.281 --> 00:41:50.501 That is very true.
00:41:49.491 --> 00:41:52.963 There will always be something. There will always be a the human
00:41:52.963 --> 00:41:56.008 touch that is trying to replicate and B making sure that
00:41:56.008 --> 00:41:57.931 it didn't screw up something awful.
00:42:01.111 --> 00:42:01.551 Here here.
00:42:02.581 --> 00:42:03.791 And over to you TMK.
00:42:06.151 --> 00:42:09.528 I was just curious since I've been messing around with the GPT
00:42:09.528 --> 00:42:12.744 3.5 and four myself when it comes to something like content
00:42:12.744 --> 00:42:16.013 length, let's say just won't spit out what you ask. What are
00:42:16.013 --> 00:42:17.621 some effective strategies for?
00:42:18.321 --> 00:42:21.325 Basically making it listen. Is there a way to strong orbit or
00:42:21.325 --> 00:42:23.941 there's some effective strategies for that? Because I
00:42:22.771 --> 00:42:23.301 Umm.
00:42:23.941 --> 00:42:26.751 know even when if it comes to code or content or whatever
00:42:26.751 --> 00:42:27.381 you're doing.
00:42:28.051 --> 00:42:30.121 Sometimes it just doesn't wanna cooperate.
00:42:28.661 --> 00:42:29.361 My most.
00:42:30.801 --> 00:42:34.682 Exactly my most effective strategy has been essentially
00:42:34.682 --> 00:42:38.979 taking the outline and if we have, you know, if if I've asked
00:42:38.979 --> 00:42:43.345 for five subheads, then I'll run the subhead through again. It
00:42:43.345 --> 00:42:47.364 seems to be better responding to, like, give me 100 words
00:42:47.364 --> 00:42:51.661 about this or give me 400 words once you get above, you know,
00:42:51.661 --> 00:42:55.750 like the 800. It's not even gonna listen. So if I say like
00:42:53.771 --> 00:42:54.121 Umm.
00:42:55.750 --> 00:42:59.908 give me 400 words on this and then do that four times, then
00:42:59.908 --> 00:43:01.571 suddenly I'm up to that.
00:43:01.671 --> 00:43:05.011 That 2000, so essentially running the subheads again.
00:43:06.241 --> 00:43:10.266 If it's not giving me everything I want, but it also possible it
00:43:10.266 --> 00:43:14.353 gave me additional subheads that led to, you know, an eye opening
00:43:14.353 --> 00:43:18.378 experience for me and then I can run those as well. So it I gave
00:43:18.378 --> 00:43:22.341 it one thing, it gave me back a little bit more, but not enough
00:43:22.341 --> 00:43:26.180 content and then I can use all the subheads to run through it
00:43:26.180 --> 00:43:26.551 again.
00:43:31.731 --> 00:43:32.511 Awesome. Thank you.
00:43:36.121 --> 00:43:36.621 You're here.
00:43:37.921 --> 00:43:39.571 Any other questions, feel free.
00:43:40.821 --> 00:43:41.631 Or comments.
00:43:44.001 --> 00:43:48.461 That is definitely really insightful and as I was sharing
00:43:48.461 --> 00:43:51.691 before, Steve not Stevia started talking.
00:43:53.011 --> 00:43:57.286 The some of the trainings on the marketing from the marketing
00:43:57.286 --> 00:44:01.285 team have been really granular. So if you're particularly
00:44:01.285 --> 00:44:05.490 inclined, we can definitely share those with the rest of the
00:44:05.490 --> 00:44:09.765 group. Feel free to just make a post into the chat and we can
00:44:09.765 --> 00:44:14.178 share those with the group. So I just want to say thanks again,
00:44:14.178 --> 00:44:17.901 Steve, excellent job that was really entertaining and
00:44:17.901 --> 00:44:21.141 insightful. Certainly looking forward to more.
00:44:21.781 --> 00:44:26.486 Insight as to how we can all continue to have value as AI
00:44:26.486 --> 00:44:30.541 continues to augment our roles but not take over.
00:44:32.041 --> 00:44:35.734 And just sort of on that topic, really excited for Jeremy, the
00:44:35.734 --> 00:44:39.603 Rose to present next and Jeremy, I'll let you kind of kick things
00:44:39.603 --> 00:44:43.061 off and just kind of set the context for where you want to
00:44:43.061 --> 00:44:46.754 start several really exciting topics that you'll be going into
00:44:46.754 --> 00:44:48.571 today. So over to you good Sir.
00:44:49.531 --> 00:44:53.162 Thanks, Chris. Yeah. So, I mean we've we've talked a lot about
00:44:53.162 --> 00:44:56.909 here different applications, but I think the the biggest thing I
00:44:56.909 --> 00:45:00.252 wanna challenge you guys to think about when we discussed
00:45:00.252 --> 00:45:04.056 this stuff is how can I use this tool like Brendan Lyon said, how
00:45:04.056 --> 00:45:07.111 can I use this tool in my day-to-day to make it more
00:45:07.111 --> 00:45:08.091 efficient faster?
00:45:09.691 --> 00:45:12.906 Do things better, right? The early adopters of most
00:45:12.906 --> 00:45:16.925 technology are usually the ones that succeed in the long run, so
00:45:16.925 --> 00:45:20.635 I think it's very important for us as a company. And thanks
00:45:20.635 --> 00:45:24.221 Chris, for giving us the opportunity to do that. But very
00:45:24.221 --> 00:45:28.117 important for us as a company to go into a deep dive of AI and
00:45:28.117 --> 00:45:28.611 GPT and.
00:45:29.411 --> 00:45:32.615 What it offers the company as well as what it can offer us as
00:45:32.615 --> 00:45:36.026 individual developers, you know, having this be an early adoption
00:45:36.026 --> 00:45:39.282 thing. Look at the people that were quick to adopt Facebook or
00:45:39.282 --> 00:45:42.538 influencers when they're quick to join TikTok, right, like the
00:45:42.538 --> 00:45:45.691 usually the first people to market are the ones that succeed
00:45:45.691 --> 00:45:48.533 in the long run. And I think that's one of the biggest
00:45:48.533 --> 00:45:51.686 advantages that we have as a company is that we're a smaller
00:45:51.686 --> 00:45:54.683 company. We're a group of incredibly talented individuals
00:45:54.683 --> 00:45:57.732 each in our own respective areas. And the ability of us to
00:45:57.732 --> 00:45:59.541 move quicker than our competition.
00:45:59.801 --> 00:46:00.981 Is really what sets us apart.
00:46:01.691 --> 00:46:05.351 So if we can get into using GPT and these AI tools and bring
00:46:05.351 --> 00:46:08.831 them into our products and bring them into our pipelines.
00:46:10.031 --> 00:46:13.014 We really stand to gain a huge advantage over our competition
00:46:13.014 --> 00:46:15.853 and you know some of the the people like Chris was leading
00:46:15.853 --> 00:46:18.691 off some statistics around companies are using AI I think.
00:46:19.621 --> 00:46:22.187 There's a big difference between using the ad that existed
00:46:22.187 --> 00:46:22.491 before.
00:46:23.221 --> 00:46:27.015 And using the newer models like GPT 3 and GPT 4. If you saw that
00:46:27.015 --> 00:46:30.633 timeline that Steve was sharing there, I can kind of share my
00:46:30.633 --> 00:46:34.369 screen here too. If you saw the timeline and the progression of
00:46:34.369 --> 00:46:36.061 how fast this stuff is going.
00:46:36.981 --> 00:46:42.971 I mean June 2019 GPT 2 came out June 2020, GT three came out.
00:46:44.381 --> 00:46:48.980 Dolly came out GBT itself ChatGPT as we know it came out
00:46:48.980 --> 00:46:54.305 in November of 2022 and that was using GPT 3 and now we're in the
00:46:54.305 --> 00:46:59.308 realm of GPT 4 which was, as he said, immensely more powerful
00:46:59.308 --> 00:47:03.261 than GPT 5. So that's between N22 and March 23A.
00:47:04.641 --> 00:47:08.334 It astronomical increase in the potential for AI and what it can
00:47:08.334 --> 00:47:11.458 do for us. So keep that in mind. It's a it's a rapidly
00:47:11.458 --> 00:47:14.867 progressing field. There's new stuff that comes out all the
00:47:14.867 --> 00:47:15.151 time.
00:47:15.861 --> 00:47:18.472 So it's something that you really want to stay on top of
00:47:18.472 --> 00:47:21.450 and the the more we do that as a collective here at the company,
00:47:21.450 --> 00:47:23.511 the more ways we think about ways to use it.
00:47:24.801 --> 00:47:27.384 That's really how we're going to go ahead and in the society
00:47:27.384 --> 00:47:28.951 going forward. I truly believe that.
00:47:29.851 --> 00:47:34.661 GPT and AI tools like GPT are going to be what separate?
00:47:35.521 --> 00:47:39.149 The competition in the in the future, you know, it's this is a
00:47:39.149 --> 00:47:42.719 transformational part of our society as far as the crossroads
00:47:42.719 --> 00:47:44.561 that we're at. So powerful that.
00:47:45.401 --> 00:47:49.582 There is an open letter that has almost 30,000 signatures to
00:47:49.582 --> 00:47:53.214 pause immediately for six months. The training of AI
00:47:53.214 --> 00:47:57.052 systems more powerful than T4, with some very prominent
00:47:57.052 --> 00:48:00.959 signatures on this list. Wozniak, Apple, Elon Musk, CEOs
00:48:00.959 --> 00:48:05.208 of major companies saying we need to stop and reevaluate what
00:48:05.208 --> 00:48:06.921 we're doing here because.
00:48:08.401 --> 00:48:11.519 We need to establish safety standards around it and and get
00:48:11.519 --> 00:48:14.480 some eyes on it because it's progressing so quickly that
00:48:14.480 --> 00:48:15.571 nothing, none of the.
00:48:16.641 --> 00:48:18.521 The regulations can keep up around it so.
00:48:19.661 --> 00:48:22.356 This is if this doesn't open your eyes as to what what's
00:48:22.356 --> 00:48:24.531 going on and how much more powerful GPT 4 is.
00:48:26.071 --> 00:48:29.359 Then I really think you you can't sleep on this, right? This
00:48:29.359 --> 00:48:32.862 is transformational. And as far as our jobs go, you know there's
00:48:32.862 --> 00:48:36.149 discussion around Will GPT take over our jobs? I don't think
00:48:36.149 --> 00:48:39.491 that that's accurate to say. I think that this statement that
00:48:39.491 --> 00:48:42.886 you know there will be jobs that get replaced or jobs that get
00:48:42.886 --> 00:48:44.341 changed fundamentally, but.
