Exploring AI in Content Creation: Tools, Techniques, and Future Possibilities with Dave Sobel

[00:00:13] Today is all about AI, but not how I normally talk about it. Today, you'll learn how I'm using it and how I'm considering using it in the future. Trust me, we're going to have some fun with this show.

[00:00:27] Welcome to the Business of Tech Lounge, the live version of the Business of Tech Podcast. It's Wednesday, December 4th, 2024, and I'm Dave Sobel.

[00:00:35] Now, I will distinctly be taking questions and comments throughout the show, so make sure to put them in chat. If you have a question, I will happily respond to it.

[00:00:46] Now, I want to thank Sales Builder, our Patreon sponsor whose support makes this show possible. Focus on your IT sales workflow with the power of automation and visit them at salesbuilder.com. That's B-U-I-L-D-R.com.

[00:01:00] Now, reminder, I am definitely watching chat and I will watch for questions and comments.

[00:01:06] So the inspiration from this was actually a recent presentation I did. I was asked sort of how AI is being used in the field by MSPs and their customers.

[00:01:17] I did a whole presentation on this and actually I ended on talking a little bit, just briefly, about how I'd use AI to prepare for that particular presentation.

[00:01:28] And interestingly, it was one of the most popular portions of the discussion.

[00:01:33] And I figured, particularly with some of the things I'm going to show you a little later and as I was thinking about it, it would be useful for you as my audience to get a sense of what exactly I am using.

[00:01:43] So the first I thought I'd do is I'll walk you through the workflow of preparing the daily news show.

[00:01:49] So obviously for listeners, I put out a roughly 10 minute show every business day where I cover three to four stories and I ask, why do we care?

[00:01:58] Let me walk you through how I put that all together and how I make the show.

[00:02:06] So the first step really is pulling together all the news.

[00:02:11] And the way I do that is through RSS.

[00:02:14] Excuse me.

[00:02:16] I use a lot of RSS feeds to pull that all together.

[00:02:19] I'll tell you, I use Feedly as my RSS reader.

[00:02:22] One of the things I've used AI in Feedly to do is that it's grouping stories for me so that I can pull lots of sources together and I can see groups of stories.

[00:02:33] So, for example, Amazon reInvent would be a topic area and it gets clustered together and I get to see areas around a particular product launch, particular press release.

[00:02:44] And Feedly is helping group that and show related stories.

[00:02:49] Additionally, I'm going to note that there's also a translate button that's available to me when I'm reading stories so I can read foreign language versions.

[00:02:56] And in fact, I subscribe roughly to five, five hundred and fifty different news sources and I'm using AI translation for a portion of that.

[00:03:05] Now, what I do then is I take all of the stories and I throw them into a notion database.

[00:03:09] If you're familiar with the product notion, that's how I track this.

[00:03:12] The cool thing for me is this allows me to do story processing in lots of different ways.

[00:03:18] In the morning, I tend to do it on my iPad and by using the extension, I can throw the web clip and throw it into the database.

[00:03:24] So all of the stories that I'm interested in get thrown into a database.

[00:03:29] Notion is actually doing summarization.

[00:03:32] There's a specific AI summary field.

[00:03:34] But what's particularly powerful is I've built a custom set of instructions designed to say exactly the way I want the story summarized.

[00:03:42] I talked about the fact that I wanted in a readable script.

[00:03:45] I've given examples of the way that I like it done, including specifics like spell out the numbers to make it readable so that when I put it in a teleprompter,

[00:03:54] I actually can read the script.

[00:03:56] So it formats on, I focus it on data and the format that I want it done so that each of the stories comes into a standardized way that I like to use that.

[00:04:05] So then in Notion, I have the ability to pick and choose all of the stories that I'm going to consider for a particular day.

[00:04:10] I do groupings.

[00:04:11] This is all manual and I do all the curation myself.

[00:04:14] So I make sure that everything is the way that I want it.

[00:04:17] When I start working, I just work in Word.

[00:04:20] I'm a little old school on that way.

[00:04:21] So I work in Word to create my scripts.

[00:04:23] And the first part I do is I pick the stories and I organize the stories.

[00:04:27] I tighten it up and I use Grammarly, which of course is AI and machine learning, to focus the language.

[00:04:33] I've told Grammarly the style I like to write to, the way that I want it to check my grammar, the kind of tightening that I want on the story.

