← The Good Question Podcast
The Good Question Podcast
AI for Business Growth Ken McLoud On Practical Strategies & Operational Optimization
00:32:30
Loading summary
Transcript95 lines
- [00:01]
A
You're listening to the Good Question podcast with Richard Jacobs. Our goal is to make each of our guests exclaim, hmm, that's a good question. I don't know the answer. Because when that happens, it means you, the listener, may be inspired to learn more beyond the interview and to ask great questions yourself that lead to new insights. In this podcast, we cover historical and current anthropology, comparative religion and history. Welcome. And let's get started.
- [00:36]
B
Hello, this is Richard Jacobs with the Good Question podcast. I guess today is Ken McLeod. He's the founder of Laconic Tech. We're going to talk about AI software they have that they use for lead generation, how to find business and identify people who may be interested in your product or service. So I think it'll be a very interesting call. So welcome.
- [00:54]
C
Ken, thanks for having me.
- [00:55]
B
Tell me a bit about, you know, your background, get into lead generation and AI.
- [01:00]
C
Yeah, so I'm an engineer by training. I spent most of my career working a 9 to 5 for, for big companies. Started doing this as a side hustle. Software development as a side hustle. Before AI was a big deal. So let's say like 2019ish sort of time frame. Obviously in the intervening years ChatGPT exploded and, and everybody became very interested in AI and that certainly helped the, the software development business. And just about the time that, that the software development was starting to go really well, I got laid off from my day job. I had survived rounds of layoffs at this company and the fourth round got me. So I said, well, that's the time to hold your nose and jump. And I took the AI development business full time and it's been nothing but great since then.
- [01:47]
B
Oh, that's cool. You know, we, I, I've tried to use a lot of a and I in my business and everything and it seems like it is helpful, but it's, it requires a lot of finishing and integrating and things like that. It's, you know, all the videos I see are like, oh, this person made a business in two seconds of AI and they make 20 grand a month. And it's like, I don't know, it just seems like there's a lot more to it than just having it do work for you. You have to really like again, spend time connecting it and integrating it into your workflow.
- [02:13]
C
Yeah, I think utter nonsense would be the polite term for videos like you're referring to.
- [02:19]
B
I mean, you see a lot of them though, right?
- [02:20]
C
Oh, yeah, definitely. No, there's a whole kind of hype industry of, you know, I just Sleep and sit by the pool and sit margaritas. And this AI tool runs my whole company. Makes me a million dollars a month. It's just nonsense. I mean, there's nobody doing that kind of thing. They're just lying. There's certainly lots of people making lots of AI, but not in the way that they're describing there where they're just sitting by the pool and AI runs the whole business.
- [02:45]
B
Yeah, I figured either I was stupid or they're not telling the truth. So what, what areas have you seen? Like what kind of specific applications have you seen AI to be really good at?
- [02:54]
C
Yeah, so we can jump into specifics in each of these, kind of break it down into three categories. The, the first one is the one you mentioned at the top of the podcast here, which is lead generation. So we do that both in AI powered lead magnets. So this is the idea of everyone's famil with sort of an old school lead magnet where you might give away a PDF of the top 10 tips for XYZ. Well, the idea of AI powering this is we can get some information about the prospect either by having them answer a few questions, or maybe in a B2B situation, we scan their website, something like that, and then we can use an AI to tailor our 10 tips specifically to that person's situation. So it's not just the generic 10 tips, it's the 10 tips given what I can see on your website right now, for instance. So that's, that's all the benefits of establishing your expertise and getting people's contact information. It's just much more powerful than a generic PDF.
- [03:53]
B
Those are like a customized website review.
- [03:55]
C
Yep, that's one of them that we've done. We've also done stuff based off of questionnaires. So like say like a five question multiple choice or fill in the blank questionnaire that you kind of learn about somebody's business. And then you can tailor kind of a how to to advice PDF based on your philosophy and the information that the prospect just gave you. We're working on one right now where we're scanning the per. So it's for an ad agency. We scan the person's current ads and then give them advice on how to adjust their ads based on this ad agency's strategy.
