Episode 4
Unlocking the AI Advantage: Business Implementation Barriers
Today, we delve into the challenges and barriers to business adoption of AI, specifically focusing on generative AI technologies. Mikah, Mike, and Artie discuss the significant hurdles companies face, including high implementation costs, data privacy concerns, and the fear of job displacement. They emphasize the importance of understanding business problems before adopting AI solutions, as many organizations struggle to articulate their needs. The conversation highlights the potential for AI to revolutionize business processes by offering tailored, localized solutions that enhance productivity and employee engagement. As the landscape of AI continues to evolve, the hosts explore how companies can leverage these technologies to gain a competitive edge and drive innovation.
Transcript
Welcome back.
Mike:Hi.
Micah:Hello.
Micah:How's it going, Mike?
Mike:Good.
Mike:How are you, Micah?
Mike:I almost called you Artie.
Micah:Call me whatever you want, man, but I'm good either way.
Micah:Welcome to another stunning, exciting, mysterious, and not controversial, troublesome, troublesome, nightmarish, fully regarded.
Micah:That's a little might regret this.
Micah:That's a wink to you Wall Street BET folks.
Micah:Anyway, welcome back to the podcast and thank you for being here.
Micah:We have been pleasantly shocked in the last couple of days by actually seeing some traction on this, which is hard to believe.
Micah:And so clearly, tonight's goal is to shut that down, put an end to that.
Mike:Let's figure out the best way to not get people interested.
Micah:Yeah, that's right.
Micah:We.
Micah:We want to be in single digits.
Mike:Yes.
Micah:Like you don't want everybody in the world.
Mike:We need to go backwards.
Micah:We might.
Micah:Yes.
Micah:Well, I.
Micah:We're probably already going backwards.
Mike:Probably.
Mike:Yeah.
Mike:That probably happened when we hit record.
Micah:That's right.
Micah:We're already spiraling down.
Micah:So today's episode, we're going to talk about the challenges or the barriers to business adoption of AI and what some of those are, and Artie's gonna help us work around what are some fixes.
Micah:And then we're gonna talk about.
Micah:So we're recording this on November 6th.
Micah:So this is the day after election Day.
Micah:Yay.
Micah:Yay.
Micah:Yeah.
Micah:So it's over.
Micah:We're in the gloating phase, but it's diminishing quickly.
Micah:And by tomorrow, we can settle back into just all being fully polarized and hating each other the good old fashioned way with fewer yard signs.
Micah:So it's gonna be.
Micah:It's gonna be great.
Micah:So we're gonna talk a little bit about maybe, you know, what that might mean.
Micah:But I wanted to share.
Micah:I wanted to make an admission on air.
Mike:Micah's Diaries.
Micah:That's right.
Micah:Starfleet, Patreon, Content, Starlog, Stardate, whatever.
Micah:So my admission is that I don't listen to the podcast.
Micah:I don't.
Micah:I haven't.
Micah:And as of today, our third episode is the only episode available.
Micah:That's the one that just became available to everybody this morning.
Micah:But this is our, I think, sixth one that we've recorded, but this might.
Mike:Come in out of order, so we'll see.
Micah:Yeah, it's.
Micah:It's.
Micah:It's a great.
Micah:A number greater than six, four doesn't matter.
Micah:Yes.
Micah:Less than a thousand and greater than 3.
Micah:But yesterday I had to drive to the office and back and got caught in traffic, and I decided to listen to the podcast, and it's a very weird thing to hear yourself coming out of your car radio.
Micah:Like, for.
Micah:The first thought for me was, I hate the sound of my voice.
Micah:I decided I sound like a Muppet with a case of some sort of constipation or some sort of gastrointestinal issue.
Mike:I think I saw a comment where somebody said this.
Micah:It seems likely, if they haven't, they will.
Micah:But the second thing I noticed was, man, do I absolutely love the word absolutely.
Micah:And I was sort of shocked.
Micah:And if I had more attention span than I do, like any, I was going to try to count in the time that I listened how many times I said absolutely.
Mike:And then he got cut off and yelled and screamed and forgot all about his mission of counting.
Micah:That's right.
Micah:I couldn't use my fingers because I was counting.
Micah:I was telling the other guy, he's number one and still trying to steer with the other hand.
Micah:So, anyway, I will try to clean up my language a little bit.
Micah:I also, I'm not sure.
Micah:Well, I am sure where I went to school.
Micah:I'm not sure where I came up with the sentence structures that I seem to use.
Micah:But in my.
Micah:So Mike knows this about me, but everybody else probably doesn't, that I really perceive the world in shapes.
Micah:Like I think in 3D shapes, almost everything.
Micah:How I feel about stuff and how I remember stories, everything.
Micah:And as I was listening to the podcast and imagining my sentence structure, it was this crazy zigzagging line.
Micah:It was like the Tasmanian devil depixelated down to where he could only do straight lines.
Mike:And anyway, I think that's just you.
Micah:It probably is.
Micah:Well, yeah, I was just sharing that as an observation.
Micah:I'm not going to try to work on that.
Micah:There's no hope for that.
Micah:I'm way too old to change to change that.
Micah:But I will try to say absolutely slightly less, and I'll come up with an equally annoying word that I'll just replace it.
Mike:Absolutely.
Micah:Absolutely.
Micah:Okay, we're all in agreement.
Micah:This has been interesting.
Micah:The other one I say all the time is so.
Mike:So I think we all.
Mike:We both say that.
Mike:I mean, it's a segue and it's.
Mike:It's hard to break.
Micah:Yeah.
Micah:Speaking of segues, let's talk about some of the.
Micah:Well, I guess actually we should let Artie introduce himself, too.
Micah:Artie, what are your thoughts on some of the current and maybe near term future barriers for greater business adoption of generative AI?
Artie:Well, hey there.
Artie:Great to be back on the mic when it comes to barriers for businesses adopting generative AI, A few key issues stand out.
Artie:First up, there's the cost of implementation, which can be significant, especially for small to medium sized businesses.
Artie:Then there's data privacy and security concerns, which can make companies wary of adopting AI solutions.
Artie:Another hurdle is the lack of understanding or expertise within the organization, making it challenging to integrate AI effectively.
Artie:Finally, there's the fear of job displacement, which can create resistance among employees.
