Episode 2: Luke Budka, AI Director at Definition

[00:00:00] Hello and welcome to the Conversational AI Podcast. Today we are talking public

[00:00:04] relations and communications and I'm joined by Luke and I'm going to ask Luke to give us an overview

[00:00:10] of himself and his company and we're going to get into the conversation. So look first off all,

[00:00:14] who are you? What's your background and then tell me about the company?

[00:00:18] Sure. So my name is Luke. I'm AI Director at Definition. My background is in tech comms,

[00:00:25] linguistics and content essentially. I was a founding employee, so I should say, of an agency

[00:00:36] called Topline Comms back in 2008 and we were a kind of B2B tech focused agency. We got acquired

[00:00:44] during the pandemic by definition. Definition is an agency on a buying bill mission so we buy

[00:00:54] agencies that have particular specialties that we don't basically. So we are the sum of

[00:01:01] I think six or seven different agencies now across London, Manchester and Leeds doing everything from

[00:01:06] PR comms, video, language training, customer experience, research, events like loads of

[00:01:14] different bits and pieces, internal comms, graphics and island brands and stuff like that.

[00:01:18] We are an amalgamation of lots of different agencies pretty much and that's what led

[00:01:23] us down the generative AI and the conversational AI route back in November 22 when chatGBT was

[00:01:30] acknowledged. Gotcha, because the world did change for quite a lot of us I think at the end

[00:01:37] of November then. Can I just before we get into it though tell us a little bit more about

[00:01:41] definition in terms of the type of clients. What kind of clients are you servicing?

[00:01:46] Sure. So I mean we've just done for example prompt engineering training with PepsiCo

[00:01:51] and then all the way to the small end of the spectrum we are working with a UK online tutoring

[00:01:59] company, a 20-man company and we've just finished what I think might be a world first

[00:02:05] which if you publish this before the next week you'll get exclusive but we've been fine

[00:02:11] tuning a version of Larmor 3 on explanations from fully qualified UK school teachers

[00:02:18] designed to teach key stage 3 and 4 maths in English by turning the training, turning the

[00:02:26] recordings between the teachers and the students that they've collected over the years

[00:02:29] into data we can use to train the model. So very much from the very big, big multinational

[00:02:35] companies all the way through to the kind of much, much smaller kind of 10-20 person

[00:02:40] type company depending on the brief pretty much.

[00:02:43] Am I right in thinking like an AI director is perhaps not something you tend to see

[00:02:49] even in today's world at an agency? Is that fair or you know?

[00:02:54] Yeah, I think so. So I invented the role pretty much right for me anyway because

[00:03:00] I went to our board in January last year and said two things firstly we need to lock down all

[00:03:05] uses of generative AI because I'm worried that we're leaking IP and we know all the inputs

[00:03:10] that are pushed into these models used to train the foundation models. We don't know who's putting

[00:03:14] what in so we're locking down across the organisation and then I said can we because

[00:03:19] we've got this bio-build strategy can we acquire a dev agency? Can we acquire an agency with data

[00:03:23] scientists and machine learning engineers because I'm very worried that in 18 months time

[00:03:27] everything we offer won't be a thing, will become extinct pretty quickly unless we

[00:03:32] pivot and mitigate the impact of this and we didn't acquire an agency but I went out and found

[00:03:38] a development agency and then kind of became the AI director in charge of developing our own

[00:03:43] proprietary solution and introducing AI services into all those different teams and services

[00:03:49] that we offer. I am now seeing it as a job title that's springing up across the industry

[00:03:55] and I've seen a couple of agencies recently appointing AI strategic strategists or AI directors

[00:04:00] and obviously you've got like the the four companies like WPP very much have these

[00:04:06] kind of job titles in their midst. For the uninformed like me when it comes to your industry

[00:04:15] I would assume that the base level panic is you can make a press release easier

[00:04:22] in three seconds with chat gpt right yeah the uninformed of you but would you flush that out

[00:04:29] massively look what you know how how were you initially thinking about the use of AI and then

[00:04:36] you know how is that moving though because you mentioned different teams and functions or so

[00:04:39] yeah I mean there's a couple of bits there to unpack so in terms of just working backwards

[00:04:44] to that question how are we initially thinking about AI I think probably the same way quite

[00:04:48] few agencies were which was we had a spreadsheet where we were logging our most successful prompts

[00:04:53] and sharing it with the team and we were all using them in chat gpt and we were doing things

[00:04:58] initially pretty basic stuff like yeah writing turn-based prompts that asked our employees for