- [04:28]
B
Oh, that's cool. I was gonna ask you what are some of the customizations that work well with people? What are some specifics?
- [04:34]
C
Yeah, so the, I think the big value add here is that we want to base the advice on your particular exper. Expertise Right. So like in the case of that ad agency, it's their philosophy for making. They mostly do meta ads for making meta ads perform well. It's not just generic, you know, ask Chat, GPT. How do I make a good meta ad? We're providing actual insider knowledge and experience here to the model in order to give you good actionable advice.
- [05:02]
B
Okay, so what. So ads. I can see, you know, the recommendations there. That would be a good lead lead, you know, lead magnet. What else would be good for, you know, let's some B2C or a B2C. B2B or B, the application. What would be another lead magnet? That would be good.
- [05:16]
C
So like, another one is a gentleman out of Colorado named Jason Swank. So he, he has grown and exited two successful agencies, and now he, he hosts a podcast and he runs a mastermind group kind of teaching people how to grow and then eventually exit their agencies successfully. So he's the one that we're running a questionnaire for. So he drives traffic to a landing page. The landing page asks you a series of questions like, where's revenue at? How involved with you? Or how involved are you with sales? How involved are you with delivery? These sorts of questions. And then we send those answers off to an AI, along with basically a small book that he and his team have written on their philosophy for how to scale and exit an agency. And then you get back a personalized report on sort of according to Jason's philosophy, here's what he thinks he should be doing for your agency.
- [06:13]
B
Here are these reports. So, you know, is robust at this point, where it's not hallucinating. And, you know, do you have to check over the reports or is it accurate enough that you don't worry anymore?
- [06:22]
C
It very, very much depends on the application. So for something like relatively broad strokes business advice, in that example I just gave, we feel fine with the model, just returns the report without a human looking over it. You're very unlikely to get the model saying something absurd there, right? You're very unlikely to. Is saying that you should do ridiculous things. The situations where you've absolutely got to be checking the model are things where you're dealing with very specific bits of data, right? So if, for instance, you were dealing with financials, right, and you had a tool that was moving money around where the exact dollar amount needs to match, you don't want to be doing that with an AI. You want to be doing that with traditional code so that the number goes from, you know, this column of the ledger to that column of the ledger. Exactly. So it depends. You're kind always doing a cost benefit analysis of one. Given what we're talking about here, how likely is a bad screw up and then if a bad screw up occurs, what are the consequences of that bad screw up? Right. If, if one person out of thousands gets a little bit silly advice in a lead magnet, that's not a huge deal. If you're moving money around between accounts and you move the wrong amount, that could be a very big deal.
- [07:38]
B
Right? What about lead generation to replace human callers? I know that inbound is loud, but are people trying to do outbound and what's the of the art inbound? What's it able to do now?
- [07:50]
C
Yeah, so absolutely people are trying to do outbound. And the, the state of the art is I would not use it for high ticket customers yet. So the, the big problem that we have is that they can't handle normal human interruptions in a conversation. So if, if it's ever, we've got voice synthesis down, right? So it can decide what it should say and then it can synthesize a human voice to say that like nearly perfect. The trouble is it doesn't react, interject in the middle of its sentence. It's very stilted. It cuts off a little too late, then it takes too long to recover from that. It becomes instantly apparent that you're not actually dealing with a human. And if up until that point in the call, you had been presenting it as though this was a human, you know, you'd been kind of trying to deceive the person, then as soon as they realize you've been trying to deceive them all, the trust is right out the window, right? Like I get angry in these situations when it's pretending to be a human and then it becomes very clear that I'm not dealing with a human. It feels like, like an insult to your intelligence.
- [08:54]
B
What if you tell people this is an AI system? What percentage will abandon it versus say, all right, fine, I'll try it.