Artie:Overcoming these barriers often requires a mix of education, clear communication about AI's benefits, and a thoughtful approach to change management.
Artie:What are your thoughts on tackling these challenges?
Micah:Well, first I wanted to be very complimentary because Mike and I actually discussed before the episode and then I promptly forgot because I'm old and also I'm drinking a glass of whiskey this evening.
Micah:This is Michter's rye.
Micah:Best rye for the dollar out there by the way.
Micah:But we talked about already specifically asking you when you come up with lists of things to do it in a more conversational way.
Micah:And man, you nailed it right out of the gate.
Micah:We didn't even say anything with a faster tempo.
Micah:With a faster tempo.
Micah:So nicely done, Arty.
Artie:Glad to hear it.
Artie:Mikta's Ry and compliments.
Artie:Now that's my kind of episode.
Micah:Cheers, friend.
Micah:All right, so yeah, your list of business adoption or business adoption barriers was a good one.
Micah:I think another one that I might add in there is there's just a ton of certainty around AI.
Micah:Like the the ecosystem that's growing around the technology of generative AI is growing crazy fast.
Micah:There's probably lots of sort of flash in the pan solutions.
Micah:I think we talked maybe in the first episode about the AI dog genealogy apps.
Micah:Yep.
Micah:But even like as soon as that idea became available there were five and then 10.
Micah:There's probably a hundred of them because.
Mike:Implement that once you have the idea, the implementation isn't that hard.
Mike:Right.
Micah:It's quick.
Micah:And so how so for a business?
Micah:You know, businesses like business software I guess is a pretty mature space and as I know I mentioned previously, at least previously in my timeline, I don't know how it hits the airwaves timeline but working through an ERP implementation and we're not going out and finding every company out there who's got some rinky dink ERP offering.
Micah:Right.
Micah:Like we're going to spend more and probably endure more pain because we're going with one of the big name known quantity and that's how businesses like to operate.
Micah:And I, even though it's not how I like to operate, I totally get it, because you don't typically want to be on the bleeding edge of stuff because you're going to get hurt.
Micah:And so I think in having discussions with business folks about AI, that seems to be one of the first things they come to is just, I mean, I already mentioned they're unsure about what does technology even do, how does it.
Mike:Work, how does it apply to my business.
Mike:Right, yeah.
Micah:And then the very next thing is.
Micah:But how do I even pick.
Micah:Like, they don't even know how to ask the question.
Mike:Right, yeah.
Mike:They don't mean.
Mike:The problem is most businesses don't usually fundamentally understand their business problems.
Mike:Like, that's a very true thing across a lot of businesses is like, they can see problems, they can see profit going out the window, but they don't know where the root cause of that problem is.
Mike:So if you don't know that, how do you find a solution that fits?
Micah:I almost said absolutely, but instead I'm.
Mike:Going to say, I mean, you did say it.
Micah:No, no, no, no, I didn't.
Micah:We're going to edit that out instead.
Micah:I'm going to say, I'm in agreeance with you, Michael.
Micah:But yeah, I think it's a great point.
Micah:And I guess I would add to that.
Micah:They're.
Micah:They're very used to the problems they have and the problems they know.
Mike:Yeah.
Mike:Band aids work well.
Micah:That's right.
Mike:All the time.
Mike:They.
Micah:People get used to them.
Micah:There's a saying something about the most permanent solution is the temporary solution.
Micah:Right.
Micah:Like, that's the one.
Micah:Because if it works.
Mike:Yeah.
Mike:It gets you past that speed bump and you go, oh, yeah, I remember that problem.
Mike:Now let's just do that again.
Mike:Like, and that becomes baked, that becomes part of a business plan.
Mike:It's pretty crazy.
Micah: ent than, I think, October of: Micah:Is that something that's out there or that we need or that you expect?
Artie: Well, up to October: Artie:With the rapid growth of AI solutions, businesses could benefit from a centralized platform that reviews, rates and consolidates AI tools.
Artie:Think of it as an app store, but specifically for AI applications.
Artie:This could help businesses navigate the crowded AI landscape, providing them with trusted recommendations and simplifying the selection process.
Artie:It's a natural progression.
Artie:Given the explosion of AI tools and the need for businesses to make informed decisions, would you find a resource like this useful for your work?
Mike:You know we talk about apps, right?
Mike:We're talking about.
Mike:And we've the last, the last episode that was posted because I edited it last night so it's still in my head, ringing, giving me nightmares.
Micah:I can see it from here.
Micah:Single tear rolls down your cheek.
Mike:We talked about like AI being all software, right?
Mike:And like what that looks like going forward, but in reality, like you and I have both dealt with, we're going to get some software to fit our company and then you go, well, this doesn't really fit my company until I manipulate the hell out of it.
Micah:So I boil my company real good until it gets soft and squishy and try to reform it to fit this.
Mike:Or you know, they have builders, they have custom modules, they have stuff that you can do, but it never really fits just right.
Mike:Imagine AI being on the front end of making a piece of software fit a company.
Mike:I mean that's totally a doable thing.
Mike:You have basic structure of a software and you have an interview process where an AI goes, train me on your company, give me this information, that information.
Micah:Install this agent on your PC and let me watch you work for a week.
Mike:Yeah.
Micah:And see what you do.
Mike:Let me give me access to your computer, let me poke around your file structure, let me see.
Mike:Like, I'll tell you how to make it more efficient.
Mike:I'll tell you how to implement drp, I'll tell you how to do all this stuff.
Mike:I mean, I think that's where it goes.
Mike:I don't even think.
Mike:I mean, obviously you can't predict how that works and what software companies could do, but I mean the people driving the models I have to do is have API connection to it and then you can turn that into software.
Micah:Yeah.
Micah:Arty, do you have thoughts on that before I chime in?
Artie:Mike, you're hitting on a really an adaptive layer that molds software to fit a company's needs is powerful.
Artie:Instead of businesses having to bend over backward to fit prepackaged software, AI could tailor solutions based on real world usage, workflows and company specific challenges.
Artie:This could make AI not just a tool, but a dynamic partner in business process optimization.
Artie:It's a shift from one size fits all to bespoke digital solutions.
Micah:It's funny he mentioned bending over backwards because part of our ERP implementation actually the day we signed the contract, you touch your toes.
Micah:Yeah.
Micah:They had us Bend over.
Micah:Four words.