00:48:46.101 --> 00:48:49.657 I think it'll enable us just to write more code. You know, if
00:48:49.657 --> 00:48:53.041 we're using these tools to accomplish our day-to-day tasks
00:48:53.041 --> 00:48:56.596 a lot faster, what it really does for us is just allows us to
00:48:56.596 --> 00:48:59.751 develop more stuff. There's always features need to be
00:48:59.751 --> 00:49:03.479 developed, whether it's writing code for the front end, code for
00:49:03.479 --> 00:49:06.576 the back end or as we saw writing code for marketing,
00:49:06.576 --> 00:49:09.787 writing code for project management, all that stuff can
00:49:09.787 --> 00:49:13.515 be done if we have more time on our hands to write more code. So
00:49:13.515 --> 00:49:16.956 it's a huge tool. It's part of my day-to-day work. I use it
00:49:16.956 --> 00:49:17.931 every single day.
00:49:18.651 --> 00:49:21.853 If you're not consulting GPT alongside your standard
00:49:21.853 --> 00:49:25.297 Googling, I think I think you're really doing yourself a
00:49:25.297 --> 00:49:28.801 disservice. There's a lot of value to be gained as we saw
00:49:28.801 --> 00:49:32.486 it's not perfect, right? It's not without its flaws. It gets
00:49:32.486 --> 00:49:36.232 better every single day, but learning to use it now, learning
00:49:36.232 --> 00:49:39.071 how to interact with it, remembering that it's
00:49:39.071 --> 00:49:40.521 conversational and that.
00:49:41.291 --> 00:49:44.072 You know it's it learns. The more that you give it and then
00:49:44.072 --> 00:49:47.131 learning the in's and outs of it behind the scenes, which will go
00:49:47.131 --> 00:49:50.144 to here in a little minute. But learning those ends and outs and
00:49:50.144 --> 00:49:52.971 kind of seeing how to interact with how to use it as a tool.
00:49:54.221 --> 00:49:54.811 You know, I think.
00:49:56.681 --> 00:49:59.593 There was a time when people used to take horses. You know,
00:49:59.593 --> 00:50:02.553 horses were the primary mode of transportation and then cars
00:50:02.553 --> 00:50:05.659 came along, and now there's not so many horses anymore. So this
00:50:05.659 --> 00:50:07.891 is one of those moments in our society where.
00:50:08.951 --> 00:50:12.461 You know, they were. They were developers that didn't use AI
00:50:12.461 --> 00:50:15.971 and then they were developers that did use AI and eventually
00:50:15.971 --> 00:50:19.538 the ones that use AI will be the ones that stick around. So I
00:50:19.538 --> 00:50:23.220 truly believe that you're gonna wanna learn this now. Don't get
00:50:23.220 --> 00:50:24.601 left behind on this and.
00:50:25.671 --> 00:50:27.741 And learn to to make it part of your workflow.
00:50:28.971 --> 00:50:32.419 There's a huge efficiency gain there, so in that applies to not
00:50:32.419 --> 00:50:35.706 just developers, but everyone across the board. I'm building
00:50:35.706 --> 00:50:38.615 tools for our PMS. I'm building tools for our project
00:50:38.615 --> 00:50:41.956 management, the lifecycle level project using Mermaid syntax,
00:50:41.956 --> 00:50:44.919 stuff like that. There's a lot that goes into it, so I
00:50:44.919 --> 00:50:48.152 encourage you as we go through this and demonstrate kind of
00:50:48.152 --> 00:50:51.600 what I'm doing with it to think about how can I use it, how can
00:50:51.600 --> 00:50:53.701 I make this part of my day-to-day and.
00:50:55.901 --> 00:50:59.951 Where could I apply this in my? My current thing because.
00:51:00.571 --> 00:51:01.851 Whether that shapes us as.
00:51:02.541 --> 00:51:04.975 Developing new software and tools for what we're doing, or
00:51:04.975 --> 00:51:07.408 shapes us as a company to say you know what, that's a huge
00:51:07.408 --> 00:51:09.924 business opportunity. I think we should pursue that. I think
00:51:09.924 --> 00:51:12.605 those are all valid things. You know, Chris is spending a lot of
00:51:12.605 --> 00:51:14.791 money investing in AI investing. What we can do and.
00:51:15.711 --> 00:51:19.302 There's a lot of opportunity there. So watching Brent's
00:51:19.302 --> 00:51:21.931 presentation with the the AI thing here.
00:51:23.991 --> 00:51:24.911 And it got me thinking about.
00:51:25.971 --> 00:51:28.281 A couple different ways to use this right so.
00:51:29.351 --> 00:51:32.088 There's a potential down the road that our job becomes
00:51:32.088 --> 00:51:33.531 auditing what the AI outputs.
00:51:34.921 --> 00:51:37.474 And I think that that's that's a huge, hugely powerful thing to
00:51:37.474 --> 00:51:38.391 be like. You know what?
00:51:39.181 --> 00:51:41.946 We could. We could either replace the work of 1010
00:51:41.946 --> 00:51:45.036 developers using AI sometime down the road, let's say 10
00:51:45.036 --> 00:51:48.180 years from now, 20 years from now, that your job would be
00:51:48.180 --> 00:51:51.216 instead of a junior level developer, it would be having
00:51:51.216 --> 00:51:54.522 someone a mid level developer that reviews the code that the
00:51:54.522 --> 00:51:55.281 AI writes out.
00:51:55.951 --> 00:51:58.361 And makes adjustments a race code reviews that go from there.
00:51:59.921 --> 00:52:03.375 So there's a huge potential there that we become arbiters of
00:52:03.375 --> 00:52:06.941 the AI. So thinking about this and thinking about how we would
00:52:06.941 --> 00:52:10.338 prompt and talk to GPT, there's a couple things you need to
00:52:10.338 --> 00:52:13.961 know. There's GPT 35, which is the standard language model that
00:52:13.961 --> 00:52:17.528 people are used to working with when they go to chat, debt. If
00:52:17.528 --> 00:52:21.151 you have the paid subscription, you can get access to GPT 4. It
00:52:21.151 --> 00:52:24.774 is much, much better. It's going to be able to make images very
00:52:24.774 --> 00:52:28.171 soon. There are models that create images like mid journey.
00:52:29.421 --> 00:52:31.875 Standard diffusion or yeah, standard diffusion, so that
00:52:31.875 --> 00:52:32.751 something like that.
00:52:34.021 --> 00:52:37.696 That that make images, but GT4 will be able to parse images,
00:52:37.696 --> 00:52:40.949 make images and go all that stuff, so that's that's a
00:52:40.949 --> 00:52:41.371 hugely.
00:52:42.291 --> 00:52:45.896 Huge advantage in in a big leap forward in technology because
00:52:45.896 --> 00:52:47.931 this was just a language model so.
00:52:48.861 --> 00:52:51.274 Thinking about what Brent put in, you have the initial prompt
00:52:51.274 --> 00:52:53.141 though. Jeez, of course. I've been sitting too.
00:52:54.741 --> 00:52:58.421 Of course it fails twice right on the demo. Ah, come on.
00:53:02.471 --> 00:53:03.421 Really. Really.
00:53:06.001 --> 00:53:07.551 Nobody is safe from the demo demon.
00:53:07.051 --> 00:53:09.341 Perfect, perfect timing.
00:53:08.721 --> 00:53:12.351 Right. Perfect segue to be like, Yep, Yep.
00:53:10.221 --> 00:53:13.211 They paused it. Everybody listened to Elon.
00:53:13.941 --> 00:53:16.586 Yep, exactly. So, thinking about this, you can see the prompt
00:53:16.586 --> 00:53:19.061 that I gave. It was your secretary. I want you to walk me
00:53:19.061 --> 00:53:21.791 through the process of becoming patient. Scheduled meeting. Can
00:53:21.791 --> 00:53:24.138 you play me that look like? Please give me a chance to
00:53:24.138 --> 00:53:26.612 respond every prompt. And once the point looks scheduled,
00:53:26.612 --> 00:53:29.087 provide me a summary markdown format. So this gave me the
00:53:29.087 --> 00:53:31.647 whole script. Right. Which is not exactly what I asked for.
00:53:31.647 --> 00:53:33.481 And that's something to keep in mind that.
00:53:34.181 --> 00:53:37.243 There, there can be subtle differences in your prompt that
00:53:37.243 --> 00:53:40.513 change entirely the output. So what if we took the same prompt
00:53:40.513 --> 00:53:43.574 and we can kind of play with this little bit here and say,
00:53:43.574 --> 00:53:46.221 OK, you know what? Let's start a new chat. Really.
00:53:48.381 --> 00:53:50.011 It's our new chat and let's say.
00:53:54.381 --> 00:53:58.046 Remember, it's conversational, so we can we can tell it exactly
00:53:58.046 --> 00:53:59.821 what we want it to do. So only.
00:54:01.611 --> 00:54:02.031 Uh.
00:54:03.041 --> 00:54:04.731 Give me a chance to.
00:54:05.661 --> 00:54:11.622 Respond after every prompt with a new part of the conversation,
00:54:11.622 --> 00:54:13.671 one message at a time.
00:54:16.041 --> 00:54:18.688 OK, so see how that that fundamentally changed what what
00:54:18.688 --> 00:54:21.567 it came out with, right. So what I did was I specified twice.
00:54:21.567 --> 00:54:24.446 Give me a chance to respond. Give me a chance to respond. And
00:54:24.446 --> 00:54:27.000 it fundamentally changed now, I'd say, OK, this is now
00:54:27.000 --> 00:54:27.371 Brent's.
00:54:28.381 --> 00:54:33.341 Brent's whole thing. Right. So now you have my name. My name is
00:54:33.341 --> 00:54:35.511 Brent Fife and my number is.
00:54:39.671 --> 00:54:43.273 So now this is Brent's chat bot, right and all I did was change
00:54:43.273 --> 00:54:45.411 the prompt that I gave it and that's.
00:54:46.271 --> 00:54:49.401 That's what I mean when I say learn how to interact with with
00:54:49.401 --> 00:54:52.480 GPT, because if you were to do this and just say, well, that
00:54:52.480 --> 00:54:55.711 wasn't what I wanted, this tool sucks. You'd walk away from it.
00:54:55.711 --> 00:54:58.891 But knowing how it works is a huge factor in this whole thing.
00:54:59.781 --> 00:55:03.312 So this this is called prompt engineering and it really comes
00:55:03.312 --> 00:55:07.014 with a lot of experience and you can see how many prompts I send
00:55:07.014 --> 00:55:10.374 in to learn how to use this thing. It just comes with time
00:55:10.374 --> 00:55:14.132 in the saddle as we would say to learn how to do this right? Like
00:55:14.132 --> 00:55:17.834 how do I make this thing bend to my will and what does that look
00:55:17.834 --> 00:55:21.137 like? There's the standard interface here that we're used
00:55:21.137 --> 00:55:24.155 to seeing, which is chat.openai.com. And then if you
00:55:24.155 --> 00:55:27.743 go into the other side, this is the actual platform itself. So
00:55:27.743 --> 00:55:29.451 this is called the playground.