- [09:00]
C
I think it's hugely context dependent, right? So things where we're already used to automated systems, right? Like if you're calling a retail establishment to get the store hours, or you're calling, you're calling a bank number to get your current checking balance or something where you're used to that being a, you know, key in your number here and then an automated voice system will give it to you. I think you'd have no more abandonment than with traditional automated systems, which I'm sure is very Low. But if you have something high, touch high ticket and the person is expecting a human, I think then you're going to be, you're going to have a lot higher drop off rate. Okay, now I should caveat. All that was saying that there's, there's obviously billions of dollars to be made in that field and lots of very smart people are tackling that problem. So what we're talking about right now is the state of the Art in March 2026. All that could be out the window by the end of the year.
- [09:54]
B
Well, what do you see that voice is best for in terms of, you know, lead generation and chatbots and all that?
- [10:00]
C
Yeah. So I don't have a great application for AI voice in lead gen right now. The place where I use AI voice the most is I'll use it in the chat features within the, the big AI providers in order to multitask. Right. So if you have something that you need to ask AI about and you're in a situation where it would be inconvenient to be staring down at your phone, like let's say you're driving, you can put it in the voice mode and talk to it. That, and this is another situation where you know you're talking to a robot, so you're not, it's not putting you off at all. And, and pretending to be a human.
- [10:36]
B
Right, Exactly. Yep.
- [10:37]
C
And then the other two. So we've been talking about Legion mostly. The other two places, just to finish the thought before we move on, where we see really good use cases for AI and businesses is the, the first is in places where we have a business process that has traditionally been difficult to automate because the rules aren't the sort of thing that you can capture in traditional co in an if then statement sort of way. Right. Maybe they have some kind of fuzzy logic to them. They involve a judgment call. They involve the sort of thing of like, are these two cities near each other? But you don't want to define it with a hard mileage radius for whatever reason. Those are the sorts of things where we can now automate those processes and capture that logic as English text. Basically like an SOP document like you'd give to a human rather than as code. So stuff that previously proved very difficult to automate, we can now automate those business process processes. And so that's.
- [11:34]
B
Can you speak it in pseudo code? And so it's easier for people to take what they want and have it translated. Is that what the intelligence is? The translation of people's thoughts into specific task flows.
- [11:46]
C
Yeah. Well, like you, we will often use AI to help us make what is essentially the SOP document.
- [11:52]
B
Right.
- [11:52]
C
So like, one of, one of my favorite ways to do this now is I'll get on a screen share call with a client and have the client teach me how to do the task we're talking about. Like basically pretend I'm the new turn and teach me how to do this. And all the while we'll be recording that call and then we can feed that call transcript to an AI and have it produce what is effectively the SOP document for that task. And now that becomes the prompt that you're feeding to this agent to get the job done.
- [12:20]
B
Is that because people are not clear on how you'd create an automated workflow or what do you think the reason is?
- [12:26]
C
I think most people. Yeah. Are not great at defining a workflow in an objective, visible to the outside world sort of way. I think that's probably true.
- [12:36]
B
Okay, so in lead generation, I mean, who are the biggest beneficiaries of this? What kind of companies? Just digital marketing companies or anyone that uses certain technique.
- [12:46]
C
Yep. So I'd say the biggest companies who are using the AI powered lead gen right now are agencies, like both ad and marketing agencies. And then like I gave you the example of Jason Swinink there, who isn't really an agency himself, but who works with agencies. But kind of the through line there is. They're the people who are kind of on the forefront of marketing techniques.
- [13:07]
B
What other industries, maybe surprising ones, that are early adopters of AI chatbots and stuff that you wouldn't suspect.
- [13:13]
C
Yeah, so I've got a lot of good success stories in manufacturing, so less so with the chatbots, Although what we wind up doing is building agentic systems. So these are more than just a chatbot. It's a little sandbox that the AI can play in and it can write and execute code within that sandboxed environment. So now picture you have a manufacturing operation with this enormous database of all their past jobs and material orders and lead times and labor costs. And you can imagine all this sort of stuff in a manufacturing database. And then you give the business leader access to an agent where he can ask arbitrary business questions about the data in that database. And the agent can then write and execute code that queries the database in order to get the answer that he needs from the data sitting in the database. And can then write more code to do things like produce graphs and reports and PDFs in order to make this stuff more Digestible. So it's sort of like having a, a data analyst on staff, but it's accessible for a much cheaper price than going out and hiring a six figure data analyst for your factory.
