Companies replacing staff with AI are idiots

Yeah, it's great to be here.

So, I mean I'll I'll give you the condensed version 'cause I'm I'm quite old now. So but I started off in recruitment and then I've always been a bit of a geek. So I I it while you you know, but in recruitment I always use like LinkedIn. so I think I've had my LinkedIn account since two thousand and four. So I was always looking at ways that could help me to do my job. so after recruitment I kind of moved into talent acquisition. So work for a big logistics company as using social media to find talent. then I kind of moved into social media marketing, which progressed into digital marketing, and then about four years ago started using AI. initially it was Jarvis, but then it had to change its name to Jasper. So it was mainly you know, using it to help write. and then obviously the advent of large language models came about. So I started looking at how I could use them to help with my marketing, then started developing products. and then a year ago I decided to set up on my own. So I it's Igentic, so it's as you mentioned, it's got the AI in it. but it's agentic wasn't available for some reason, so I had to do a little play on the on the word just to say what it is. but I basically do workflows and optimization, so look at problems within companies. you know, processes that they have and then I'll look to see whether there's an automation needed or if it requires an AI agent. So yeah, that's that's basically it.

So yeah, that's so I mean obviously I come from a marketing background. So one of the things within marketing is kind of don't spread yourself too thin. So I've never worked I mean, I've not really worked in construction before, but there was a local company to me, a construction company that I got in touch with just to say, would you be interested in any automation? And he was like, Yeah. And so basically I'm I'm building them a timesheet management tool that uses clawed OCR so that it's actually able to read paper timesheets and extract the data from it. So it's my you know, a lot of my background has not really been within construction. You know, I've I've worked for logistics companies, I've worked for insurance companies doing marketing. so kind of as a company, I'm not really I could work on any industry. I'm not really that. you know, it it's basically if there's a problem, I'll look at it and see if I could build an automation or a workflow for it. So, but however, from a marketing perspective, that's not great because 'cause it's kind of I'm not focusing on an industry, I'm kind of leaving it quite open.

Yeah, I think for me, when I'm like if I'm speaking to a company or if someone's considering using AI or doing an automation is I would look at what firstly keep it simple. if you try and go in and make it as overcomplicated as possible, you've got so many points that it can break. So what I would suggest is you start off simple. So It could be look at something which is using like taking up the most man hours of someone's time, you know. So it could be, I don't know, analysing spreadsheets, for instance. So that's AI's bread and butter is is like, you know, it's got the data's there, it just needs to analyze the data and output a response. Now one of the things there is as long as the prompt is decent, it's not gonna make a mistake. Whereas if a human is going through all that data, there's a high chance that it will make a mistake. So that's that would be my suggestion is keep it simple and look for the most time consuming task that someone is spending the, you know, the most amount of time working on.

So on that instance, I I would I would say there needs to be a a happy medium. So the people that are refusing to use AI, I think that is the scenario where AI will take your job. And it's not the fact that AI's gonna take their job exactly, but someone that uses AI. will come in and replace the person that's refusing to use AI because the it's a tool. At the end of the day, it's just a tool. It's it's so somebody that's using that tool is going to be able to do the the job quicker than the person that's refusing to use AI because they're using this tool to allow them to do more time doing their job by doing tasks That are time consuming, like admin or something like that. So that's on that end of the scale. But then people or companies that are replacing their staff with AI, there's no nice way of saying it. They're idiots. Because AI doesn't know the nuances of a job. It's, you know, there's the certain things that AI cannot do. So it because it's it's basically it's ones and zeros. It's just it's you know it's just code. So you need someone that n understands the job, that can then allocate certain parts of that job to the tool. You know we've had scenarios where, you know, some of the big AI companies had got rid of staff and all of a sudden they're now having to hire them back. Because they've worked out just having AI do the s the work. It's it it you know, it it's not feasible. You need someone that can control the AI. You know, people's roles will change slightly. You know, a developer's role you know you know, is not they're not you still need developers. You know, if once I'm at a point where I can recruit, I will recruit a developer. It's just the developer's role will change slightly. That's all.

Yeah. So I think the thing for I mean, even now I think if companies aren't using AI, it's not too late. It's not like you've missed the boat and it you know, it's it's I think there are a lot of companies people think that, you know, companies are a lot further ahead with AI than they actually are. so it's not it's not a case of we've missed the boat now and everybody's using it. They're not. You know, it's and if they are using it, they're potentially not using it properly. there they may only be getting like ten percent out of it and nothing is organized within a company. You know, it's like people are maybe using their own Chat GPT account or the company has a a business company. Copilot account part of you know office. It's there's so much more that it can do than just work on a spreadsheet or answer a question. Because the thing is also is that it can make things up. If you don't prompt correctly, and if you don't tell the AI where to search for the information that you require, AI Large language models at the moment are people pleasers. So it's never gonna tell you I don't know. It will t tell you an answer, even if it's made that up. So that's the worry for me is that if companies don't know how to use it and their staff are off doing whatever they want on AI without knowing the process of prompting. they're gonna they're gonna get wrong answers, they're gonna get things like that. And that could then make a business decision from that. And that's where the damage will come for a business is once you start getting all of that information. So my advice is if your staff are using it, make sure there's a policy in place and make sure that you've you're training your staff on how to use it. You know, because another big thing is someone could go in and put some customer data into Chat GPT. Now that is no longer secure data that you've just given OpenAI some of your customers' private information. So that's a big issue is knowing the security, what you can put into an LLM without it breaching GDPR.