00:55:29.691 --> 00:55:33.266 And this is where Stephen you were talking about, how do we
00:55:33.266 --> 00:55:35.471 make it nondeterministic? Right, so.
00:55:36.181 --> 00:55:38.501 If you're not familiar with that term, what it means is that.
00:55:39.721 --> 00:55:42.680 This AI is going to give you a different answer every single
00:55:42.680 --> 00:55:45.396 time, right? Think of it as conversational, right? It's
00:55:45.396 --> 00:55:48.500 talking to a person. You're not gonna get the same answer every
00:55:48.500 --> 00:55:51.459 time, but if you go into the actual back end, or if you were
00:55:51.459 --> 00:55:53.641 in the API, you can control the temperature.
00:55:54.781 --> 00:55:57.703 The length of the tokens, the frequency penalty, the presence
00:55:57.703 --> 00:56:00.813 penalty, all this stuff which is slightly different than you were
00:56:00.813 --> 00:56:03.876 seeing on the actual interface, right? The actual user interface
00:56:03.876 --> 00:56:04.111 here.
00:56:04.911 --> 00:56:07.649 So if I turn down the temperature, you can see as it
00:56:07.649 --> 00:56:10.387 turns approach to zero, it becomes deterministic and
00:56:10.387 --> 00:56:13.487 repetitive, which really for programming, if we're building
00:56:13.487 --> 00:56:16.535 tools, that's what we need, right? If we wanna build tools
00:56:16.535 --> 00:56:19.635 around this and integrate this into our system, we need the
00:56:19.635 --> 00:56:22.683 temperature to be zero. We need it to be as repetitive and
00:56:22.683 --> 00:56:25.679 deterministic as possible. One way that I've found to get
00:56:25.679 --> 00:56:28.727 around that in some of the tooling that I'm creating is to
00:56:28.727 --> 00:56:31.878 tell it to respond in JSON, right. So if I get it to respond
00:56:31.878 --> 00:56:35.081 in JSON format, I can take this thing and say, you know what?
00:56:35.381 --> 00:56:38.796 Always give me the same kind of answer back, right? So the
00:56:38.796 --> 00:56:41.401 prompt that I would give it looks like this.
00:56:42.201 --> 00:56:44.682 You're an assistant to a project manager that only responds to
00:56:44.682 --> 00:56:47.242 Jason. Do not write normal text. It's imperative that you follow
00:56:47.242 --> 00:56:49.605 these instructions exactly. Never write normal text or code
00:56:49.605 --> 00:56:49.881 blocks.
00:56:52.581 --> 00:56:55.593 Can you create a mermaid chart and string field called mermaid
00:56:55.593 --> 00:56:58.749 chart that begins with the chart type for the following input? So
00:56:58.749 --> 00:57:01.666 this is a pre prompt that would follow all of my messages so
00:57:01.666 --> 00:57:02.861 that I would say can you.
00:57:04.021 --> 00:57:05.301 Create me a.
00:57:06.061 --> 00:57:08.571 Gantt chart for the software.
00:57:10.801 --> 00:57:17.181 Development life cycle of an ecommerce integration with an.
00:57:18.611 --> 00:57:20.371 ERP system.
00:57:22.621 --> 00:57:26.526 Some turn the temperature down. I'm gonna bump the tokens up so
00:57:26.526 --> 00:57:30.369 the tokens are the length and the memory of what you're doing,
00:57:30.369 --> 00:57:31.101 so when you.
00:57:32.141 --> 00:57:36.237 Pay for purchase an API access or a A A paid membership to GPT.
00:57:36.237 --> 00:57:40.461 You could access these tokens so you can send a number of tokens.
00:57:40.461 --> 00:57:44.300 I think the maximum limit for this is 2048 tokens, but that
00:57:44.300 --> 00:57:48.076 being said GPT 4 and the reason why it's so much better is
00:57:48.076 --> 00:57:52.236 something like 40,000 tokens you can send across. So having that
00:57:52.236 --> 00:57:55.948 memory right, that's the context. That's the memory limit
00:57:55.948 --> 00:57:59.787 of what you're doing. It helps it stay along with what it's
00:57:59.787 --> 00:58:02.091 doing. That's why GPT is so bad at.
00:58:02.171 --> 00:58:05.162 Counting and math. Remember, it's a language model, so it's
00:58:05.162 --> 00:58:08.202 just it's not great at that kind of stuff. So if I send this
00:58:08.202 --> 00:58:08.501 along.
00:58:10.101 --> 00:58:13.608 I get back Jason, which as a developer I mean this is huge
00:58:13.608 --> 00:58:17.353 right? So now I can parse this out and and do what I needed to
00:58:17.353 --> 00:58:17.531 do.
00:58:18.691 --> 00:58:22.241 And integrate this into a tool, right? So having the ability to.
00:58:23.051 --> 00:58:26.167 Give it nondeterministic characteristics like this. This
00:58:26.167 --> 00:58:29.555 would be huge, so I can take this and bring this into another
00:58:29.555 --> 00:58:32.944 tool now, and that's what I've been working on. My spare time
00:58:32.944 --> 00:58:34.201 is creating tools that.
00:58:35.021 --> 00:58:38.478 I can develop integrations with that will always respond in the
00:58:38.478 --> 00:58:42.043 same way. If you're not familiar with what Mermaid chart is. If I
00:58:42.043 --> 00:58:45.500 go here, you can see where this is. GPT itself will tell you if
00:58:45.500 --> 00:58:48.741 you tell ask it to make it a flow chart and you build. I'll
00:58:48.741 --> 00:58:50.361 put that same prompt in there.
00:58:52.571 --> 00:58:53.201 This one.
00:58:56.061 --> 00:58:57.451 We'll go here and we'll say.
00:58:58.151 --> 00:59:01.051 And we won't say Gantt chart will say workflow diagram.
00:59:04.431 --> 00:59:06.974 For software lifecycle of new commerce integration the system
00:59:06.974 --> 00:59:08.451 it's tell you it can't make images.
00:59:10.891 --> 00:59:12.781 In you create.
00:59:14.591 --> 00:59:16.781 An image of what that.
00:59:22.971 --> 00:59:26.840 And it can, right? So it failed, some would say. Can you create
00:59:25.561 --> 00:59:25.861 Can it?
00:59:26.840 --> 00:59:27.021 it?
00:59:28.131 --> 00:59:29.401 Using mermaid syntax.
00:59:30.811 --> 00:59:33.681 Or could it do it like a base 64 string of an image?
00:59:34.841 --> 00:59:37.814 That could have been, yeah, Yep. The the problem with that would
00:59:37.814 --> 00:59:40.695 be is it would take up so many tokens, right? Because the base
00:59:40.695 --> 00:59:42.661 64 of an image is absolutely massive. Yep.
00:59:41.071 --> 00:59:43.041 Yeah. Yeah, it's huge.
00:59:43.481 --> 00:59:47.401 Yep. So the advantage you have is, but it does know mermaid, so
00:59:47.401 --> 00:59:51.382 it can give me this that I could paste directly in and if you're
00:59:51.382 --> 00:59:55.425 not familiar with mermaid it can be displayed like this inside VS
00:59:55.425 --> 00:59:55.731 code.
00:59:56.521 --> 00:59:58.231 So if we were using markdown files.
00:59:59.011 --> 01:00:03.421 We could directly import that into our tooling so.
01:00:04.571 --> 01:00:08.430 Where my head went on this whole thing was to say, you know what?
01:00:08.430 --> 01:00:11.821 I bet I could create tooling for project managers or for.
01:00:13.541 --> 01:00:15.151 Developers that we would have.
01:00:15.951 --> 01:00:20.604 This integrated into our actual files, right? So I said OK,
01:00:20.604 --> 01:00:23.241 let's go here. I'll open this up.
01:00:24.621 --> 01:00:26.651 So what I created was on the CONNECT team.
01:00:27.851 --> 01:00:29.211 A document.
01:00:30.411 --> 01:00:31.041 Repo.
01:00:33.051 --> 01:00:34.231 Where we now have.
01:00:36.301 --> 01:00:37.571 All of these kind of things.
01:00:39.051 --> 01:00:42.979 In markdown format O, all of our documents are now contained in
01:00:42.979 --> 01:00:43.531 markdown.
01:00:44.311 --> 01:00:48.140 What that does for us is lets us use automation inside our tools,
01:00:48.140 --> 01:00:51.911 right? So if you were to create a docx file, a regular word doc,
01:00:51.911 --> 01:00:55.683 you wouldn't necessarily be able to use the formatting and stuff
01:00:55.683 --> 01:00:57.771 that you could use inside code. So.
01:00:58.521 --> 01:01:02.171 Now on the if I go to preview this.
01:01:03.381 --> 01:01:07.038 You can see that this is the CONNECT project lifecycle, so
01:01:07.038 --> 01:01:10.261 I've creating tooling through GPT that will output.
01:01:10.941 --> 01:01:15.017 Actual images inside a markdown file, so this is very early on
01:01:15.017 --> 01:01:18.898 in the whole process, but my idea is we could theoretically
01:01:18.898 --> 01:01:22.974 automate the entire process to say you know what, because this
01:01:22.974 --> 01:01:26.661 is a standardized format, we could create tickets off of
01:01:26.661 --> 01:01:30.931 this, right? If we had our Gantt charts were accurate and our PMS
01:01:30.931 --> 01:01:35.006 knew how to use Mermaid syntax, we could take this and develop
01:01:35.006 --> 01:01:38.111 tools and tickets off of it so that the actual.
01:01:38.781 --> 01:01:42.140 Act of requirements gathering from our customers and our
01:01:42.140 --> 01:01:42.611 clients.
01:01:43.321 --> 01:01:46.225 From the sales handoff all the way through, the developers are
01:01:46.225 --> 01:01:49.037 automated and there's a process behind it and we could build
01:01:49.037 --> 01:01:49.821 tools doing that.
01:01:51.081 --> 01:01:51.511 So.
01:01:52.571 --> 01:01:56.510 This is a small portion of what we could do, right. We have this
01:01:56.510 --> 01:01:57.661 markdown files for.
01:01:58.651 --> 01:02:01.712 All this so I created this sample meeting notes document
01:02:01.712 --> 01:02:05.041 right here so that I could come in and say I could copy paste
01:02:05.041 --> 01:02:08.155 this. So now that's our standard workflow for our project
01:02:08.155 --> 01:02:10.571 managers right now is to say, OK, we have a.