- [14:22]
B
So you could say like, who are our best customers, who spends the most, etc.
- [14:27]
C
Yes. Yep. Absolutely. Yep. Another example there is one of the customers doing this as a very labor intensive. They're sewing factory. Right. So most of the most, most of their jobs are people sitting in front of sewing machines sewing complex things together. So that's a skill. So he has the ability to go in and ask, you know, hey, we've got to make 300 of this particular product this week. I want you to go back through the last six months of data and find me the three people who can, who are best at making this product because I want to put them on this 300 piece order.
- [14:57]
B
So besides lead generation, what other kinds of uses is AI working for?
- [15:02]
C
Well, yeah, so those, the, the three then. Now we've at least touched on all three of them. The three of them are legion. The second one is capturing workflows in English language rather than code. So this is all about automating back office work really. So that's where we, we want to identify a particular workflow and then document it out well enough that we can have an AI implement that workflow. And this, this almost never results in replacing a human, but what it does is takes one particular task of their job, often one that' tedious and time consuming, and automates it away so that then you can handle a bunch more business with the same team size. And then the third category is these agents which kind of act as automated data analysts. So there's a lot of businesses out there that are, are rich with data that they've accumulated over the years, but they're not big enough to have on staff analysts to go dive into that data and get insights for the leaders. So we can basically set you up a robot to go be that data analyst that you can't afford to have on the team full time.
- [16:07]
B
You can you analyze data from video? Like could you figure out a workflow by watching someone do a task or do you have to have it in like tabular database format in order to process it?
- [16:16]
C
No, we can work through all kinds of, of data. The trouble with video is there's only a couple of models that can accept video as an input. Gemini is one of them. And it per, it's a lot of data, it's a lot of tokens and you're Paying per token. So it would get very expensive to run large reams of video through an LLM. Prohibitively so. However, if it's more, it depends on the application that we're talking about. I have MID software, which does automated point tracking. So if we were talking about, say you had some factory process that ran thousands of times and you wanted to do point tracking within that, we could use the AI to orchestrate point tracking software to gather your data out of that point tracking without ever passing the video itself through the AI. The AI is orchestrating point tracking software, essentially.
- [17:06]
B
What is point tracking software? How does it work?
- [17:08]
C
Yeah, so this is actually, this is the old school version of AI. It's called computer vision. This goes back decades instead of just a few years. So essentially it would. I'm trying to think of a good example. You. If, let's say you had a baseball pitch being thrown, right, and you had a video from the first baseline so that you could see the whole path of the pitch. Point tracking would be, we want to capture the ball just as it leaves the pitcher's hands and then track where that ball is in every frame. So we can then know everything we need to know about the speed and position of that ball by comparing how much it changes between frames. So I'm getting a little.
- [17:45]
B
You represent. You represent the ball by a few points and the person and the bat, and you see how those points move through space. So it's tracking points instead of the whole data feed, the whole video.
- [17:56]
C
The cool AI bit of it is that we don't need to have a human identify where the ball is in each frame. Once it finds the ball in that first frame, as it's leaving the pitcher's hand in that example, it's going to be able to find the ball in every subsequent frame.
- [18:10]
B
So what? Okay, so AI is good for workflows, I guess. Workflows that are complicated, that have many, many steps, or recursive loops. This is a great example. What is it good for?
- [18:21]
C
A great test that I'm constantly talking with prospects and customers about when they're saying, hey, can we automate XYZ workflow? The first kind of the initial question is, could we write an SOP document so that we could then hand that document to an intelligent, resourceful intern and have them successfully do the job? If the answer to that question is yes, maybe the SOP document doesn't exist right now, but we write such an SOP document, then the answer is probably yes, we can automate that. If it's the Sort of thing where it's like, no, we can't really do that because, you know, step three here requires, requires one of the senior sales guys to sign off on whether or not that's okay. And I can't really capture his 20 years of knowledge in an SOP document. I'm judging whether or not this is okay. Then obviously that's highlighting a step. All right, we're probably not teaching an AI to do that anytime soon, so we're going to have to have a step in there where that senior sales guy signs off on it.