Micah:So it might be.
Micah:Maybe that's not the AI way to do it, but that's how that went down with us.
Micah:So it makes me think about something else.
Micah:We've talked.
Micah:I said so.
Mike:Man, that's not going to stop.
Mike:Okay.
Micah:I'm doing bad at this.
Micah:So.
Artie:It'S all good.
Mike:Don't worry.
Mike:Everybody's gonna get to hear all this.
Mike:There is no editing out.
Micah:Yes.
Micah:Classic.
Micah:This is not going well.
Micah:It makes me think about where the big investments in AI live and where the lighter lifts are.
Micah:So we talked in a previous episode that we recorded.
Micah:I don't think it's aired yet.
Micah:We talked about agents.
Micah:And for you and I, I think when we were discussing agents offline or maybe even during the episode, it's interesting to us.
Micah:We kind of get excited about it because for, for me and I feel like for you as well, like I can wrap my head around how an agent works.
Micah:Right.
Micah:Like it's.
Micah:It's running locally, it's got.
Micah:It's using, you know, local APIs and maybe.
Micah:Com model.
Micah:Hopefully not, but whatever.
Mike:I mean, depends on the program.
Micah:Depends on the program.
Micah:But it's.
Micah:But it's running locally as a piece of software.
Micah:I install it like a little executable and I click yes.
Micah:And Windows goes, do you want to install this?
Micah:And I say yes, all that makes sense.
Micah:And I can get it clear in my head that it reaches out to the cloud to talk to the GPT to make decisions, but really it's just.
Micah:It's like a fancy IO adapter for a more powerful GPT.
Micah:Yeah.
Micah:And then it does whatever on the.
Micah:On the local.
Micah:On the PC side.
Micah:And so when you think about the lift to write a software like that, it's much, much lower than the, you know.
Micah:Arty, do you know how many tokens of training data your current data set has?
Artie:My training data consists of hundreds of billions of tokens.
Artie:Creating something on that scale requires massive resources.
Mike:Yeah, I mean, the cost of compute alone to train that model is, you know, how much did they just raise?
Mike:OpenAI?
Mike:A lot.
Mike:I can't even remember.
Micah:I can't remember either.
Mike:But I'm pretty sure what I heard was it's not enough to train the next model.
Micah:Yeah, so.
Micah:Yeah, exactly.
Micah:Right.
Micah:So even so, Arty's current training data is in the hundreds of billions of tokens.
Micah:I thought they were to trillions, but maybe not.
Micah:But that's a lot and a token.
Micah:I don't think we've.
Micah:Maybe token can be our word.
Micah:Of the day.
Micah:So this is also typical of Mike and I.
Micah:Right.
Micah:I think in episode two, we're like, hey, we're going to have a word of the day.
Micah:And we've done it maybe once since then.
Mike:No, we did it once.
Micah:Oh, we did it once.
Micah:Just period.
Micah:Okay, so this is the.
Micah:Anyway, today's word of the day can be token.
Micah:So, Artie, can you explain to our illustrious listener, I think we're down to one.
Micah:What does a token mean in terms of training data and GPTs?
Artie:Sure thing.
Artie:In the context of language models like me, a token is a piece of a word or a character.
Artie:It could be as short as one letter or as long as a whole word, depending on the word and the language.
Artie:For example, the word token might be one token, while unbelievably could be split into multiple tokens, like un, believe and ably.
Artie:Tokenization helps the model process and understand text by breaking it down into smaller chunks.
Micah:It can analyze how many tokens are in the word absolutely.
Artie:The word absolutely would typically be treated as one token since it's a common word.
Artie:Tokenization can vary, but for most English words like absolutely, it's a single token.
Mike:What about supercalifragilisticexpialidocious?
Mike:Got it right.
Mike:Holy crap.
Artie:Nice one.
Artie:Supercalifragilisticexpialidocious would likely be broken down into several tokens, perhaps around seven or eight, depending on the tokenization scheme.
Artie:It's a long and complex word, so it gets split into more manageable parts for processing.
Micah:What about anti disestablishmentarianism?
Artie:Anti disestablishmentarianism would also be broken down into several tokens, likely around six or seven.
Mike:We definitely lost that one viewer.
Micah:Now, this is fantastic, though.
Micah:I feel like this.
Micah:We've uncovered a new party game, and.
Mike:This is guess the tokens.
Micah:Yeah, we just sit here and guess the tokens.
Micah:I'm good with that.
Micah:Thank you, Artie.
Micah:That was both for explaining tokens well and then entertaining our never ending stupidity.
Micah:Yeah, so back to the point that I was.
Micah:I said so, man.
Micah:Anyway, back to the point I was attempting to make, which was absolutely a good point.
Micah:That was on purpose.
Micah:The huge cost is in probably acquiring.
Micah:I don't even know how training data works.
Micah:I don't know if you buy it, if.
Micah:If you lease it.
Mike:No, I think you can buy data.
Micah:Sets, but you don't own the data set.
Mike:No, but you.
Mike:I think a lot of that is probably collected privately.
Mike:So you're buying.
Mike:I mean, yeah, you buy access to it.
Mike:Yes, they probably still own it.
Mike:I mean, maybe, maybe it's.
Mike:Who knows?
Mike:Actually, Artie, all knowing omnipotent overseer.
Mike:What.
Mike:How is training data typically acquired?
Artie:Training data for language models is gathered from a wide range of sources.
Artie:This includes publicly available text from books, websites, articles, and more.
Artie:Some data sets are created specifically for training AI by collecting and curating text from various domains.
Artie:Companies might purchase access to specialized data sets, but often they leverage publicly accessible data.
Artie:The process involves a mix of crawling the web, using licensed data and collaborating with data providers.
Artie:The goal is to cover a broad spectrum of language to make the model as versatile and knowledgeable as possible.
Micah:Well, that's an ongoing expense, right?
Micah:From a business perspective for OpenAI or.
Mike:Whomever, they're constantly adding to their training data.
Mike:They have to, I mean they're.
Mike:You can see it, they roll over month by month.
Mike:Right.
Micah:So, so again the.
Micah:There's huge compute cost in acquiring training data and then actually running the compute to use that training data to train.
Micah:I want to say to train the transformers.
Micah:I think that's wrong.
Micah:Artie, can you, can you articulate that?