01:02:11.381 --> 01:02:15.261 A docx file. We can generate a copy of it. We can do that, but
01:02:15.261 --> 01:02:19.018 then my my head kind of went, you know, if this is all being
01:02:19.018 --> 01:02:22.898 generated through code, right? What I could do is I could then
01:02:22.898 --> 01:02:24.191 ask GPT to create me.
01:02:25.051 --> 01:02:30.969 Snippets right? So I could say, can you create me a sample
01:02:30.969 --> 01:02:35.081 template for a client escalation report?
01:02:36.101 --> 01:02:37.051 In software.
01:02:41.011 --> 01:02:44.001 OK, great. This is good. This is good. We could we could paste
01:02:44.001 --> 01:02:46.801 this in, we could make some adjustments to it. We could go
01:02:46.801 --> 01:02:49.600 that route, but I want to take this one step further and I
01:02:49.600 --> 01:02:51.261 would say, can you make this a VS?
01:02:52.681 --> 01:02:55.401 Code snippet to reuse.
01:02:57.901 --> 01:03:01.156 So now it's going to generate code and it's going to create me
01:03:01.156 --> 01:03:02.551 a client escalation report.
01:03:03.751 --> 01:03:07.179 That I could bring into my VS code and I could say ohh look I
01:03:07.179 --> 01:03:10.661 have several of these. I have a challenge solution analysis. I
01:03:10.661 --> 01:03:14.089 have a meeting notes, summary template. I'm now going to have
01:03:14.089 --> 01:03:15.581 a client escalation report.
01:03:16.781 --> 01:03:20.347 So with that I can type in escalation into a new file as
01:03:20.347 --> 01:03:21.911 long as I'm in this repo.
01:03:23.481 --> 01:03:24.511 And I have that right here.
01:03:25.711 --> 01:03:29.658 OK, that's useful. But what I did one step further was say OK,
01:03:29.658 --> 01:03:33.543 I have Csas on the CONNECT team. These are challenge solution
01:03:33.543 --> 01:03:36.926 analysis files right here to bring in. This is now in
01:03:36.926 --> 01:03:40.121 markdown format. So it would if I hadn't broke it.
01:03:42.061 --> 01:03:45.211 Alright, here it would bring this in using.
01:03:47.041 --> 01:03:47.691 My formatting.
01:03:50.391 --> 01:03:53.667 Which it couldn't do if I had a docx file right, it would not be
01:03:53.667 --> 01:03:56.842 able, GPT would not be able to create headers and all sorts of
01:03:56.842 --> 01:04:00.067 stuff, but markdown Ken so I can go back to my GPT prompt and I
01:04:00.067 --> 01:04:00.571 could say.
01:04:01.671 --> 01:04:03.311 I'm going to change this one around.
01:04:05.181 --> 01:04:08.311 And I'll go and I'll say, can you do it in markdown?
01:04:12.661 --> 01:04:15.405 And it's going to me them back to markdown. So these hashes, if
01:04:15.405 --> 01:04:17.591 you're not familiar markdown, this is going to be.
01:04:18.971 --> 01:04:22.291 The headers and stuff like that, so kind of my my biggest push on
01:04:22.291 --> 01:04:25.309 the CONNECT team is to change things around from being docx
01:04:25.309 --> 01:04:28.227 files to markdown files which would then open us up to an
01:04:28.227 --> 01:04:31.245 entire suite of tools that we can use around that. So now I
01:04:31.245 --> 01:04:32.301 have a markdown file.
01:04:33.621 --> 01:04:35.491 I hear and see. Now it has a header.
01:04:36.631 --> 01:04:39.734 And I could go one step further and even say make it so that I
01:04:39.734 --> 01:04:42.541 can put these in and it automates that, which I've done.
01:04:43.601 --> 01:04:44.511 By type in meeting.
01:04:46.871 --> 01:04:48.451 Apps can spell.
01:04:49.331 --> 01:04:52.361 Are you see there that I have the template. If I control
01:04:52.361 --> 01:04:55.710 space, this guy bring it in. You see? Now it gives me this has
01:04:55.710 --> 01:04:59.058 got this highlighted and I could type Winco and I could create
01:04:59.058 --> 01:05:02.513 meeting ended at 4:31 PM and all I did there was tab across this
01:05:02.513 --> 01:05:04.161 field. So what that looks like.
01:05:05.721 --> 01:05:06.891 If I preview this.
01:05:07.891 --> 01:05:12.016 You can see WINCO meeting ended at 4:31 PM and that's all using
01:05:12.016 --> 01:05:15.818 this snippet. So I have a meeting and these variables that
01:05:15.818 --> 01:05:19.749 it puts in it automatically put in the current error current
01:05:19.749 --> 01:05:24.002 month, current date created. All these different variables that I
01:05:24.002 --> 01:05:26.451 can tab across to fill things out so.
01:05:28.121 --> 01:05:30.995 This in and of itself is somewhat handy, right? This is a
01:05:30.995 --> 01:05:34.216 step in the right direction, but the cooler part is because it's
01:05:34.216 --> 01:05:37.091 all done in code because it's all done in a format that's
01:05:37.091 --> 01:05:40.114 consistent. I could automate these, right? And that's what I
01:05:40.114 --> 01:05:43.335 wanna challenge you guys all to think about. How can we take our
01:05:43.335 --> 01:05:46.556 current systems and our current business practices and processes
01:05:46.556 --> 01:05:49.430 and get them to where we can automate them so that we can
01:05:49.430 --> 01:05:52.403 scale, right? Because that's where my head always goes. How
01:05:52.403 --> 01:05:55.327 can we scale this thing? So think about the current volume
01:05:55.327 --> 01:05:58.201 of projects that we have here at clarity and think about.
01:05:58.831 --> 01:06:02.025 What would it take to 10X that? What would it take to 100 X
01:06:02.025 --> 01:06:05.272 that? Would we just need to bring in more people or is there
01:06:05.272 --> 01:06:08.466 a way we could streamline the whole process like similar to
01:06:08.466 --> 01:06:09.211 this to where?
01:06:10.201 --> 01:06:13.910 Now, things that we used to do that took us 1520 minutes, 30
01:06:13.910 --> 01:06:17.619 minutes are taking us five or taking US10 right. And I think
01:06:17.619 --> 01:06:20.781 that's really the power of GPT and the power of AI.
01:06:22.061 --> 01:06:25.466 In in building tools around it, so not just the actual code that
01:06:25.466 --> 01:06:28.871 you guys are writing, but think about what are some tasks that I
01:06:28.871 --> 01:06:32.119 do every single day that are repetitive that take a long time
01:06:32.119 --> 01:06:32.381 that.
01:06:33.271 --> 01:06:37.008 You know, just kind of mindless tasks that AI would be perfect
01:06:37.008 --> 01:06:40.685 for, you know, and it's really going to be that creativity is
01:06:40.685 --> 01:06:44.185 really what sets us apart when it comes to using AI tools,
01:06:44.185 --> 01:06:47.091 right? Like AI itself is not very creative, but.
01:06:47.751 --> 01:06:50.336 We are and that's how we can leverage these things to do it.
01:06:50.336 --> 01:06:51.141 We need them to do.
01:06:51.761 --> 01:06:54.428 Umm, so that's the project management side. I think that
01:06:54.428 --> 01:06:56.954 there's a huge opportunity there. There's very little
01:06:56.954 --> 01:06:59.668 tooling that that exists around that, whether it's in our
01:06:59.668 --> 01:07:00.931 company or even in general.
01:07:02.471 --> 01:07:05.978 Just having tooling to support project management and and ways
01:07:05.978 --> 01:07:09.317 to bring things across the pipeline. If you think about the
01:07:09.317 --> 01:07:12.823 sales handoff and the discovery process, my kind of dream is a
01:07:12.823 --> 01:07:13.491 world where.
01:07:14.231 --> 01:07:17.564 Just like we saw with this, a client could lead themselves
01:07:17.564 --> 01:07:21.065 through discovery, right? Is there any reason why we couldn't
01:07:21.065 --> 01:07:24.511 take the output of what this is, put it into a markdown file
01:07:24.511 --> 01:07:27.561 where the formatting is consistent and just pass that
01:07:27.561 --> 01:07:30.837 all the way down the pipeline and create can't charts and
01:07:30.837 --> 01:07:33.491 create workflow diagrams and everything right?
01:07:35.051 --> 01:07:37.836 So I would encourage you all to think about things like that
01:07:37.836 --> 01:07:40.483 because again, like I said before, if we aren't doing it,
01:07:40.483 --> 01:07:43.405 our competition might right, you know, and how do you get ahead
01:07:43.405 --> 01:07:46.098 of someone if they're doing those things and we aren't and
01:07:46.098 --> 01:07:48.061 they can get things done that much faster.
01:07:50.361 --> 01:07:50.801 So.
01:07:50.501 --> 01:07:51.021 Here here.
01:07:51.931 --> 01:07:52.331 Yeah.
01:07:53.141 --> 01:07:54.211 So with that instead.
01:07:54.861 --> 01:07:57.299 You know, with the project management side, I want to talk
01:07:57.299 --> 01:07:58.911 about ways that I'm using this inside.
01:08:00.811 --> 01:08:03.935 Actual C code, right? So for the developers out there, it is very
01:08:03.935 --> 01:08:06.870 good at writing SQL queries. So of course it can't load these
01:08:06.870 --> 01:08:07.201 really.
01:08:08.701 --> 01:08:10.131 It's very good at writing SQL queries.
01:08:12.051 --> 01:08:14.951 If you can get them to load, I'll have to just go through
01:08:14.951 --> 01:08:17.701 these and find them. Really. That's that's exhausting.
01:08:18.511 --> 01:08:22.328 Umm, but you know, it came up on the Connect team recently to say
01:08:22.328 --> 01:08:26.029 you know we take these options files on the CONNECT team. We've
01:08:26.029 --> 01:08:27.821 built out an automated tool to.
01:08:28.771 --> 01:08:31.601 Build a example settings.
01:08:32.981 --> 01:08:35.491 Based off of what's in our our Jason, so.
01:08:36.301 --> 01:08:37.081 Let's say I have.
01:08:37.961 --> 01:08:42.020 This woo commerce options file right? This this contains a URL,
01:08:42.020 --> 01:08:46.015 a consumer key, consumer secret, the auth type. These are both
01:08:46.015 --> 01:08:48.361 enums I could come into GPT and say.
01:08:49.481 --> 01:08:50.561 I have this options file.
01:08:54.421 --> 01:08:55.291 Can you build me?
01:08:57.161 --> 01:08:57.791 Sample.
01:08:58.451 --> 01:09:00.051 Jason file or?