[00:04:40] So that gets me another version of the draft.

[00:04:42] Next, what I do is I'll start working with ChatGPT and I've built a custom GPT around my why do we care segment.

[00:04:49] What it does is that it actually asks a series of questions to me.

[00:04:53] And additionally, specifically designed to consider the negative case as well as alternative arguments so that I can then have a discussion with myself essentially around why do we care?

[00:05:05] I then use, I work with that.

[00:05:07] That helps me craft the couple of paragraphs that become why do we care?

[00:05:11] And again, I use Grammarly to tighten the language.

[00:05:13] At that point, I've got three or four stories in the script.

[00:05:17] I then have another custom GPT that I'm using to do headline generation so that I put it in there and say, give me ten different options for headlines.

[00:05:26] And then I start picking the ones that I'm interested in.

[00:05:30] When I have completed the entire script, it goes into another bot which helps me with episode title generation where I've told it specifically the way I like titles of the episodes to be done.

[00:05:41] I'm factoring in the YouTube title limit to make sure that I've done that correctly.

[00:05:45] Additionally, I then have another set of prompts that pulls out a show description and then I've produced the show.

[00:05:52] So I create it in Premiere Pro.

[00:05:54] I export it.

[00:05:55] I have an audio and a video version there from the same source file.

[00:05:58] That AI then will then take that in another tool, will chalk that up into shorts, generates all the subtitles you see in my social media feeds, and it helps me with my LinkedIn post generation.

[00:06:10] That's all of the different components that at every point in the process, I'm checking the work.

[00:06:16] So I'm reading it.

[00:06:17] I'm making sure that it's what I'm comfortable with.

[00:06:19] Am I like that?

[00:06:20] Do I not like that?

[00:06:21] I make adjustments the entire way.

[00:06:23] Now, I also do a lot of interviews on this show.

[00:06:25] And I will tell you, I've built another custom GPT that helps me with interview prep.

[00:06:30] I put all of the information that I've gathered on the person, including if there's a PR pitch, why I was offered the person, it pulls in their LinkedIn profile, and it pulls it all together in order to create for me both a biography of the person and a list of 10 starter questions that I never use.

[00:06:49] But what they do is they frame my thinking so that when I'm going into the interview, I'm prepared with, these are some of the kinds of things that might have been asked or might want to ask to give me topic areas.

[00:07:01] For those of you watching my IT Nation Connect interviews, I used that same generation.

[00:07:06] That's what's on the cards that I'm holding in all of those.

[00:07:10] I'm using AI in the creation of the audio itself.

[00:07:14] It does splicing the footage, remove silence.

[00:07:17] It does AI and voice enhancement, and I then put it into the same episode process.

[00:07:22] And finally, I use it generally for anything that I'm presenting on.

[00:07:26] So any kind of presentation, I take everything that I'm doing and I do iterations through it.

[00:07:31] First, I pull data points from my AI, the AI in Notion where I keep all my stories.

[00:07:37] It helps me pull all the data points.

[00:07:39] And any of my writings I put into ChatGPT and I ask it to then ask me questions about it.

[00:07:45] So that I say, like, have I thought of this? Have I thought of that?

[00:07:48] You can see that throughout the entire process, AI is my real-time partner.

[00:07:53] So that's the way that I'm using it in a lot of bits.

[00:07:55] My guiding principle for this the entire time is, it is a tool to assist me with the creation of content at an accelerated pace.

[00:08:04] It gives me ideas and it challenges mine and helps as an editor, but it never produces without any level of checking on my behalf.

[00:08:14] Now, reminder, if you're watching, if you've got questions, throw them into the question area and I will definitely respond to them.

[00:08:21] And I will let you know more of my thinking.

[00:08:24] Now, the other reason I was thinking about this is besides, I thought it would be interesting to you as a listener to get a sense of what I'm doing.

[00:08:30] I've also been doing some experimentation with a cool set of AI tools, and I'm really thinking about the way that I might use them.

[00:08:37] In fact, one allows for the generation of an AI video version of myself, and the other actually is a complete clone of my AI voice.

[00:08:47] So I thought it would actually be a lot of fun to welcome the intrepid producer for the show, Sharon of Picture This Video, to come onto the stage.

[00:08:56] Sharon, join me on stage here.

[00:08:58] Hello, here I am.

[00:08:59] So you're normally behind the scenes, but I thought it would be super fun to have you for a little bit of a fun test, because ultimately you're one of the people that knows my video and voice probably better than anybody.