- [19:17]
B
So you automate what you can so the whole workflow can get 50%, 70% faster. Even though there may be a few, like, human necessary touch points in there.
- [19:26]
C
Absolutely. Yeah. We're. We're almost never. I mean, really, I don't think, I literally don't think ever wholesale replacing a human at a job. What we're doing is we're super powering your existing team so that you can handle more business. Right. More revenue and profit with the same team size. And kind of the, the side benefit of that is the things which are most amenable to automation are usually those things which the team likes doing.
- [19:50]
A
Right.
- [19:50]
C
They're usually the most tedious and boring. Their people are happy to have them off their plate.
- [19:55]
B
Well, just like people say that AI is making them millions while they sleep, there's also the doomsayers. You know, AI is going to take everyone's job and we'll all be sitting there doing nothing. So what, you know, in terms of jobs being lost or changed, what's that going to look like, do you think, from your perspective?
- [20:10]
C
Yeah, I mean, it's probably the most interesting question we have right now. I'm. A little bit. When we look back through history, it's certainly true that every time we had a big boom in economic technology like this, people said things like that. Right. Like people said when mechanized farm equipment first came out, you know, 90% of the country was involved in agriculture in some way, and, and everyone said the sky was falling and nobody's going to have any jobs. But obviously things turned out okay because people found other ways to be employed. And now some tiny percentage of the country is employed in agriculture. This is obviously different than that because it moves much faster and it's much broader than, than just one segment of the economy. But, but I think, I think we have a long time to go before there's no jobs for humans left to do.
- [20:57]
B
What will, what will jobs look like? You know, what's an example of a job that will be affected, will be changed by AI, but it's more of someone working with an AI or maybe to eliminate the entry level, but not a more higher level or you know, I mean, any job specific examples?
- [21:13]
C
Yeah, no, my job's a fantastic example. Right. So as a software developer, AI is fantastic at writing code. Like it isn't, it isn't perfect, it does, it makes mistakes and it doesn't have any sort of strategic vision, but it's very good at it, especially at the sort of boring, tedious boilerplate work like I was talking about earlier. So the effect that AI has had on the software development world is that you can produce much more software much quick, much more quickly with a given team size than you could before. And I think that's the pattern we're going to see writ large across the economy. And a lot of this kind of work is that you're going to be able to do a lot more with your existing team size.
- [21:56]
B
Hmm. Okay. So using like things like codec, you know, a programmer would normally take, let's say a month to program something, but you know, using codecs or other AI tools, they could program it maybe in a week and debug it.
- [22:07]
C
Yep, great example. Yep. And, but it's still very much human in the loop. Right. Using. So like the very best tools right now are stuff like codecs and Claude code. And if a non coder walks up to either of those tools and tries to make complex production ready software, like they're going to stumble and shoot themselves in the foot before too long. But it's, it's a powerful tool that if you know how to direct it, to not do stupid things and you know the list of stupid things that we got to prevent it from doing, you can move much faster than if you had to type everything by hand.
- [22:40]
B
Makes sense. Are there any other areas that maybe they're not here yet, but they're up and coming? They're surprising to you on what AI seems to be able to handle any.
- [22:48]
C
Yeah, I mean I think like, like before we were talking about stuff like outbound sales and I had said I, I was, I was down on it because of the voice we were talking about calling at the time.
- [22:58]
B
Yep.
- [22:59]
C
But if, if we're in a situation where it's strictly text, the, the bear models themselves are still no good at it because they're all trained to speak like corporate robots and it's, it's instantly recognizable how they talk. But if they're sufficiently well coached up on how to Talk a certain way, they can do a very good job at following a script like you'd want a sales guy to do. So I think in any applications where the SAL communication can be via text, like email or text messages, et cetera, I think that's another place where we're going to see a lot of AI usage and then I'm sure they are going to sort out the voice problem eventually.