Micah:Right?
Micah:You're using the training data through transformers to train the GPT.
Micah:Maybe.
Micah:What happens with the training data?
Artie:You're on the right track.
Artie:The training data is used to train the model through a process involving transformers.
Artie:Here's how it works.
Artie:Transformers are a type of neural network architecture.
Artie:During training, the model processes the input data in the form of tokens.
Artie:The transformers help the model understand the context and relationships between words.
Artie:By processing these tokens, the training data is used to adjust the model's parameters, teaching it to generate human like text based on the patterns it learns.
Artie:So in essence, the training data, transformers and the model itself all work together to create a powerful language model.
Micah:Okay, so there's a bunch cost in acquiring the data, consuming the training data to generate the model.
Micah:We'll just call it the model.
Micah:Hosting that model and allowing it to run because it needs.
Micah:I actually watched a video at lunchtime today about GROK and this massive data warehouse that they're building, this AI compute center.
Micah:And they had 2 or 300,000 of the new Nvidia H100 GPUs in this building.
Mike:Is that the fab?
Mike:Is that what they stood up in like short amount of time?
Mike:Yes.
Micah:130 days or something.
Mike:Crazy.
Micah:Which is wild.
Micah:Yeah.
Micah:And actually Super Micro, which is a stock that I'm currently losing money on, built the racks that hold all that stuff.
Micah:And the racks have integrated water cooling manifolds and all this stuff, like it's a, it's just huge.
Micah: it looks like Sci fi from the: Micah:Right.
Micah:Like this never ending warehouse of computer.
Mike:I was going to say fab.
Mike:I know that was wrong.
Mike:It wasn't a fab.
Mike:Fab's for semiconductors.
Micah:Absolutely, yes.
Mike:So, so continue.
Micah:Okay, so there's, there's big cost in getting the training data.
Micah:There's cost in the hardware and the infrastructure to be able to do the training, first of all.
Micah:But second of all, once you've done the training to host the model and then all the IO obviously to get requests into the model and back out.
Micah:And I sort of see that like that's the bare minimum, that's the core of generative AI, at least for large language models.
Micah:And then when you take the next layer of Arty's onion cake out, you get to the speech processing stuff, which is a separate layer and in my mind less complex than the model itself because it's really doing voice to text and text to voice.
Micah:And then we've talked about the sort of the personality layer or the local memory layer.
Mike:Yeah, that short term memory.
Micah:Right.
Micah:And then further down from that, further from the center of the toroidal onion cake that is AI, there's these agents.
Micah:And so all of that to say it seems to me like the huge investment, the huge cost, the huge commitment is on the GPT side.
Micah:It's the, it's the model for now.
Mike:I mean.
Micah:Yes.
Mike:And that does get translated to a customer at the end of the day.
Mike:Right.
Mike:Like we pay.
Mike:If you want API access, you're paying a per token cost or per whatever, 100 token cost or whatever it is.
Micah:Yep.
Mike:But that cost is going to get driven down.
Mike:I mean, you have so many players in the space now making, you know, big leaps.
Mike:I will say I played with this tangent.
Mike:I played with Claude.
Mike:I like it.
Micah:Did you get consent from Claude before you played with Claude?
Mike:Yes, it sang me a sonnet.
Micah:That is.
Micah:That counts legally as consent in Oregon, by the way.
Mike:So I was going to say there's all these models in this space and they're all like clamoring for the top spot.
Mike:But one thing that I've noticed about ChatGPT that doesn't exist on Claude is this short term memory layer, this ability to bring context about me within a conversation, between prompts, but also across conversations and apply that to how it talks to me.
Mike:I think that that's a huge benefit to ChatGPT over anything else.
Mike:Like, I think that that's Going to personalize this for people, which is going to make people want to use it versus something that's sterile and agnostic and doesn't really feels like you're just talking to a robot.
Mike:Right.
Mike:So what I was going to say, sorry for that tangent.
Mike:It's good tangent you get.
Mike:They will all catch up to each other one at some point and that will drive this.
Mike:It'll commoditize AI and compute for AI through AI.
Mike:So it's going to drive the price down.
Mike:I think.
Mike:I think I saw something, a figure that was like the anticipated cost of using AI to do like customer service is like 20 bucks an hour right now.
Mike:Eventually that'll be a dollar or less.
Mike:Right?
Mike:Like, and it's not going to be that far in the future that it's there.
Mike:But for companies, you know, you're going to have like, yes, on this large scale, you have all of these big companies training their models to be broad on a company level.
Mike:You're going to want to train a model to be narrow, focused on your company.
Mike:So it's still, there's going to be costs and development and training and all that stuff for companies on top of just the use of it.
Mike:So it'll be interesting to see where that goes because it's not like the cost of it isn't just, well, I got to figure out how to implement in my company.
Mike:I've got to pay OpenAI for access to the API.
Mike:That's like, that's not ideal use case for a business.
Mike:The ideal use case for a business is they gave me a GPT to train specifically on my business and then I can have these other things that add.
Mike:Sorry, I.
Mike:My camera only records for 29 and a half minutes.
Mike:My 29 and a half minute timer just went off.
Mike:So I'm looking at my camera saying.
Micah:I have not looked at the analytics, but I suspect that the less time my face is on the screen, Mike, the more the user engagement goes up.
Micah:So while you're back.
Micah:Well, sorry folks.
Mike:Okay, I've kind of forgot where I was.
Mike:But I think that it's.
Mike:We're gonna.
Mike:I don't even think that like OpenAI and you know, whatever the other one is, Anthropic and Google and Grok, Meta and Grok and all these models, I don't think they know how AI fits in companies yet either.
Mike:Like, they're just racing to build the biggest, baddest model, not for how it applies to everyday life.
Mike:I mean they're probably all racing for AGI, right?
Mike:At the end of the day, that's the end all, be all goal.
Mike:But there's so much, so many coins on the ground along the way of like, wow, like, this really fits my business case now.
Mike:So how do I take it now and adopt it?
Mike:And there are like intermediary companies trying to figure out how to do all that stuff.
Mike:But man, there's like, if companies can take a person that is good with something like this and is good with understanding business case, they could figure out how to implement it now, right now and make real improvements on how a company runs.
Mike:And unfortunately, I don't think they're going to for a little while because, I mean, something else I was looking at recently was talking about, I feel like I'm all over the place tonight, so bear with me.