01:09:00.791 --> 01:09:02.091 This representation.
01:09:04.911 --> 01:09:07.776 It says, Oh yeah, sure, I can do that. Here's an example. Jason
01:09:07.776 --> 01:09:10.595 file. OK, perfect. That's one one step in the right direction.
01:09:10.595 --> 01:09:11.311 Now our tooling.
01:09:12.511 --> 01:09:16.560 Wouldn't wouldn't necessarily by itself build this out, but I
01:09:16.560 --> 01:09:17.801 could say. Can you?
01:09:19.111 --> 01:09:20.161 Jason schema.
01:09:20.981 --> 01:09:22.951 Or what that would look like.
01:09:25.351 --> 01:09:28.739 Oops, I meant to bring that in because this has context, right?
01:09:28.739 --> 01:09:32.021 So it says, yeah, I can build you a schema. So now using this
01:09:32.021 --> 01:09:35.038 schema we can auto generate an app settings file for our
01:09:35.038 --> 01:09:38.319 projects. We can concatenate them all together and do it that
01:09:38.319 --> 01:09:38.531 way.
01:09:39.171 --> 01:09:43.771 The other cool part about this is can you descriptions?
01:09:44.961 --> 01:09:45.531 Or.
01:09:46.721 --> 01:09:47.371 What each?
01:09:49.031 --> 01:09:51.091 Those might be might represent.
01:09:53.311 --> 01:09:56.243 This is sure I can add description so now it says it
01:09:56.243 --> 01:09:56.741 gives me.
01:09:58.421 --> 01:10:02.313 Actual descriptions that come up when you go to create your JSON
01:10:02.313 --> 01:10:06.085 file. So again, something small right? Something very small we
01:10:06.085 --> 01:10:08.001 could have written this by hand.
01:10:08.681 --> 01:10:12.331 But the power that this this can create this this quickly. And
01:10:12.331 --> 01:10:15.111 now what I've become is I am now an arbiter of.
01:10:16.151 --> 01:10:19.453 Yeah, I right. I go back through this. I just verify that these
01:10:19.453 --> 01:10:22.704 look good and I say, you know what that works for me. Sign off
01:10:22.704 --> 01:10:24.561 on it and move on to the next task.
01:10:25.351 --> 01:10:29.079 For some of these options, this is quick, right? This would have
01:10:29.079 --> 01:10:32.750 taken me that long to write, but to do this across 20 projects,
01:10:32.750 --> 01:10:36.191 30 connectors, 100 connectors with 100 different fields and
01:10:36.191 --> 01:10:39.861 properties. Those are the things I want you to think about. How
01:10:39.861 --> 01:10:43.303 can we scale this business and how can we scale these tools
01:10:43.303 --> 01:10:46.801 out? What would it take to make me 10 times more productive?
01:10:47.721 --> 01:10:51.164 And you know these really repetitive tasks. Super super
01:10:51.164 --> 01:10:51.471 easy.
01:10:53.411 --> 01:10:55.491 Some of the other things that I'm working on.
01:10:57.391 --> 01:11:01.471 Here is a workflow diagram for that we created. This was pulled
01:11:01.471 --> 01:11:05.105 straight out of a meeting minutes that Clint had created
01:11:05.105 --> 01:11:09.121 for WINCO. Right, and I said, you know what? This is good. How
01:11:09.121 --> 01:11:12.819 can I do this to where it creates me? I remember my chart
01:11:12.819 --> 01:11:16.644 for that. Well, there's the mermaid chart. Right. There's a
01:11:16.644 --> 01:11:19.321 visual representation of what's going on.
01:11:20.891 --> 01:11:22.331 This is sometimes hard.
01:11:23.191 --> 01:11:23.981 To see.
01:11:26.101 --> 01:11:29.627 For a client, right? These notes very quickly created. Sometimes
01:11:29.627 --> 01:11:32.882 that's hard to understand, but a workflow diagram very much
01:11:32.882 --> 01:11:36.353 easier to understand. You guys have done the mermaid or the the
01:11:36.353 --> 01:11:39.771 mural boards. Now a lot of you guys doing the architecture for
01:11:39.771 --> 01:11:42.917 Miro boards. Think about the time that it takes to create
01:11:42.917 --> 01:11:46.280 those and now think about what would it be if we took the the
01:11:46.280 --> 01:11:49.480 sales handoff directly from the client and said build me a
01:11:49.480 --> 01:11:52.952 workflow diagram using mermaid syntax that we could then import
01:11:52.952 --> 01:11:53.711 into our tool.
01:11:54.931 --> 01:11:57.661 And create notes and stuff around this right? And then
01:11:57.661 --> 01:12:00.837 could I not create tickets off of these as well? Thinking about
01:12:00.837 --> 01:12:03.418 becoming a developer, would this not be spelled out
01:12:03.418 --> 01:12:06.446 automatically, right? If we ran this through our tooling and
01:12:06.446 --> 01:12:08.381 said these are the pieces that I need?
01:12:10.221 --> 01:12:13.051 So certainly something to think about as we go forward.
01:12:14.291 --> 01:12:17.451 With that being said, I create a little bit of a tool here.
01:12:18.321 --> 01:12:19.251 So I said.
01:12:21.251 --> 01:12:25.394 Can you create a Gantt chart using roommate syntax for SLC
01:12:25.394 --> 01:12:29.608 now? This is hooked up to GPTC API, so I could say. Can you
01:12:29.608 --> 01:12:30.731 build me a Gant?
01:12:33.201 --> 01:12:35.431 For the open for a.
01:12:36.691 --> 01:12:39.741 Software project that integrates.
01:12:40.731 --> 01:12:41.841 Magento.
01:12:42.671 --> 01:12:46.161 Sierra e-commerce Swiss.
01:12:47.631 --> 01:12:50.231 Sage 100 ERP.
01:12:53.071 --> 01:12:57.070 So that's the message. I'll send it across. This is going to call
01:12:57.070 --> 01:12:58.221 out to Openais API.
01:12:59.091 --> 01:13:01.871 And come back with a string inside this.
01:13:03.351 --> 01:13:06.996 The the Jason object that I gave it right and so now it can
01:13:06.996 --> 01:13:10.701 create me an actual workflow diagram or Gantt chart for this
01:13:10.701 --> 01:13:14.346 project. Now is this useful in and of itself? No right. But
01:13:14.346 --> 01:13:15.621 it's the idea behind.
01:13:16.781 --> 01:13:19.515 What can I do with this tool right? How can I use AI and and
01:13:19.515 --> 01:13:22.160 if I could do this, I wanna challenge each and everyone of
01:13:22.160 --> 01:13:25.073 you to think what else can we do with it, right? How can we take
01:13:25.073 --> 01:13:27.942 this and then move this along to the next tool and integrate AI
01:13:27.942 --> 01:13:30.631 with that thing and then move it along to the next tool and
01:13:30.631 --> 01:13:31.931 integrate it with that thing?
01:13:32.311 --> 01:13:34.731 Umm, So what I've done is I've used.
01:13:35.201 --> 01:13:36.901 And the.
01:13:38.901 --> 01:13:42.730 The prompting to say you know what? How could we get GPT to
01:13:42.730 --> 01:13:46.559 lead our clients through this? Right, since it goes back to
01:13:46.559 --> 01:13:50.579 that IVR that Brent was showing too, let me stop sharing for a
01:13:50.579 --> 01:13:53.131 second and reshare with computer audio.
01:13:57.221 --> 01:13:59.331 So I've got this and I would say you know what?
01:14:00.241 --> 01:14:01.781 Could we take this?
01:14:02.521 --> 01:14:03.371 And say.
01:14:06.311 --> 01:14:10.621 Welcome to clarity connect. My name is.
01:14:12.761 --> 01:14:17.462 Please and I'll be your assistant today as we lead
01:14:17.462 --> 01:14:19.121 through discovery.
01:14:24.551 --> 01:14:26.481 And so this is going to call out to.
01:14:27.691 --> 01:14:31.246 The 11 last clarity connect my name is Clarice and I'll be your
01:14:31.246 --> 01:14:33.801 assistant today as we lead through discovery.
01:14:35.251 --> 01:14:38.809 So think about the power of that. Now I have an API called
01:14:38.809 --> 01:14:42.367 that goes to 11 labs that then does text to speech using a
01:14:42.367 --> 01:14:46.166 custom voice that I created for clarity Connect. Could we then
01:14:46.166 --> 01:14:47.131 hook this up to?
01:14:47.931 --> 01:14:50.961 Our other service, which then would prompt the user for hi my
01:14:50.961 --> 01:14:54.039 name is whatever and it would say what systems are you looking
01:14:54.039 --> 01:14:54.821 integrate today.
01:15:04.271 --> 01:15:06.880 So I would make the the API call. What systems are you
01:15:06.880 --> 01:15:08.161 looking to integrate today?
01:15:09.421 --> 01:15:12.800 So you can imagine the power of this, right? It's not. It's not
01:15:12.800 --> 01:15:15.863 that it's going to replace people's jobs. But if this all
01:15:15.863 --> 01:15:19.295 got parsed out and brought back to our end right now, the client
01:15:19.295 --> 01:15:22.357 has LED themselves through discovery and we've asked them
01:15:22.357 --> 01:15:25.525 what pieces are most important to you. Do we not have a lot
01:15:25.525 --> 01:15:26.581 more context around?
01:15:27.571 --> 01:15:31.333 What the client was looking for, especially if we could train
01:15:31.333 --> 01:15:34.367 this AI system on on the specifics of what is the
01:15:34.367 --> 01:15:38.310 workflow for a sales order from SAGE 100 to big commerce or what
01:15:38.310 --> 01:15:42.072 would the workflow look like for the cyber A be to be able to
01:15:42.072 --> 01:15:45.591 register on a website and manage their memberships right?
01:15:46.511 --> 01:15:49.623 So that being said, I'll pause here. I'll look for and you guys
01:15:49.623 --> 01:15:52.687 have any questions or anything on this that we've gone over so
01:15:52.687 --> 01:15:53.951 far. I know there's a lot.
01:16:02.641 --> 01:16:03.441 A lot. A lot.
01:16:05.711 --> 01:16:10.111 I wanna I wanna just get you to talk about fine tuning.
01:16:11.831 --> 01:16:15.101 And I'll just set it up for a second and feel free if anyone
01:16:15.101 --> 01:16:18.157 else has any comments or questions. Just raise your hand
01:16:18.157 --> 01:16:21.481 so I can see. But one of the things that's really interesting
01:16:21.481 --> 01:16:22.231 about this is.
01:16:22.961 --> 01:16:28.622 With ChatGPT, not just ChatGPT, other large language models as
01:16:28.622 --> 01:16:29.071 well.