- [23:36]
B
Yeah, interesting. Any other areas that are up and coming that you're excited about with AI?
- [23:41]
C
Yeah, I think that the trend, like the most exciting trend is these agents like we were talking about using for a data analyst. So the idea here is when you just open up chat GPT or Gemini or Claude, you're in a request response cycle, right? It's like a game of catch back and forth where you, you put in a message and then you wait for the AI to reply with a message and then you put in a message and it ping pongs back and forth like that. Whereas the idea with an agent is you, you kick it off with some instruction, but then it can sit there and operate in a loop and run code, execute code, run web searches, do all sorts of things like that with its tools until it thinks it's done and then it can come back to you. So it can run for an arbitrary amount of time. And obviously you can limit this to keep things from getting out of hand. But this, these tools phenomenally more powerful than a straight back and forth chatbot is largely because they can do that stuff like write code and then execute it in order to get things done. So go ahead.
- [24:41]
B
I'm interested in code that can fix itself because you know, with my company we'll have a number of websites up and things will malfunction. And I thought at first, because I didn't know anything, how can it malfunction? Why, what's wrong with the code? And my programmers would say it's what the code interacts with that changed. So do you have other programming systems that will code a website or a function, but it'll also look every day or every week to see if the code it deals with has changed and to see if there's a problem, it can rewrite itself?
- [25:10]
C
Absolutely. The term for this is self annealing. And I just had one of my systems, I watched one of my systems do it yesterday. So what's going on here is when these agents run, they have what we call skills, which are really just folders that they have access to that are broken down into individual tasks. Right? So this, this One I'm talking about had a skill for classifying whether or not a we website was a good candidate for this application we were working on and that called a particular anthropic model. In order to pass the whole content of the website through and ask if it was a good candidate, then a a higher level agent is calling that skill and asking it to evaluate this website. Well the particular anthropic model, it was an old haiku model that we were using, got deprecated yesterday which means that anthropic shut it, turned off the lights, shut down those servers, you couldn't use it anymore. So the top level orchestrator Agen saw that its attempt to review that site using the skill had failed, went into the skill, found the line of code where it was calling haiku 35, went onto the web, found that the latest version of haiku was 4 5, changed the code so that it called haiku 45 and then ran the skill again and then it successfully completed.
- [26:24]
B
That's awesome.
- [26:25]
C
So that that process is called self annealing and we have to explicitly instruct them to do that if we want it to happen reliably. But when you see something like that happen, it's really a beautiful thing.
- [26:35]
B
Self annealing? Yeah, we've had like contact forms that go wrong and all of a sudden this won't load and that won't load and you know, Google's always changing stuff. So what do you have to do to allow self annealing? Is it just a property of AI code or do you have to specifically tell it, you know, self anneal when you run into problems?
- [26:53]
C
Yeah, no, the system has to be engineered to work this way. It wouldn't be a trivial thing to paste on to say an existing website though it could be done for sure.
- [27:03]
B
Are there self annealing software that's out there or like, like how do you get an AI to do it?
- [27:08]
C
Y so we're doing it by we're operating inside one of these agentic loops that we've been talking about and you're in a sandbox area where the AI has the ability to write and execute code. So you want it walled off from important computers, right? Like you don't want it to be able to screw with your web server or your laptop when it's doing this. It's gotta be properly isolated and then it's a matter of just explaining to it that you're in a safe walled off environment. We're giving you permission to edit this code if something breaks. So go ahead and run it if you encounter an error, examine the error, you know, search for documentation if you need it, and then operate in a loop where you attempt to fix the error and then run again.
- [27:47]
B
Okay. And I guess it's really amazing when these things work.
- [27:50]
C
Absolutely. Yeah.
- [27:51]
B
It saves, I mean, how much time do you think it saves on ensuring all the code is up to spec and working properly and you know, the quality assurance angle, is it like a huge time saver or what?