[00:09:12] Having edited a ton of my videos, having looked at it a lot, I feel like you'd be exactly the right person to play with a little test.

[00:09:20] You up for it?

[00:09:21] Sure.

[00:09:22] Okay, so here's the deal.

[00:09:25] I have created three versions of the same video with the same script.

[00:09:31] Okay, it is a 45 second segment that played earlier this week, and I regenerated it in different ways.

[00:09:38] The first is entirely me.

[00:09:40] The second is so it's completely real.

[00:09:43] The second way I made it was my voice, so a recording of my voice, but an AI generated me.

[00:09:51] And then the last version of it is entirely AI.

[00:09:55] Both the voice and the graphic is entirely done with artificial intelligence.

[00:10:02] So your mission, should you choose to accept it, is to figure out which one is which.

[00:10:08] We're going to watch each one, and you'll watch carefully, and you'll tell me which one feels real and which one feels AI and why.

[00:10:16] And you'll try and identify which is of each of the three.

[00:10:19] No pressure.

[00:10:21] Hmm.

[00:10:21] But I am really curious to see how well you can tell them all apart.

[00:10:26] Okay, and just for the viewers, I just wanted to be disclosed that I have not seen any of these in advance.

[00:10:34] That was part of the fun.

[00:10:35] I wouldn't even let her see it in rehearsal.

[00:10:37] Now audience, I want you to play along too.

[00:10:39] So remember, we're looking for is it fully real?

[00:10:42] Is it a real voice with AI avatar version of Dave or is it all AI?

[00:10:48] And if you're listening to the podcast recording and listening to the audio listeners, I've put a link in the show notes to take you to the video of this so you can actually see this exercise entirely in video.

[00:11:00] So Sharon, you ready?

[00:11:02] Sure.

[00:11:04] All right.

[00:11:04] Let's roll the first AI taste test.

[00:11:08] Okay.

[00:11:08] Microsoft has refuted allegations that it collects data from its Microsoft 365 applications, such as Word and Excel, for artificial intelligence training.

[00:11:18] This statement follows a viral Tumblr post claiming the company used a feature called Connected Experiences to scrape user data.

[00:11:25] A Microsoft spokesperson clarified that customer data is not utilized to train large language models and emphasize that the Connected Experiences feature, available since April 2019, is designed to enhance productivity with tools like real-time grammar suggestions and co-authoring.

[00:11:42] Users can choose to turn this feature on or off within the settings of their Office applications.

[00:11:48] The company reassured customers they have full control over their data privacy settings.

[00:11:57] And the new views on or off within the Microsoft Office of Microsoft Office, is designed to create a defective which was recommended to do to get to the web- whose hjitssi for immersive reality-based budget-based Zap circa 2021.

[00:11:58] Do you want to guess at the end and see each one?

[00:12:01] No, no, no, no.

[00:12:02] I want to guess as we go.

[00:12:04] Okay.

[00:12:24] thinks that this is the you think it is a real voice i think it is a real voice and ai dave

[00:12:32] all right let's do let's roll on to the second one all right microsoft has refuted allegations

[00:12:41] that it collects data from its microsoft 365 applications such as word and excel for artificial

[00:12:47] intelligence training this statement follows a viral tumblr post claiming the company used a

[00:12:53] feature called connected experiences to scrape user data a microsoft spokesperson clarified that

[00:12:59] customer data is not utilized to train large language models and emphasize that the connected

[00:13:04] experiences feature available since april 2019 is designed to enhance productivity with tools like

[00:13:10] real-time grammar suggestions and co-authoring users can choose to turn this feature on or off within the

[00:13:17] settings of their office applications the company reassured customers they have full control over

[00:13:22] their data privacy settings hmm well for that one i'm gonna say that the audio was artificially

[00:13:37] intelligent just made up a word you did okay uh and i'm only saying that because there's like

[00:13:48] background noise that i didn't hear in the first one okay which might mean that there that's completely

[00:13:56] opposite but all right but this is your reaction what i think and the video so that was tricky because

[00:14:04] there were some parts of the video that reminded me of things i saw in the first one where i said i don't

[00:14:10] think that's dave dave okay but it was not as obvious to me in this one as it seemed to be obvious in the first one okay so what's your

[00:14:23] guess so if you're telling me it's a uh audio or video or neither i'm gonna say audio