Well, I had I had a scenario, a friend of mine, he was trying to work out so he he owns a business and he was trying to work out a calculation, like so a finance calculation. Rather than putting it in the calculator on his phone, he went and put it into Chat GPT. And the answer it gave wasn't correct. I was like, Well, why did you do that? Why didn't you just do it on the calculator? And he was just like, I'll just do it. But it's like it's it's again with the prompting thing. If you're not prompting it correctly, if you just ask it something, it may get it wrong. In this case, it wasn't a just a simple I'm not gonna say it, you know, it got one plus one incorrect. There was a bit of working out to do with this sum, but still he could have used a tool that is on his phone that's still just one click away. And that's one task is that is what it does. Whereas

Yeah. Well the th the thing is as well, it's not only and this is where the difficulty comes in with large language models, because not only is it the prompt needs to be right to get the right outcome. So but it's also what model you're using. So for instance, you know, you could use Claude Opus five point five, which is Quite expensive, it takes a lot of tokens, but it's a thinking model. So if you just want something to be a yes or no answer, and you're giving it the data to look, then using Opus 5.5 might not be the right model to use. So it that's this is where the difficulty comes in because you need to look at what the process is you're asking your model. or your workflow, what task is it going to do? And then that's where you then decide what model. Could I go with a lesser model or a free model to do this task? Because it's not doesn't require much thinking. So that's a bit of a minefield on that one. And that's where you'll probably need someone with a bit more expertise if you're trying to do something slightly more complicated.

Yeah, I use open I yeah, yeah.

But the thing with that is that you need to so when you build anything, you need to make sure that you are optimising everything because the new models are coming out all the time so quickly. Like for instance, I think it was was it Opus four point five before Opus five point five? and it was not very good. It was really expensive. and, you know, people were going no, sorry, it wasn't, I think it was Opus five. And people were going back to Opus four point five because Opus five was they they just found it n not very good at all and really expensive. and so when you're building anything, if you're using open router, is you need to have something in place so that you're constantly checking the performance of your workflow or or whatever process your LLM is. And that it also can update to the latest model. So if you are using Opus 5.5, so say they bring out Opus six, which is cheaper to run and you know, and better, then you you could automatically switch to to that. And sometimes, I mean, I've built one which will actually analyze and then swap out a model to a completely different one for me. Depending so it could be not Opus, it could be GPT, you know, five point six or whatever, but it'll swap it out for me by looking at the performance and the results.

Yeah, so yeah, so some of the open weight ones. So so the apologies, but this might sound a bit political, but with with the things going on in the US at the moment, obviously most of the frontier models are in the US. You know, with Gemini, Chat GPT, Claude. so what you what you've What you find is that the concern with the open weight models is most of those are in China. And so at first the concern was, well, what are they going to do with my data? But unfortunately, with the US government that it is at the moment, well, you could safely say that that's a concern for the frontier models as well. so it's kind of made it less of a threat to go with the open weight models because they're potentially in the same situation where in the US you're not sure where what's happening with your data. Well it's the that's the same concern with the Chinese models. So I think as long as you put things in place that you can that you're only giving certain information to, you know, that you're not a lap Allowing everything to go in, which you wouldn't do anyway. But a lot of these open weight models you can host yourself. Whereas the Frontier models you can't. They're you can't host them yourself. Whereas you could get, you know, all right, you can get Gemma, which is Google's, but a lot of these open weight models you can just go and subscribe on the site and actually install them if you had a big enough server. And therefore there's no they're not going outside of your server. So you could ask it any questions you want and it's all encased on your server. So there's no security issues.

Yeah. So one of my favorite things to do is actually put on so use your voice instead of typing. So one of the there's a free one I use called Fluid Voice. And what you can do is you can use your own API key for whatever model you're using. So you could if you've got an anthropic API key, you can use that to improve so it'll listen to your voice and then it uses your own model. So you could even use Open Router and find a model off there. But you put the API key in and then it will update the the transcript. Type using your voice and waffle as much as you want. Just talk to it. But when you are talking to it, say all the things that you want within the prompt, but also all the things you don't want. Because prompting's not just about I would like this, I would like that, I would like this. It's also about, but don't do this, and I don't want this, and I don't want you to do this, and I don't want you to look here, and I don't want you to go there. I only want you to look in this on this web domain or whatever it is that you're doing. And that is the best way to prompt. But it it feels weird at first because you're literally talking to your computer. But once you get the hang of it and once you're used to it, I just sit there, tap a button, talk away, return, off it goes. and it is far better because you can speak more words than you can type. And I think humans are inherently lazy. So when you're typing, you try to type as

little as possible to get your point across. But when you're talking you just waffle away for five minutes and then it will read that and off it goes.

So you can either find me on LinkedIn. So it's I mean, I you can I'm sure you'll share a link on that, but it's just Dale Taylor and the company's Igentic. and then obviously if anybody's interested in wanting to speak to me or if they're looking at AI and automation, then they can get me on my email, which is Dale at Igentik-T. ai dot co dot uk.

You're very welcome.