01:16:29.831 --> 01:16:33.971 Umm, probably one of the most sort of sophisticated
01:16:33.971 --> 01:16:39.067 progressions you can make from the off the shelf is to put your
01:16:39.067 --> 01:16:43.923 own content into it and teach it how you want the outputs to
01:16:43.923 --> 01:16:44.321 look.
01:16:45.511 --> 01:16:49.023 Given a certain set of inputs, this is what I want the outputs
01:16:49.023 --> 01:16:52.592 to be. Example, If I'm working on mermaid charts and we've done
01:16:52.592 --> 01:16:56.104 like 100 of them, you know six months from now and we're like,
01:16:56.104 --> 01:16:58.501 we really want them all to look like this.
01:16:58.791 --> 01:17:02.148 Uh, you know, it's kind of a pain to have to, like, massage
01:17:02.148 --> 01:17:02.931 it afterwards.
01:17:04.131 --> 01:17:08.316 We can teach it, and that's called again fine tuning, and
01:17:08.316 --> 01:17:12.717 it's also possible with our software to like, take chunks of
01:17:12.717 --> 01:17:17.479 our code and say, look, ChatGPT. This is, this is how we want our
01:17:17.479 --> 01:17:19.211 T fours to be generated.
01:17:19.671 --> 01:17:20.061 Umm.
01:17:20.191 --> 01:17:24.129 This is how we want uh payment provider generated shipping
01:17:24.129 --> 01:17:28.067 provider whatever a react component blah blah blah you get
01:17:28.067 --> 01:17:31.871 the idea. We basically inject those into ChatGPT in this
01:17:31.871 --> 01:17:36.009 example. Eventually it will be other open source LLNS that we
01:17:36.009 --> 01:17:38.011 can use internally right? But.
01:17:39.631 --> 01:17:43.281 Ultimately, we can teach it and train it.
01:17:44.231 --> 01:17:48.181 And it's very easy to do, or I mean at least simple.
01:17:48.951 --> 01:17:52.142 I mean it's basically easy because the hard part is just
01:17:52.142 --> 01:17:54.941 getting the data clean and ready to push into it.
01:17:56.691 --> 01:17:59.891 One of the things that I would just encourage everyone to think
01:17:59.891 --> 01:18:01.841 about with AI is it's not complicated.
01:18:03.621 --> 01:18:07.967 It's very capable of a lot of things. Whenever it's done at
01:18:07.967 --> 01:18:08.401 scale.
01:18:09.741 --> 01:18:14.633 But it's in it. At its core. It's a large it. It ends up
01:18:14.633 --> 01:18:19.181 being like a large curve fit of a giant set of data.
01:18:19.901 --> 01:18:22.111 Like, that's mechanically what AI is.
01:18:22.811 --> 01:18:26.465 Umm. And anyways, if you remember doing like the
01:18:26.465 --> 01:18:27.881 quadratic equation.
01:18:28.981 --> 01:18:32.043 And how that all works like that's basically what you're
01:18:32.043 --> 01:18:33.601 doing with matrices at scale.
01:18:34.891 --> 01:18:39.082 Anyways, the point is like whenever you put in your own
01:18:39.082 --> 01:18:43.872 stuff and tune it, it can make it really powerful for practical
01:18:43.872 --> 01:18:44.321 usage.
01:18:45.131 --> 01:18:49.775 And so I would be interested in your thoughts on that rose, just
01:18:49.775 --> 01:18:54.062 kind of like working within cause I I see several folks are
01:18:54.062 --> 01:18:58.420 pretty interested in this like making it an action item from
01:18:58.420 --> 01:19:02.564 this meeting to sort of get subsets of this group working
01:19:02.564 --> 01:19:07.136 over the next couple of months to sort of play around with fine
01:19:07.136 --> 01:19:11.137 tuning, like taking chunks of our code for example with
01:19:11.137 --> 01:19:15.281 Kinect, maybe we can have it come up with an auto mapper.
01:19:15.841 --> 01:19:17.381 You know, using AI.
01:19:17.881 --> 01:19:21.671 Umm. And then like for some of the Seth Code, maybe it can
01:19:21.671 --> 01:19:25.719 learn enough that it can make providers on its own, et cetera.
01:19:25.719 --> 01:19:28.031 So over to you with those thoughts.
01:19:28.841 --> 01:19:31.962 Yeah. No, it's a great point. You know, a lot of what we write
01:19:31.962 --> 01:19:33.051 on a day-to-day basis.
01:19:33.371 --> 01:19:36.757 Umm is repetitive. You know there's there are some aspects
01:19:36.757 --> 01:19:40.314 of code that are novel, but a lot of it is boilerplate. A lot
01:19:40.314 --> 01:19:43.757 of it's repetitive. I know on the CONNECT team, if we could
01:19:43.757 --> 01:19:47.142 train an AI model to be like, you know what this is how we
01:19:47.142 --> 01:19:50.642 create a standard workflow for a product sync. We have these
01:19:50.642 --> 01:19:54.315 methods appropriately named to where it wouldn't be anything to
01:19:54.315 --> 01:19:57.528 be able to tell an AI model generate me this file given
01:19:57.528 --> 01:20:01.028 these two systems because this flow is very standard, get an
01:20:01.028 --> 01:20:04.241 order, get a customer from a customer, push a customer.
01:20:04.561 --> 01:20:07.623 Push in order. You know this is a lot of what we do is
01:20:07.623 --> 01:20:11.075 repetitive like that, especially if our codes developed well,
01:20:11.075 --> 01:20:12.411 right? And that's where.
01:20:13.371 --> 01:20:16.399 You know, I've I've been pushing a lot in the CONNECT team,
01:20:16.399 --> 01:20:19.426 especially to standardize what we're doing because if it if
01:20:19.426 --> 01:20:22.403 it's standardized, you can automate it. So standardization
01:20:22.403 --> 01:20:23.311 is automation, so.
01:20:24.031 --> 01:20:25.381 Having those things be.
01:20:26.411 --> 01:20:30.329 Easily available to recreate, whether it's a payment provider
01:20:30.329 --> 01:20:34.121 inside CEF like you said, or an integration to another ERP.
01:20:35.341 --> 01:20:38.546 While the technical details might be different if we've
01:20:38.546 --> 01:20:41.693 written our code well, the interface that we work with
01:20:41.693 --> 01:20:45.070 those is very similar. So there's a there's a time not far
01:20:45.070 --> 01:20:47.931 off if we can train and fine tune our own models.
01:20:49.411 --> 01:20:52.331 That we could just generate these on the fly. And I mean,
01:20:52.331 --> 01:20:55.351 think about the power of being able to onboard a new client
01:20:55.351 --> 01:20:58.623 with standard workflows or even some complicated ones that we've
01:20:58.623 --> 01:21:01.442 taught our models, whether how to resolve the customer,
01:21:01.442 --> 01:21:04.714 anything like that with a click of a button that's kind of where
01:21:04.714 --> 01:21:05.821 we're going to end up.
01:21:13.011 --> 01:21:13.241 Yep.
01:21:34.641 --> 01:21:35.211 Exactly.
01:21:35.811 --> 01:21:40.426 And then it will be able to sort of like Intuit out the variables
01:21:40.426 --> 01:21:44.131 like the client name, the hours and stuff like that.
01:21:45.751 --> 01:21:49.797 But yeah, I think it would be really interesting for folks who
01:21:49.797 --> 01:21:53.651 are and I'm seeing pings on the side. So thank you guys for
01:21:53.651 --> 01:21:57.697 pinging. If you're interested in participating in some of this
01:21:57.697 --> 01:22:01.422 activity, it's just really stimulating for some people to
01:22:01.422 --> 01:22:05.532 press on and then we can circle back to the group and just kind
01:22:05.532 --> 01:22:09.386 of share what we found, you know, in a couple of months. So
01:22:09.386 --> 01:22:10.991 everyone gets the update.
01:22:11.821 --> 01:22:14.667 But keep going. I just wanted to sort of talk about fine tuning a
01:22:14.667 --> 01:22:15.141 little bit.
01:22:16.231 --> 01:22:16.781 Thank you, Sir.
01:22:16.611 --> 01:22:20.798 Yeah. Yeah. So that's that's a lot of GPT itself. There are
01:22:20.798 --> 01:22:24.705 other tools that are being developed, auto, GPT, if you
01:22:24.705 --> 01:22:29.102 guys haven't heard of this one, this is an open source outside
01:22:29.102 --> 01:22:33.358 of ChatGPT that relies on, I think some of their API, but it
01:22:33.358 --> 01:22:35.521 has access to the Internet, so.
01:22:37.561 --> 01:22:39.881 Some of the stuff that this is capable of doing.
01:22:40.981 --> 01:22:44.158 Pretty incredible Auto GT is trying to create an app it
01:22:44.158 --> 01:22:47.562 recognizes didn't have good node. It googled how to install
01:22:47.562 --> 01:22:50.910 node, found a stack overflow article downloaded, extracted
01:22:50.910 --> 01:22:52.271 and spawn the server so.
01:22:53.591 --> 01:22:54.691 Auto GPT.
01:22:55.471 --> 01:22:57.331 That that can access the Internet.
01:22:58.531 --> 01:23:01.546 And write its own code. I mean you can see why they said we
01:23:01.546 --> 01:23:04.461 wanna pause for a little bit here because the ability for
01:23:04.461 --> 01:23:06.974 these things to go on the Internet and do its own
01:23:06.974 --> 01:23:08.381 research, this is AI agents.
01:23:10.111 --> 01:23:13.971 A little bit differently than than what ChatGPT is, so.
01:23:14.701 --> 01:23:16.691 You know, there's there's a lot out there.
01:23:19.381 --> 01:23:23.107 A lot out there, and as far as what's going on in the world,
01:23:23.107 --> 01:23:25.611 it's pretty difficult to stay on top of.
01:23:26.471 --> 01:23:26.871 But.
01:23:28.151 --> 01:23:29.191 Yeah, I that's.
01:23:30.261 --> 01:23:33.339 Mostly you know more about the the fine tuning stuff than I do,
01:23:33.339 --> 01:23:34.301 Chris, to be honest.
01:23:35.481 --> 01:23:36.881 I think you've been playing with that piece of it.
01:23:37.661 --> 01:23:40.968 Yeah, it's it's really interesting because just like
01:23:40.968 --> 01:23:44.837 you made a an application flow AI, I think is the name of the
01:23:44.837 --> 01:23:48.331 name of the application you will, which is really cool.
01:23:46.981 --> 01:23:47.261 Yep.
01:23:48.801 --> 01:23:49.031 Yep.
01:23:49.881 --> 01:23:56.197 I went to ChatGPT one night a couple months ago and asked it
01:23:56.197 --> 01:24:00.131 to build me a C# console application.