[00:14:35] audio and i'm gonna hope the third one is just all right all right so let's roll running out of options

[00:14:43] let's roll the last one okay microsoft has refuted allegations that it collects data from its microsoft

[00:14:50] 365 applications such as word and excel for artificial intelligence training this statement follows a viral

[00:14:56] tumblr post claiming the company used a feature called connected experiences to scrape user data

[00:15:02] this person clarified that customer data is not utilized to train large language models and

[00:15:08] emphasized that the connected experiences feature available since april 2019 is designed to enhance

[00:15:14] productivity with tools like real-time grammar suggestions i'm not hearing that echo thing in this

[00:15:18] one users can choose to turn this feature on or off within the settings of their office applications

[00:15:23] the company reassured customers that they have full control over their data privacy settings

[00:15:30] yeah so so you 100 dave on the third one is my guess all right you ready you ready for the answers

[00:15:39] all right so the third one is 100 me that is all that is all real so you correctly identified that

[00:15:47] however the first one is actually all ai that you thought was my real voice okay so the the first

[00:15:56] one was your real voice and then they chopped it up and made new words out of it but they had to start

[00:16:01] with your real voice in the beginning i know this so fair but but so so the voice synthesization is so

[00:16:07] good that it tricked my producer now the second one is actually my voice and an ai avatar well you should have

[00:16:17] added a filter says your producer you needed a noise uh eliminator on that on that filter but you didn't

[00:16:26] have me produce it so how would you know see that's all fair but but but but so you it was in fact an ai

[00:16:32] graphic dave but the voice was was actually me so all right dave without a producer number two

[00:16:42] dave without a producer dave without a producer so there is that humorous but i think i think what

[00:16:47] it amusingly you've made the argument that it's at least really usable uh in that but particularly on

[00:16:56] the voice front yeah the first one first one pretty darn good it it uh it actually it actually fooled you

[00:17:05] so that's uh really kind of insightful to learn that the voice synthesis is that good well sharon

[00:17:10] thank you for being such a good sport in this one well you're welcome now i want to show the audience

[00:17:16] the the something else that i've been playing with on on this as well and i want to start this by

[00:17:22] saying so listeners i don't speak spanish let's roll the fourth ai clip microsoft ha refutado las

[00:17:31] alegaciones de que recopila datos de sus aplicaciones de microsoft 365 como word y excel para el entrenamiento

[00:17:38] de inteligencia artificial esta declaración sigue a una publicación viral en tumblr que dice que la

[00:17:44] empresa usó experiencias conectadas para recopilar datos de usuarios un portavoz de microsoft aclaró

[00:17:50] que los datos de los clientes no se utilizan para entrenar modelos de lenguaje grandes y enfatizó que la

[00:17:55] función de experiencias conectadas disponible desde abril de 2019 está diseñada para mejorar la

[00:18:01] productividad con herramientas como sugerencias gramaticales en tiempo real y coautoría los

[00:18:07] usuarios pueden elegir activar o desactivar esta función dentro de la configuración de sus aplicaciones

[00:18:12] de office la empresa aseguró a los clientes que tienen control total sobre la configuración de

[00:18:17] privacidad de sus datos now what's fascinating that particularly if you're listening on audio i would

[00:18:24] really encourage you to watch that is that is the real video of me converted to speak spanish in a

[00:18:32] voice that sounds pretty much like mine and again while i took three years high school and one year of

[00:18:38] college spanish i am not fluent in spanish uh it matches the mouth movements perfectly so i let the

[00:18:45] video run a little bit longer too is is that it is such a good job of doing that uh i was kind of had my

[00:18:52] mind blown a little bit this uh when i saw that version of it uh and it immediately has some real

[00:18:59] implications for the creation of localized content in various markets uh what i wanted to pose to the

[00:19:08] audience is i've actually been i'm bringing this to my audience because i've been thinking about lots

[00:19:12] of ways to do this i'm going to work on a spanish language version of the show as that seems like an

[00:19:18] incredibly natural extension uh i'm going to disclose that it's ai generated in the description of the

[00:19:24] show and in the episode notes uh because i do not think it is the intention is not to deceive the

[00:19:30] intention is to offer my content in additional ways uh i still believe that fundamentally that the value

[00:19:36] that i bring you as an audience is the curation the ability my ability to curate the stories that i