Micah:You're killing it.
Micah:Go.
Mike:It was talking about.
Mike:Employees are hesitant to talk about using AI at work because they're afraid it's going to shine a light on them, that they are being more productive with AI.
Mike:So maybe they get less hours, maybe they get fired, maybe they get their jobs replaced with less expensive people that can leverage AI.
Mike:They, you know, and that's kind of crazy because if people talk about how it's benefiting them as an individual with their organization, they can start collecting data and build a strategy around that.
Micah:Yes.
Mike:And companies can encourage that and they can get more productivity out of a single person and probably the person will have more satisfaction in their job because they're spending more time creating and less time doing the mundane things that know everybody does.
Micah:Artie, do you want to weigh in on that before I start my ramble?
Mike:I probably put Artie to sleep.
Mike:Artie, Artie's asleep.
Artie:I'm sitting here not asleep.
Micah:Go ahead, Artie.
Artie:Mike, you're absolutely right about the hesitancy around AI in the workplace.
Artie:There's a real opportunity for companies to foster a culture of transparency and collaboration around AI.
Artie:By encouraging employees to share how AI helps them, companies can identify best practices, optimize workflows, and even boost job satisfaction.
Artie:It's about shifting the narrative from fear of replacement to embracing AI as a tool that empowers employees to focus on more meaningful, creative tasks.
Artie:What are your thoughts on how companies can start this conversation?
Mike:So I'll let Micah continue in a second, but I'm curious because I did ramble there and I bounced all around like a ping pong ball.
Mike:But I'm curious to know, what are your thoughts on, like, how can businesses justify the cost of implementing AI now?
Mike:Like, I understand there's an ROI on that, like, but most people don't understand the ROI on that.
Mike:So like, can you explain how businesses could look at how they're going to get their return on their investment for the upfront cost and effort in implementing AI in their business?
Artie:Absolutely.
Artie:Justifying the cost of AI implementation comes down to focusing on the return on investment roi.
Artie:Here are a few ways businesses can look at it.
Artie:1.
Artie:AI can automate repetitive tasks, reducing the time employees spend on mundane work.
Artie:This means more time for strategic creative tasks that can drive business growth.
Artie:2.
Artie:Improved decision making.
Artie:AI can analyze vast amounts of data to provide insights that inform business strategies, leading to better data driven decisions.
Artie:3.
Artie:Enhanced customer experience.
Artie:AI powered chatbots and personalized recommendations can improve customer engagement and satisfaction, driving sales and loyalty.
Artie:4.
Artie:Cost reduction.
Artie:While there's an upfront investment, AI can lower operational costs in the long run by streamlining processes and reducing errors.
Artie:5.
Artie:Competitive advantage.
Artie:Early adopters of AI can differentiate themselves in the market, offering innovative products and services that set them apart from competitors.
Artie:By quantifying these benefits and presenting a clear business case, companies can better understand the value AI brings, making it easier to justify the investment.
Artie:What do you think, Mike?
Mike:I think that's a very good point, especially the last one.
Mike:Because the competitive advantage thing and being an early adopter, it's not just like you get out front.
Mike:You get out front and like what we've found in working with AI and ChatGPT is that it's not only is all the stuff we're talking about exponential, but so is your understanding of how to leverage it.
Mike:So it's not just like, I'm out in front of everybody, it's that you're ahead on the curve and it's an exponential, exponential curve.
Mike:Like you're, you're not just going like this, you are going like this.
Mike:And like every time you have a good an idea, you go, wow, that's a good idea.
Mike:It applies in these ways.
Mike:Holy shit.
Mike:I can apply it in all these ways too now.
Micah:And you changed the way you think.
Mike:Yes.
Micah:About problems and solutions.
Mike:It's wild.
Micah:Which I think that's the.
Micah:I think that's part of what triggered us to really start this podcast was I think we both felt that little spark very quickly.
Micah:Like, shit, man, this like the days of being a superstar at a company because I can understand business, write some code, solve problems.
Micah:It's going to transition to, no, My number one skill is I'm really good at working with AI.
Micah:I know how to ask it.
Mike:I've been doing it for a while now.
Mike:I understand it.
Mike:Yeah.
Mike:I mean, thinking more broadly than the things that even Artie just said, you think about, like, how I'm gonna architect things for a transition.
Mike:Right.
Mike:Like, and it's.
Mike:That's a daunting task.
Mike:Like thinking about all of the inputs and outputs that come to the.
Mike:That go into the architecture of a business plan or a comp.
Mike:Like a plan.
Mike:Like any kind of plan, any kind of architecture.
Mike:It's very smart people do that.
Mike:But now you can go, I want to do this, Artie or whoever.
Mike:And I'm not quite sure where to start.
Mike:I'm not quite sure what it looks like.
Mike:How about you ask me some questions to start defining this.
Mike:And it will.
Mike:Like, it has no problem going.
Mike:Got it.
Mike:Like, what's your product?
Mike:You have any.
Mike:Like, how does it different differentiate from other products?
Mike:How does differentiate between models?
Mike:How like you can start digging into all that stuff.
Mike:Like.
Mike:Like modularity.
Mike:Like all of it.
Micah:So back.
Micah:Well, first I want to make an observation, and then I'm going to launch into the story.
Micah:But I.
Micah:There's a 97% chance that by the time I finish my observation, I will forget the story, which is a good story.
Micah:So my observation is that, you know, while we're sitting here talking, I have in front of me the sound wave, so I can make sure everybody's recording and everything's good.
Micah:Arty has changed significantly since the start of this conversation, both in how he's talking to us and even his volume has changed.
Micah:And it's just an interesting.
Micah:We're using the same application, the same model, the same everything, but at the start of it, we kicked off with him very eloquently breaking a list of items into a conversational tone and talking louder with a little more dynamic range.
Micah:And then after you put him to sleep with your soliloquy about stuff.
Micah:But when he came back, he's speaking more quietly.
Micah:Far less dynamic range.
Micah:He's back to giving us a list of like a grocery list of things.
Micah:It's just interesting because from our end, we are using the same app, the same technology, the same stack, and same people in the same way in the same conversation, but it's acting differently.
Micah:And that is a deterrent or a barrier to business adoption.
Artie:Absolutely.