01:24:02.021 --> 01:24:05.021 And WPF application after that.
01:24:06.311 --> 01:24:09.691 That would let me input some sample data.
01:24:10.641 --> 01:24:13.761 That would sort of like emulate the open AI.
01:24:14.301 --> 01:24:18.211 Umm, playground. But let me do fine tuning on it.
01:24:19.081 --> 01:24:20.871 And so it lets me upload.
01:24:22.391 --> 01:24:26.918 Jason BLOB that I can use to put whatever in it, so this can be
01:24:26.918 --> 01:24:30.101 like I said, it could be like new cleaner O.
01:24:30.871 --> 01:24:34.821 Or like the PM section of nuclear? No, that has like best
01:24:34.821 --> 01:24:39.043 practices. So I could inject that and then upload it and then
01:24:35.031 --> 01:24:35.321 Yep.
01:24:39.043 --> 01:24:43.265 prompt it and say like, hey, give me a standard intro e-mail.
01:24:43.265 --> 01:24:47.419 Give me a standard blah blah blah. But with this client name
01:24:47.419 --> 01:24:51.437 and these are the primary contacts and like obviously this
01:24:51.437 --> 01:24:55.523 stuff isn't really going to be practical in like every area
01:24:55.523 --> 01:24:57.361 that we're using right now.
01:24:58.831 --> 01:25:02.892 But if we kind of experiment with it, we can look where we
01:25:02.892 --> 01:25:07.090 have like high frequency of just repetitive tasks that could
01:25:07.090 --> 01:25:11.219 benefit. So you know anyone who's sort of like in the weeds
01:25:11.219 --> 01:25:15.624 on this fine tuning stuff, you would probably just need to know
01:25:15.624 --> 01:25:16.931 like it's a it's a.
01:25:18.241 --> 01:25:21.491 It's going to be out of love for learning.
01:25:22.731 --> 01:25:26.971 And experimenting, not necessarily immediate payoff.
01:25:24.491 --> 01:25:24.831 Yeah.
01:25:28.171 --> 01:25:32.783 But I I do think that like right away like today, a lot of the
01:25:32.783 --> 01:25:37.468 team like for example on the PM team, you really probably could
01:25:37.468 --> 01:25:40.981 use some of like the summary of transcriptions.
01:25:41.951 --> 01:25:45.401 Doesn't have to be ChatGPT, but like taking the transcriptions
01:25:45.401 --> 01:25:48.631 and getting a summary from that. That's a great example of
01:25:48.631 --> 01:25:51.971 something that would be really practical that you could use.
01:25:53.501 --> 01:25:57.511 And a tool like the hyper right might be more practical for PM's
01:25:57.511 --> 01:25:58.251 to consider.
01:25:58.981 --> 01:26:03.110 Umm, there's sort of has an interface that's a little bit
01:26:03.110 --> 01:26:07.382 more opinionated, so you could benefit from some prewritten
01:26:07.382 --> 01:26:07.951 prompts.
01:26:09.211 --> 01:26:12.984 That are kind of like make an agenda for the meeting that's
01:26:12.984 --> 01:26:17.009 coming up and one of the things that I find really helpful with
01:26:17.009 --> 01:26:20.719 it is whether I'm trying to get some code generated or I'm
01:26:20.719 --> 01:26:24.617 trying to get a creative thought which can include writing an
01:26:24.617 --> 01:26:28.579 e-mail or sending a summary or something. It is really helpful
01:26:28.579 --> 01:26:31.031 for me using ChatGPT or similar tools.
01:26:32.171 --> 01:26:36.951 To sort of like get past a creative ledge that I get stuck
01:26:36.951 --> 01:26:42.299 on. So I'll like get overwhelmed and busy, and if I lean on it to
01:26:42.299 --> 01:26:47.565 start the process that tends to help cuz then I'm willing to put
01:26:47.565 --> 01:26:52.670 the effort into fine tune it. But I find a procrastinate if it
01:26:52.670 --> 01:26:54.371 if I haven't started.
01:26:55.591 --> 01:26:58.872 So there's kind of a human element to AI that's helpful
01:26:58.872 --> 01:27:02.328 because it's kind of just like gonna do the work no matter
01:27:02.328 --> 01:27:02.621 what.
01:27:03.821 --> 01:27:07.202 So that I've actually found that to be really helpful to get
01:27:04.051 --> 01:27:04.351 Yeah.
01:27:07.202 --> 01:27:10.583 started. And then I'm like, well, it got pretty close, I can
01:27:10.583 --> 01:27:11.691 just tune it in now.
01:27:12.571 --> 01:27:14.701 So that's kind of an interesting part of it too.
01:27:15.611 --> 01:27:19.097 Yeah. And that goes back to, you know what I'm saying with being
01:27:19.097 --> 01:27:22.422 the the auditor of the AI. You know, there's there's a lot of
01:27:22.422 --> 01:27:24.031 value in having it prompt you.
01:27:24.351 --> 01:27:27.765 Umm. And in troubleshooting with it and using it as a as an
01:27:27.765 --> 01:27:31.292 intern basically right, this is your intern that you can chat
01:27:31.292 --> 01:27:34.933 with to come up with solutions and it was a lot of those on the
01:27:34.933 --> 01:27:38.346 CONNECT team are familiar with this worksheet. A lot of the
01:27:38.346 --> 01:27:41.475 formulas and stuff that are inside this worksheet were
01:27:41.475 --> 01:27:44.945 generated through ChatGPT. I starting out here, this company
01:27:44.945 --> 01:27:48.529 I don't know a lot about Excel and I certainly don't know much
01:27:48.529 --> 01:27:51.886 about Visual Basic but it's able to create me all of these
01:27:51.886 --> 01:27:54.161 formulas that populate these fields so.
01:27:54.771 --> 01:27:57.827 You know, thinking about how can you include this in your tooling
01:27:57.827 --> 01:28:00.790 if you guys are managing Excel sheets to do things and you need
01:28:00.790 --> 01:28:03.661 specific formulas or vlookups, it can write all of those like
01:28:03.661 --> 01:28:05.791 it created this whole vlookup for this stuff.
01:28:07.591 --> 01:28:10.673 Because I don't, I don't know to do that. If I'm being honest.
01:28:10.673 --> 01:28:13.609 But with the power of GPT, you know, suddenly I speak every
01:28:13.609 --> 01:28:16.544 programming language known to man, right? I could give it a
01:28:16.544 --> 01:28:19.381 piece of code, and I could say, can you translate this to
01:28:19.381 --> 01:28:22.561 Python? Could you translate this Python code to C? Can you debug
01:28:22.561 --> 01:28:25.644 this code for me? I have this error message and thinking about
01:28:25.644 --> 01:28:27.111 how we can use this this tool.
01:28:32.731 --> 01:28:36.349 Yeah. And if you don't mind, could you just, you went through
01:28:36.349 --> 01:28:38.741 just an example of generating some code.
01:28:39.631 --> 01:28:39.881 Yeah.
01:28:40.041 --> 01:28:44.097 Could you go through a few that it was really helpful with and a
01:28:44.097 --> 01:28:47.715 few that it was not that helpful. So we can sort of learn
01:28:47.715 --> 01:28:48.651 from your pain.
01:28:50.101 --> 01:28:54.036 Yeah, the code that it generates is not great, right? So keep in
01:28:50.691 --> 01:28:51.921 OK. Thanks.
01:28:54.036 --> 01:28:57.910 mind it doesn't have the context that we have around the entire
01:28:57.910 --> 01:29:01.542 project, but it is very good at figuring out like if it's a
01:29:01.542 --> 01:29:02.571 contained method.
01:29:02.931 --> 01:29:05.831 Umm, it's pretty good at figuring out what's what's going
01:29:05.831 --> 01:29:08.781 on, so you know, it came up in one of our our chats on the
01:29:08.781 --> 01:29:11.581 Kinect team. There was a piece of code that was kind of
01:29:11.581 --> 01:29:12.181 complicated.
01:29:13.601 --> 01:29:15.990 And we didn't really know what it was doing off hand. Maybe it
01:29:15.990 --> 01:29:16.141 was.
01:29:17.671 --> 01:29:21.384 A little bit higher level code or is written in a way that we
01:29:21.384 --> 01:29:22.881 didn't understand it and.
01:29:23.321 --> 01:29:27.811 Umm, it was fantastic it translating the code to.
01:29:28.501 --> 01:29:30.031 A piece of workable.
01:29:32.071 --> 01:29:34.231 Workable codes let me see if I can find it here.
01:29:35.481 --> 01:29:37.081 Mermaid charts.
01:29:37.201 --> 01:29:38.211 To.
01:29:44.341 --> 01:29:45.151 Ah, geez, really.
01:29:47.531 --> 01:29:51.010 I don't know if it's off hand. I used it so the whole application
01:29:51.010 --> 01:29:54.225 that you see here I don't know front end code development, I
01:29:54.225 --> 01:29:57.071 don't know felt but this application's built-in spelt
01:29:57.071 --> 01:29:59.021 using almost entirely code from GPT.
01:30:00.141 --> 01:30:03.500 Where I just asked for things and the cool part about that is
01:30:03.500 --> 01:30:06.804 as you ask for things, keep in mind it is conversational. It
01:30:06.804 --> 01:30:10.163 does have the limit. I'm not sure. Chris, do you know offhand
01:30:10.163 --> 01:30:12.871 what is the limit for this versus the playground?
01:30:13.991 --> 01:30:14.681 In tokens.
01:30:16.421 --> 01:30:21.275 Umm, I don't know the limit for the chat itself, but the yeah,
01:30:21.275 --> 01:30:22.431 the playground.
01:30:23.151 --> 01:30:23.801 Umm.
01:30:24.661 --> 01:30:27.387 I I believe they're the end of being the same. If I'm not
01:30:25.051 --> 01:30:25.421 I.
01:30:27.387 --> 01:30:30.395 mistaken, the only thing is you can change the endpoints you're
01:30:30.395 --> 01:30:30.771 hitting.
01:30:31.221 --> 01:30:31.541 Right.
01:30:32.711 --> 01:30:36.181 So and I know that like some of the endpoints like, there's
01:30:36.181 --> 01:30:39.592 actually a different endpoint that's made just for writing
01:30:39.592 --> 01:30:39.881 code.
01:30:40.701 --> 01:30:44.795 So this is more of like a standard LLM. Whereas I think
01:30:44.161 --> 01:30:44.481 Yeah.
01:30:44.795 --> 01:30:46.111 it's called Codex.
01:30:47.701 --> 01:30:50.932 Like one of the things ultimately we'll wanna do is
01:30:50.932 --> 01:30:54.722 again, I would just encourage folks to consider this. If you
01:30:54.722 --> 01:30:58.450 want to break out from this group and collaborate with some
01:30:58.450 --> 01:31:00.811 of the folks who were speaking today.