[00:19:43] think are important based on my industry expertise and additionally the perspective that i offer in why do we

[00:19:49] care i think that's the bit that is the most valuable portion of what i do and so i want to let you know

[00:19:55] that i'm distinctly not thinking of replacing uh i don't think that makes any sense the what i want

[00:20:01] to get some feedback from you as an audience is the situations i am considering using the ai english version

[00:20:09] of dave uh i'm considering this in what i would consider sort of emergency type situations uh what i'm

[00:20:15] thinking is is heaven forbid i get sick uh and my voice is unusable for a day or two but i need to

[00:20:22] produce an episode uh that feels like a natural way to do this uh if i am traveling i could actually record

[00:20:29] uh on audio but link it to studio video which makes the show consistent and delivered at the high quality

[00:20:35] that i want with the lighting the sound the the the level of production that i want to produce the show at

[00:20:40] uh i could also see it necessarily if i have family emergency or or some kind of need where

[00:20:45] i can produce the show but maybe not in the with all of the technology that i have uh i thus am also

[00:20:52] considering what makes sense to my disclosure rules for using ai in the english language show uh how much

[00:20:59] disclosure makes sense uh one of my guiding principles here is that there is always a human

[00:21:04] and a human is me who approves the work but i believe that there are certain situations that i might

[00:21:10] be able to extend my coverage or be able to do a little bit more by leveraging this but the only

[00:21:16] way that makes sense to me is open transparency and hearing from my audience so one of the reasons i

[00:21:20] wanted to do this version of the show was to get a little bit of feedback you guys watching live seem

[00:21:26] to be pretty quiet on this so i'm going to focus and remind everyone put it into the comments on the

[00:21:30] youtube video reach out on linkedin if you're listening to the podcast you can reach out at

[00:21:34] question at mspradio.com if you've got questions or thoughts but i in particular want to hear

[00:21:39] your thoughts on what are the appropriate uses of ai in this situation when we think about what

[00:21:46] the content is a reminder i am distinctly following the chat and if you've got thoughts i will put them

[00:21:52] up on stage now now i also want to give you a preview of an episode coming this weekend i sat down with

[00:21:58] elliot hyman now he's the ceo of lyra technology group to discuss the unique challenges of overseeing

[00:22:05] operations for a group of disparate msps working collectively across the portfolio of evergreens

[00:22:11] companies elliot shared some insights on balancing standardization with preserving each business's unique

[00:22:17] strengths the tangible business's benefits of scale and the critical factors for sustaining growth in the

[00:22:24] mid-market msp space here's a preview of that interview a little bit about your vision of what

[00:22:32] sustainable growth needs um you know well sustainable growth is about building healthy companies that

[00:22:39] serve their employees well and serve their their uh their customers well and you know for us that means

[00:22:46] adding value to those those key stakeholders and so we do that in a lot of different ways but if

[00:22:51] we're doing that consistently then you know that's going to produce sustainable growth and you know we

[00:22:56] also know what doesn't produce sustainable growth which is trying to uh squeeze every dollar cost out

[00:23:02] of a business and you know uh run it in a way that um that isn't for the the long-term good of the

[00:23:12] employees or customers and we work very very hard to to not do that okay and you know it can be tempting

[00:23:20] but yeah we you know we're always balancing the long-term and short-terms uh in how we operate

[00:23:29] our business and we work very well as we work very hard to do that right now

[00:23:32] it'll drop on the weekend on youtube and the podcast feed if you're interested i encourage you to listen

[00:23:39] visit patreon.com slash mspradio to sign up now now i want to thank sales builder our patreon

[00:23:46] sponsor whose support makes this show possible focus on your it sales workflow with the power of

[00:23:51] automation and visit them at salesbuilder.com that's b-u-i-l-d-r dot com and vendors you too

[00:23:59] can get your name mentioned of the live show it's a simple monthly subscription visit patreon.com

[00:24:05] slash mspradio to learn more and listeners the number one thing you can do to support the show

[00:24:10] it's like share and follow on your favorite platforms support directly on patreon with our

[00:24:16] give what you want model and you set when you think the content is worth you get access to videos early

[00:24:22] if you have a question or a thought on what we talked about today and are listening to the recording you can

[00:24:27] send it in at question at mspradio.com you're watching on link on linkedin put the comments below

[00:24:33] as well as on youtube thanks for joining me for the business of tech lounge and i will see you real me

[00:24:39] next time