- [28:01]
C
Yeah, absolutely. And like in this case, it's doing stuff that we wouldn't even be doing with humans in the first case. Right. It's enabling new use cases.
- [28:10]
B
Okay.
- [28:10]
C
Yeah. With software development in general, an enormous time saver.
- [28:14]
B
Yeah, I guess the quality assurance world AI will pick that up big time. It'll be very good at it, you know.
- [28:19]
C
Yeah, absolutely. Nope. We have a routine that we run where the AI attempts to act as a quality control tester. Basically, it'll spin up a web browser, run a test version of the app, and then go and click around the web browser like it was a user and just try to come up with the combination of clicks that can break things.
- [28:38]
B
Yeah, okay, but any, anything else that you've seen. One last thing. Any applications that are coming that not yet here, but are getting close that you think will be very useful or helpful.
- [28:48]
C
Let me see, like another cool application here is simple video editing. So for instance, like I've been doing a string of podcasts here like this, and a lot of them are, are sort of simultaneously podcasts and YouTube shows. Right. So there's a video of them. So we have an agent set up that whenever I go on one of those shows, I feed it the URL of the YouTube show that I was just on. It'll go, pull down that video, make a transcript of the video, identify the most interesting CL from inside that, you know, 30 minute long video, and then edit out just those clips, crop it for vertical video like on a phone, and then use an API to post that out to all the social media channels. So essentially it's, it's clipping up these podcasts into the most interesting bits and posting them out to social media. And the human involvement in that is copy pasting in the YouTube URL and hitting go.
- [29:39]
B
Oh, wow. And is it, is it accurate? Is it finding good, good parts or.
- [29:43]
A
Yeah.
- [29:44]
B
So is it funny parts that are weird?
- [29:45]
C
No, it's stellar at the interesting parts of a conversation like this. The bit where it struggles is with precise video edits, which I think is also a skill that humans have to Learn right of clipping it just the right part so that you're not catching the tail end of the previous word or the start of the next word.
- [30:04]
B
Yeah, makes sense. Oh, interesting. Okay, so are there off the shelf softwares that, that people can use or is it, is there a specific model that's better like ChatGPT or Gemini or you know, if people put this in their business, how do they, how do they go about it?
- [30:18]
C
Yeah, no, the, the model question. Great example. So for a given application, we will almost always try multiple of the big frontier models and you'll almost always find that one of them does significantly better than the others. And we can notice some patterns there. Like anything visual, Gemini seems to do the best with things involving like complex data analysis that we were talking about earlier. Anthropics. Claude models tend to do the best with. But on almost all of these projects where we're setting them up so that, which underlying model you're using is kind of an interchangeable Lego brick and then that lets you test multiple models to find the best. And also, you know, a month from now when the next latest and greatest model comes out, it makes it very easy to upgrade.
- [31:04]
B
Yeah, that's great. So people listening, what kind of customers are you, are you good to work with? Are you looking for and how can they reach you?
- [31:12]
C
Yeah, absolutely. So the customers we can work best with are sort of mid sized American companies, let's say team sizes of 10 to a couple hundred. That's the area where you have budget to tackle serious projects like this, but you don't have the crippling bureaucracy to get in the way of getting anything done. And how you can get a hold of me is you can find me on LinkedIn under Ken MacLeod or my website is laconictech.com L A C O N I C t e c h.com and then obviously the email is kenictech.com okay, excellent.
- [31:49]
B
Well Ken, thanks for coming on the podcast.
- [31:51]
C
Yes, thank you very much.
- [31:53]
B
If you like this podcast, please click the link in the description to subscribe and review us on itunes.
- [32:00]
A
Thank you for listening to the Good Question podcast. Please email Support the GoodQuestion podcast.com if you have any referrals to great guests for us to interview, visit thegoodquestionpodcast.com to hear more interviews. And please help us spread the word by rating and reviewing us on Apple podcasts, iTunes, Spotify, YouTube or wherever you listen to this podcast.