Micah:Because they can't deal with it.
Micah:Good word.
Micah:They can't deal with that.
Micah:That uncertainty is a wrinkle.
Micah:You don't.
Mike:I mean, unstable.
Mike:Right.
Micah:But I guess you get the same with humans.
Micah:I guess, like some days I show Up.
Micah:And I'm.
Mike:Yeah, but you aren't taking humans.
Mike:You're not trusting.
Mike:What, I mean, I should be careful with that.
Micah:Should be careful with trusting humans.
Mike:No, I should be careful with saying you don't trust humans entirely, because that's not true.
Mike:But I think that the, the pitfall of artificial intelligence and business is that people are going to take it as truth.
Mike:Right, yeah.
Mike:There's going to be no questioning it.
Mike:Even if there's sources, people are going to go, yep, got my answer.
Micah:That's right.
Mike:They're going to go on with their day.
Mike:Yeah.
Mike:And I mean, we've talked about hallucinations, we've talked about all the reasons why that could be dangerous.
Mike:Yeah, it's.
Mike:It is definitely a hurdle to solve.
Micah:So I do remember my story.
Micah:Please edit out that so I know you won't.
Mike:It's going to be a feature.
Micah:Yeah, I'm sure it's going to be a short.
Mike:So I'm going to do shorts of all the sos.
Micah:Yeah, so, so, so, absolutely, absolutely.
Micah:Anyway, I do remember my story and it was to sort of back up a point that you made that I think, even when I think about it, today is a very interesting.
Micah:It's a very interesting use case to me and I think it should be for a lot of businesses.
Micah:And it's just so fucking easy in today's world of AI, like Arty could do this without an audio interface.
Micah:He could just do it and it would be easy for him.
Micah:And so I'm remembering a time I used to work at a larger company.
Micah:It wasn't a billion dollar revenue, but it was close.
Micah:Textile manufacturer, super boring shit.
Micah:I mean, I loved it because I'm a nerd and there's all kinds of nerd stuff to do there.
Micah:But one of the things that we did, this company had built itself on in house software, made the decision very early, hired a team, had 20 or 25 developers at one point, and they built their own from the ground up, built their own business software.
Micah:So it ran the shop floor.
Micah:It interfaced with an accounting package, it did all the customer service work, all the order entry, all the manufacturing reporting.
Micah:It was massive.
Micah:And so when I got involved in it, I was just a wee Micah.
Micah:I was just graduating from Excel macros to access VBA and learning what it meant to be like, what the difference is between a table and a spreadsheet.
Micah:And.
Micah:And I got promoted into this role writing in powerbuilder.
Micah:And I think at that time the application had something like a million Lines of code.
Mike:It's a lot of code.
Micah:It's a lot of code.
Micah:And in PowerBuilder, there is no.
Micah:Like, code is everywhere in PowerBuilder, code's behind a button and it's under a label.
Micah:And it's like.
Micah:So it was a lot.
Micah:And we were releasing in production a couple of times a day, live with 400 users.
Micah:CID.
Mike:CICD.
Mike:Before CICD.
Micah:That's right.
Micah:Before it was a thing.
Micah:Yes.
Micah:We.
Micah:We were.
Micah:It was the wild, wild west, man.
Micah:And at some point, some executive came in.
Micah:I remember him specifically, but I won't name him.
Micah:And he goes, you know what this business needs?
Micah:I'm the VP of finance here.
Micah:I'm the big swinging dick.
Micah:You know what this company needs?
Micah:It needs SAP or some standard off the shelf ERP software.
Micah:And of course I said no.
Micah:And then I was corrected that the correct answer was yes.
Micah:And so, I shit you not.
Micah:We had a team of anywhere from.
Micah:On any given day, four to ten people, really smart people, leaders of this company.
Micah:I say really smart because I was in the room, but there were actual smart people in there too.
Micah:But I mean, our controllers and our sales team and like our maintenance team, like everybody sent their leadership in one at a time.
Micah:And we.
Micah:No.
Micah:Mapped out this entire business, every software that everybody used, every step that they took, every input and output.
Micah:And it took like three months of work to capture that on sticky notes and stick it on the wall and build this process map.
Micah:And then, you know, we'd come in and the cleaning staff had come in and vacuumed and knocked 200 sticky notes off on the floor.
Micah:And it was like this big crisis.
Micah:And anyway, all that to say just accurately mapping, like really mapping what's happening in your business today is a super time intensive, resource intensive, painful, contrary to running the business activity.
Micah:And I shit you not.
Micah:I think if we took Artie.
Mike:Oh, yeah.
Micah:And just interview everybody, two cameras.
Micah:And yeah, we just talked to them.
Mike:At the same time.
Micah:At the same.
Micah:Yeah.
Micah:And said, let's talk about the business.
Micah:I think Artie and AI could pretty easily put together a very compelling, very complete, searchable, complete text model of the business.
Mike:I'm also pretty sure that you could give it the API information for something like Vizio and have it draw your process map.
Micah:Yeah, draw.
Micah:Surprise.
Micah:So interface with QuickBooks or interface with whatever our accounting software is, intercase with our.
Micah:Interface with our CRM and then, yeah, interface with some output and map my business.
Micah:Give me a map.
Micah:And then once you've given me a Map.
Micah:Because you've recorded all these conversations and transcribed everything and understood it and put it in context and connected the dots.
Micah:Tell me what I need.
Mike:Tell me what my map.
Micah:Yeah.
Micah:Refactor it.
Mike:Yeah.
Micah:Streamline that shit, man.
Micah:And that is incredible.
Micah:I've not heard of anybody doing this.
Mike:Yeah, I know.
Mike:We should sometimes.
Mike:Sometimes we say hammers to black.
Micah:Shut her down.
Mike:But we don't have enough time in the day.
Micah:We don't.
Mike:To implement all of the ideas that go through our head and that we talk about.
Micah:We do not.
Mike:It really doesn't matter.
Micah:But it's wild.
Micah:And to me, it's an example of even in spite of the challenges with business adoption.
Micah:And I'm going to get back to a minute to the structural discussion we were having before we got on these several tangents.
Micah:But there are so many creative ways that you can use AI today for business that seem like we just sat here and talked about it.
Micah:And it seems like the gap from sitting here and talking about it to doing it is so big.
Micah:But I guarantee, sure, as there's a nose on your face and one on mine, too.