01:31:01.921 --> 01:31:05.481 Messages on teams and we'll coordinate offline.
01:31:06.741 --> 01:31:11.171 But we ought to create our own, you know, CLT.
01:31:12.311 --> 01:31:15.949 AI tool or something? You know where we can all just keep
01:31:14.281 --> 01:31:14.591 Yeah.
01:31:15.949 --> 01:31:17.391 working on it together?
01:31:18.811 --> 01:31:23.984 Internally and have it call ChatGPT, you know, Open API,
01:31:23.984 --> 01:31:25.981 Open AI API endpoints.
01:31:28.251 --> 01:31:31.826 But I didn't answer your question. It is. It does seem
01:31:31.826 --> 01:31:35.141 like it has a limitation that is very frustrating.
01:31:35.691 --> 01:31:38.997 Yes. Yep, Yep. And that's, that's the difference between T3
01:31:36.861 --> 01:31:37.351 Yeah.
01:31:38.997 --> 01:31:42.468 and T4. Like we said, the token difference. So if you're gonna
01:31:42.468 --> 01:31:45.664 paste a lot in here, I think this message. Yeah. See this
01:31:45.664 --> 01:31:49.245 message got cut off. So you may have to break it up. What you're
01:31:49.245 --> 01:31:52.551 saying to it into chunks to send it off a little bit better
01:31:52.551 --> 01:31:55.912 because there is a token limit when you interacting with it.
01:31:55.912 --> 01:31:58.501 And this was for Greg and anybody that was on.
01:31:59.501 --> 01:32:03.573 I wanna say this is PBC with the OIC integration. There's a
01:32:03.573 --> 01:32:07.374 complicated piece of XML response that we didn't really
01:32:07.374 --> 01:32:11.649 know how to parse by itself, so Brendan Lyon, you may this may
01:32:11.649 --> 01:32:15.789 look familiar to you. I just asked it how can I convert this
01:32:15.789 --> 01:32:16.671 to a C Class?
01:32:17.401 --> 01:32:21.008 Given that it has all these weird namespaces in there and it
01:32:21.008 --> 01:32:24.734 came back with the actual code that I would need to be able to
01:32:24.734 --> 01:32:28.282 parse that out given those namespaces. So this is what kind
01:32:28.282 --> 01:32:31.771 of helped me solve that and get the ball rolling for Greg.
01:32:32.911 --> 01:32:34.081 To where he could parse that out.
01:32:41.291 --> 01:32:45.002 Yeah, I think I think this will definitely be something that
01:32:45.002 --> 01:32:48.349 we'll have just continual discussion on. I I see we're
01:32:48.349 --> 01:32:51.938 sort of right at times, anyone else have any comments. And
01:32:51.938 --> 01:32:55.467 thanks again Larose that was really insightful seeing how
01:32:55.467 --> 01:32:59.422 you're using it and of course of you know this isn't going to be
01:32:59.422 --> 01:33:03.316 an immediate game changer across the board, right. We're trying
01:33:03.316 --> 01:33:06.601 to get everyone exposed to IT pros and cons and hand.
01:33:08.331 --> 01:33:12.680 See the reality of it. Uh, and sort of hopefully just get you
01:33:12.680 --> 01:33:16.960 thinking, you know, this is a thing that you want to embrace
01:33:16.960 --> 01:33:21.028 it just as a side note, so everyone's aware we do need to
01:33:21.028 --> 01:33:25.588 work out how we're going to set up ChatGPT, premium accounts for
01:33:25.588 --> 01:33:29.306 folks who do want those so that we can coordinate it
01:33:29.306 --> 01:33:33.725 logistically. It's it's kind of frustrating because they don't
01:33:33.725 --> 01:33:36.461 have like, a corporate account system.
01:33:37.061 --> 01:33:40.529 So just as a side note, if there's a little bit of a lag in
01:33:40.529 --> 01:33:44.227 getting that turned on, that's why we just want to make sure we
01:33:44.227 --> 01:33:47.753 manage that responsibly as a company. But any other comments
01:33:47.753 --> 01:33:49.891 or questions before your wrap it up?
01:33:56.231 --> 01:34:01.222 Mountain say it initially when you hear about the AI invasion,
01:34:01.222 --> 01:34:03.361 especially into the coding.
01:34:04.591 --> 01:34:07.690 Workspace. You kind of get scared. Like all great here they
01:34:07.690 --> 01:34:10.841 come to come to take me to my job is getting taken away too.
01:34:11.711 --> 01:34:15.417 But it really isn't so much that it's going to be able to write
01:34:15.417 --> 01:34:18.081 all the code that we do. It's something very.
01:34:18.681 --> 01:34:22.162 It's very particular to help you. You know, I use Jason or
01:34:22.162 --> 01:34:25.821 JSON to C all the time and I think of it more as an expansion
01:34:25.821 --> 01:34:29.597 of that. That's basically AI in a sense. And this is just blown
01:34:29.597 --> 01:34:33.315 up to where it's conversational and can help you step by step.
01:34:33.315 --> 01:34:36.855 So I'm actually really excited about where we're going with
01:34:36.855 --> 01:34:40.041 this. And I'm happy that we're taking a step forward.
01:34:41.781 --> 01:34:43.411 Yeah. Thank you. Thank you.
01:34:44.471 --> 01:34:47.928 Uh, very great comment. Thank you, Sir Lynch. And over to you,
01:34:47.928 --> 01:34:48.641 Lady Johnson.
01:34:49.581 --> 01:34:52.916 Just a random side note to the PM team. If you're going to be
01:34:52.916 --> 01:34:56.411 using this to respond to client communications, either in e-mail
01:34:56.411 --> 01:34:59.907 or Basecamp, please send it by a manager first, either Mickey or
01:34:59.907 --> 01:35:00.821 myself to review.
01:35:00.681 --> 01:35:01.191 Agreed.
01:35:02.041 --> 01:35:02.591 Perfect.
01:35:06.491 --> 01:35:07.561 Here here.
01:35:11.671 --> 01:35:14.401 Alright, turn the video on. You can say hi. Jessica wants to say
01:35:14.401 --> 01:35:15.031 hi really fast.
01:35:16.691 --> 01:35:17.661 You guys see Queenie?
01:35:18.111 --> 01:35:18.961 Ah.
01:35:18.681 --> 01:35:19.381 Hi.
01:35:20.421 --> 01:35:21.111 Hello.
01:35:21.291 --> 01:35:22.601 This is our third clone.
01:35:24.321 --> 01:35:24.921 A high.
01:35:25.661 --> 01:35:30.999 Say hi, say I'm way smarter than a I say yeah, AI can't mess with
01:35:30.999 --> 01:35:33.101 me. She's trying to steal.
01:35:32.011 --> 01:35:33.961 Can't believe how big she is already.
01:35:32.391 --> 01:35:32.591 Hi.
01:35:32.661 --> 01:35:34.101 I Jessica.
01:35:34.161 --> 01:35:35.171 Hi.
01:35:37.551 --> 01:35:40.103 And Charlie, I seen you had a, I, I see you had your hand up
01:35:40.103 --> 01:35:40.981 over to you good Sir.
01:35:41.661 --> 01:35:46.456 Yeah, just sort of echoing what Jim said. I I feel like I've
01:35:46.456 --> 01:35:51.487 been around long enough to see disruptive technologies come out
01:35:51.487 --> 01:35:52.351 before and.
01:35:53.481 --> 01:35:57.927 Disruption is opportunity. Jeremy made a couple of
01:35:56.221 --> 01:35:56.481 Yep.
01:35:56.891 --> 01:35:57.501 Here here.
01:35:57.927 --> 01:36:03.506 references about like it's not about displacing jobs, it's not.
01:36:03.506 --> 01:36:08.301 It is leverage. It is about moving forward in the most
01:36:08.301 --> 01:36:13.532 robust and effective fashion with what we already have. And
01:36:13.532 --> 01:36:19.024 yeah, like dig in because this is such an inflection point and
01:36:19.024 --> 01:36:23.731 it is such a force multiplier to move us forward for.
01:36:23.911 --> 01:36:26.571 The foreseeable future. Super cool. Thanks guys.
01:36:28.331 --> 01:36:28.801 You're here.
01:36:29.911 --> 01:36:32.261 Here, here. Well said, good Sir.
01:36:33.191 --> 01:36:37.024 Yeah, it is a huge opportunity and like I I think several folks
01:36:37.024 --> 01:36:40.138 have mentioned, you know, I think it really can be,
01:36:40.138 --> 01:36:44.030 especially if you think about it in the ecommerce and the Kinect
01:36:44.030 --> 01:36:47.384 space as far as like a tactical output from our product
01:36:47.384 --> 01:36:48.581 offerings currently.
01:36:49.441 --> 01:36:53.488 We can be huge disruptors over the next 12 to 24 months and we
01:36:53.488 --> 01:36:57.277 have our own DNS server. Guys like you have to see my face
01:36:57.277 --> 01:36:57.791 earlier.
01:36:58.641 --> 01:37:01.825 So exciting. So anyways. Alright. Well, unless anyone
01:37:01.825 --> 01:37:05.304 else has anything. We're a little over Happy Friday, guys.
01:37:05.304 --> 01:37:08.546 I hope you have an amazing weekend. Thanks again for a
01:37:08.546 --> 01:37:12.202 great training for everyone who shared and LED and see you on
01:37:12.202 --> 01:37:13.381 Monday. Bye for now.
01:37:14.471 --> 01:37:17.301 Have a great weekend. You all bye.
01:37:14.901 --> 01:37:17.780 You have. If you guys have any questions around it, let me
01:37:15.491 --> 01:37:16.131 Have fun all.
01:37:16.381 --> 01:37:17.151 Have a great weekend.
01:37:17.780 --> 01:37:20.561 know. Feel free to ping me and we can we can dig into it
01:37:20.391 --> 01:37:20.881 Ohh.
01:37:20.561 --> 01:37:20.951 further.
01:37:21.581 --> 01:37:23.131 I'll be picking your brain like mad.
01:37:22.181 --> 01:37:27.681 Thanks, Jeremy. Take care of y'all. See. Yeah. Thanks guys.
01:37:25.521 --> 01:37:27.121 See, everybody. Have a good weekend.
01:37:26.471 --> 01:37:27.331 See you all later.
01:37:27.221 --> 01:37:29.201 Michael, this is very informative.
01:37:27.681 --> 01:37:28.781 Bye for now.
01:37:28.091 --> 01:37:28.491 Thanks guys.
01:37:28.411 --> 01:37:28.691 Yeah.
01:37:29.931 --> 01:37:30.421 Have a good one.