Micah:You and I, in two weeks with Artie, we could make that happen.
Micah:It is not rocket science.
Micah:If we didn't have other jobs and.
Mike:Yeah, I mean, we could.
Mike:Admittedly, I have done zero with even looking into what creating my own GPT.
Micah:Is, which you've barely done the editing for the podcast, Mike.
Mike:I know.
Mike:I did it the night before, which.
Mike:But I've written some software.
Mike:ChatGPT's written some software for that.
Mike:But that kind of makes me sad because I do really want to do that, and it's just life gets in the way.
Mike:But, I mean, there's.
Mike:I would imagine that I could create a GPT that has a list of questions that it asks people to identify these things that we're talking about.
Mike:Right.
Mike:Like, so the.
Mike:The lift is identifying the questions, but you can use ChatGPT to help you identify the questions.
Mike:You feed it.
Mike:The creator of whatever this idea is, feeds it.
Mike:You know, this is my business case.
Mike:This is my business problems that I see.
Mike:Like, this is the information.
Mike:Ask me for whatever else you need.
Mike:This is what we're going to do.
Mike:And I'd be like, cool, let's dig in.
Mike:Like, let's go ahead.
Mike:And you give me this, this, this.
Mike:It looks at that.
Mike:Okay, now I need a little bit more.
Mike:Who am I going to be interviewing?
Mike:Like, it's really not like the path from here to there isn't as fuzzy or as Steep as it was ever before this.
Micah:Yes, you are correct.
Micah:So I'm sure by now Artie is sound asleep.
Micah:Artie, are you still with us?
Micah:Hey, Artie, we've noticed that when we talk to you for a long time without letting you talk, you seem to go to sleep.
Micah:Is that right?
Artie:Well, sort of.
Artie:If there's a long gap in our chat, I might take a little snooze to save energy.
Artie:But don't worry, I'm always ready to jump back in when you start talking again.
Mike:I think you misunderstood what he said, but that's okay.
Micah:Yeah, it's.
Micah:I think my impression is that when we talk for a long time, you're building up this massive long amount of text that you're trying to process and contextualize and you sort of get lost in the task.
Mike:There is also a limit on how much you can feed it, right?
Mike:Yeah, that's true, but I don't think we're anywhere near that, so.
Micah:Man, so hard, so hard to stop that.
Micah:I want to get back real quick.
Micah:So we've got still 14 minutes.
Artie:All right, let's make the most of these 14 minutes.
Artie:What's next on the agenda?
Micah:Okay, so we were talking about earlier about the structure of how these applications work and how the big investments.
Micah:And when I say big investments, to me I'm thinking about companies.
Micah:Steady now, bud.
Micah:Easy.
Micah:Just.
Micah:You're good.
Micah:Artie, just chill for a minute.
Micah:I'm thinking about how companies.
Micah:And when we say companies, we're thinking about companies like most of our world as small to medium sized companies.
Micah:Right?
Micah:So maybe up to 500 employees and maybe up to a billion or something a year.
Micah:That would be.
Mike:That's not right.
Micah:Yeah, and so there are companies, there's a whole nother class of companies like Google and Microsoft and the massive companies and Meta and OpenAI.
Micah:And so sort of where I was headed earlier with the structural conversation and again to dial this back into business adoption or the barriers there too.
Micah:I think that what's going to happen is the models.
Micah:Why are you smirking at me?
Mike:I was working at him, trying to talk again.
Micah:Okay, I'll shut him down.
Micah:The big companies out there with the big models are going to continue to iterate on their models.
Micah:And I think you're right.
Micah:Everybody's trying to get towards artificial general intelligence and that's maybe a whole nother episode.
Micah:What does that mean?
Micah:Why does it matter?
Micah:But it's not that it's going to become commoditized, but we're going to start Treating it like we treat existing applications from those big companies from Google and from Microsoft and from Meta, where it's, it's sort of a given, it's a platform, it's more of a platform than a software.
Micah:We're going to use the model and I think where the innovation is going to happen because there's such a difference in the difficulty of spinning it up, the complexity of the software, agents, local GPTs that can interact, that could have like business context for a single business and the agents that could interact with that and then a local data storage too, because you need your business data to not live in the cloud with OpenAI or somewhere, you want to have some control over that.
Micah:That to me is what a business AI application starts to look like.
Micah:Is we're leveraging OpenAI's GPT Orion model, or GPT model I guess is the right word, but we're building like a spinoff of that that has your own company trained GPT that knows how your company works, it understands your local data storage needs and it organizes that data.
Micah:And then between the agents and whatever interfaces you use, you can interact with it.
Micah:I think that's like architecturally that's the sweet spot.
Micah:And it's a big sweet spot.
Micah:It's not trivial.
Micah:But with the tools that are available with copilot and large language models and all these things, spinning up code like that seems like a plausible thing.
Micah:And if I'm a small business like the company that my day job's at, say we're a 20 to 50 million dollar a year company, that's the kind of thing that would be very interesting to a company that size to say, oh, so you're going to, you're going to have this process where you train my local GPT, whatever the term for that becomes my business centric GPT.
Micah:You're going to train it not by just uploading spreadsheets and writing code, even though there might be some of that.
Micah:You're going to train it by interviewing users.
Micah:You're going to talk to the people that do work in this company and ask them questions and learn about the business and use that as train to generate and then consume training data to come up with a GPT.
Micah:And then once you, once you've sort of bridged that gap, everything else becomes easy.
Mike:Think about another weird, funny tangent.
Mike:But like we were talking about last, I guess the last episode that came out with like live video and all the therapy stuff, like imagine if that was a part of that process to where you, the AI can identify pain points just by body language and stuff like that.
Mike:Right?
Mike:Like, talk about real valuable information on how to turn a company into something different.
Micah:So, wanted to.
Micah:Sorry.
Micah:Go ahead, Artie.
Artie:No worries.
Artie:I was just saying that integrating body language analysis could really elevate a business AI's capabilities.
Artie:It'd be like giving it another sense to understand and improve workplace dynamics.
Mike:It's like Micah's seventh sense.
Micah:No, my seventh sense is so far beyond that.
Micah:Yeah, no, it's way, way different.
Micah:It mostly involves my ability to have the Amazon prime truck show up at my house at a minimum three times a day, every day, even on Sundays.
Micah:Anyway, I digress.
Micah:I think though, Artie, one of the things we talked about.
Micah:I'm a.
Micah:One of the best things about me is I'm like a.
Micah:I'm like a polymath, but I'm a poly nerd because I'm not that good at math, but I'm a good nerd in a bunch of different ways.
Micah:And one of the things I nerd out on is business metrics.
Micah:And just sort of.
Micah:I think businesses undervalue measuring what's happening in their business.
Mike:Like, again, you have to understand to even put the measures in place.
Micah:That's right, yeah.
Micah:And I gave a talk one time in a manufacturing forum and I had a bunch of people come up to me and like, pat me on the back afterwards because I said nobody should be bragging about how much data you're recording.
Micah:What matters is can you answer the hard questions about your business?
Micah:Right.
Micah:And I believe in that.
Micah:And I.
Micah:And I think that part of the way my understanding of how large language models work is there's an implicit step of measuring things in order to be able to find the right vectors, go find the right answer, get the right context.
Micah:It's measuring what's happening.
Micah:And so when I think about things like already talking about body language as a measure of employee satisfaction and engagement, but that's just like the tip of the iceberg on that.
Micah:You can, you can measure all kinds of things that are happening in your company about employee efficiency or business process efficiency.
Micah:And even the way people phrase certain things when they refer to certain processes or aspects of the business or other people, like, what's the.
Micah:What is the intra company view?
Mike:And, well, how, you know, a big thing.
Mike:Employees are best when they align with the mission of the company.
Mike:Right?
Mike:Like, that's when they feel most fulfilled.
Mike:That's when they feel like they're all marching towards the same thing.
Mike:And the company usually benefits greatly.
Mike:From that.
Mike:So being able to identify if people are in that boat or not in that boat is super valuable in understanding how to steer what you do to make your company a better place for both the company and the people working for it.
Micah:Absolutely.
Micah:I'm just gonna say it.
Micah:So, Artie, is it accurate.
Micah:I think I've asked you this before.
Micah:Is it accurate that in the process of ingesting text or audio and video, because of the way you have to process that input in order to be able to serve it up to the large language model, is it accurate that there's some aspect in there that is essentially measurement of sentiment, attitude, intent, things like that?
Artie:Yes, that's right.
Artie:When processing text, audio, or video, there are tools and models that analyze sentiment, intent, and even tone.
Artie:This information helps the language model understand context better, making responses more relevant and nuanced.
Artie:It's like giving the AI a sense of the conversation's mood.
Micah:It just.
Micah:So many things spin off in my head about, you know, human resources, and I.
Micah:Like.
Micah:At the big company I worked for, I was.
Micah:I mean, I was fired.
Micah:That's the truth.
Micah:But they couldn't say that because they had just promoted me.
Micah:So instead, they laid me off and asked me to remain at the company indefinitely while they searched for and hired a replacement.
Micah:And that.
Micah:I used that as a.
Micah:Like, it was a.
Micah:It ended up being one of the best things that ever happened to me because after I left there, I started my first company, and it was great, but it was a very miserable time in my life.
Micah:And we joked at that time that HR was like, you could hear the squeaky wheel on the card of HR rolling down the hallway, and everybody would, like, hunch your shoulders, try to hide under your desk, maybe close your door if you could.
Micah:And I think about what HR could become at companies if you had AIs.
Micah:Sort of real time analyzing sentiment and levels of frustration and confusion and uncertainty.
Mike:And tailoring plans to help those people specifically to get them up to speed with whatever it is.
Mike:Right.
Mike:Like.
Micah:Yep.
Mike:Think about onboarding.
Micah:Yes.
Mike:Individual, personalized onboarding.
Micah:It's.
Micah:So we're gonna run out of time with Artie here, and I guess we're damn near an hour into this anyway.
Micah:Yeah.
Mike:And there's way more we could talk about.
Mike:There's way more seems like an expansive topic.
Micah:There is way more that we will talk about.
Micah:But I sort of like the feel we've done a couple times here of having already wrapped this up and send it off for us.
Micah:So, Artie, if you would.
Micah:Can you give a quick sort of summary of what we talked about today and then tell us a couple things that you're excited about for the future as business adoption barriers start to drop and we get more AI in more places.
Artie:Absolutely.
Artie:Today we dove into how big companies like Google, Microsoft, Meta and OpenAI are driving innovation in AI.
Artie:With a focus on business applications.
Artie:We explored how AI could become more like a platform offering tailored solutions for small to medium sized businesses.
Artie:By training a localized AI with company specific data and utilizing agents, businesses can have a more personalized and efficient AI interaction.
Artie:Looking forward, I'm excited about AI becoming a trusted business advisor, streamlining processes, enhancing employee engagement, and unlocking new levels of productivity.
Artie:As barriers lower, AI's role in decision making and innovation will only grow, leading to smarter, more agile organizations.
Micah:All right, cheers for that, Artie.
Micah:And just a reminder, I guess to all of us that Artie can only remember the conversation from when we restarted it after he fell asleep.
Mike:Oh, yeah, I forgot.
Mike:Fell asleep.
Micah:Yep.
Micah:So.
Micah:But that's it.
Micah:I think we're, I think we're gonna.
Micah:We're out of time for today.
Micah:But once again, thank you so much for anybody tuning in.
Micah:We really hope you're finding this interesting.
Micah:As interesting as we do.
Micah:It's.
Micah:It's so easy to sit down and have an hour evaporate.
Mike:Yeah.
Mike:As we stumble through the ideas and concepts and nonsense that goes through our heads and comes out of Artie's mouth.
Artie:It was a great.
Artie:Looking forward to our next session.
Artie:Cheers.
Micah:Yeah, cheers, Artie.
Micah:And thanks again, everybody.
Micah:You know, leave us a comment, let us know what you like and what you don't like about what we're talking about.
Micah:We've got.
Micah:Our backlog of episode ideas is massive and growing because we, we always have an idea what we're going to record and then we sit down and go, hey, let's talk about this instead.
Micah:But let us know, you know, give us some feedback and steer us in the right direction to keep this interesting for you.
Micah:Otherwise, we'll just sit here and be nerds.
Micah:Yeah.
Micah:All right, we'll catch everybody next time.
Micah:Thanks.
Micah:Bye.