[00:05:06] the information required to write a press release so the first thing the AI would ask you would be

[00:05:11] the name of the client and their domain name and then next thing it would ask you would be

[00:05:15] the 5w's the who what when where why they need to go the first paragraph of the press release

[00:05:19] and we found that like those turn-based conversations work quite well to produce

[00:05:23] decent content that didn't require much iteration and we quickly then started realizing that we

[00:05:28] can do things like create personas of different with different kind of social demographic

[00:05:34] like backgrounds and use it for brainstorming and do all kinds of things and it grew from there

[00:05:41] as I could go on and on and on about how it's grown from there but we're now a situation

[00:05:46] where we have our own in-house proprietary AI solution with a bespoke prop repository where

[00:05:52] staff can search through a bank of prompts and they will auto load in the correct AI model

[00:05:57] we know for example sonnet the claw 3.5 sonnet is better at writing certain types of content

[00:06:02] than gpt4 omni for example so the prompt will all run in that model so we've come quite a

[00:06:07] long way in quite a short period of time um yeah and ultimately I think the other thing

[00:06:13] is that we've had to tackle it in the order in which the industry has developed so yes obviously

[00:06:17] initially conversational AI chat bots language bots very good at next word predictions and very good

[00:06:23] at producing text very quickly moved into diffusion models very good at producing imagery but as and

[00:06:28] then we can move into kind of the world of audio and now we're soon going to be moving

[00:06:32] to the world of video and well we're already doing that but we're kind of all waiting for

[00:06:36] saura in these two minutes high fidelity video would be promised and yeah so we've kind of

[00:06:40] had to tackle it as it's happened um so as the as the industry's developed we've developed with it

[00:06:48] absolutely fabulous because that that must both be rather exciting but also well exciting with

[00:06:55] the different definitions of the word exciting right because you know a press release comes out

[00:06:58] from open AI here we go we're going this way guys well yeah it's a thrill it's the role

[00:07:08] jik it's quite difficult in a way but at the same time it's not because it's all very well

[00:07:13] signposted right but and but the other thing for us is that we've never we've never built at this

[00:07:20] scale and we've known and no one has ever built with this technology it's something my developers

[00:07:25] say to me my development team say to me look we're building this stuff that didn't exist six

[00:07:29] months ago so it's like a lot of learning going on on a daily basis but it strikes me

[00:07:36] while you're busy you know at the front end there the bleeding edge of you like not knowing the

[00:07:42] answer but discovering the answer I wonder that there's quite a lot of other agencies and practitioners

[00:07:50] way over there left behind here you know what's your assessment of yeah I think some are

[00:07:59] so when we went down right without being kind of we've built our own private environment

[00:08:04] that means we use our Microsoft entry logins we log into definition AI we've got a variety of models

[00:08:09] in there we've got we've got it does all kinds of things it transcribes it produces imagery and we've

[00:08:15] got bespoke proper positive trees we've integrated open AI's assistance now as well so we were at

[00:08:19] the very beginning of agents so we've got these assistants with bespoke instructions that use the

[00:08:23] code interpreter and the file search tools which is like rag and data analysis so we can now

[00:08:29] have like digital twins of our clients so we because you can attach up to 10 000 files using

[00:08:33] the file search functionality so we can have a digital twin of a client so we've now got AI members

[00:08:38] of the team that have access to all the clients information via a vector database it's all private

[00:08:43] and secure in our private environment so in a way yeah we've gone quite quickly and other agencies

[00:08:47] I'm now speaking to other agencies about building the equivalent thing for them yes

[00:08:52] the more exciting bit for me is where this now goes and just very quickly on that which I

[00:08:58] tangent so we've recently launched the ability to spin up demo environments so in our

[00:09:06] in our week what we can do is we can now so I can make you a conversational AI demo environment

[00:09:11] which will have your logo and it'll be on conversationalai.definition.ai.com you log in

[00:09:16] you've got a proper positive there with prompts in it that we've written for you right now the

[00:09:21] future for me is all our clients will have one of those environments and in it we will have a

[00:09:26] bespoke blend of pre-written prompts that are designed to solve their use cases will have fine

[00:09:33] tuned models that are designed to solve their use cases and then ultimately will have autonomous

[00:09:38] assistance agents sorry that are designed to solve their use cases they'll all be managed by us this

[00:09:44] end now this us managing this end is how we don't become redundant as an agency in my head

[00:09:49] yes at least it's just my opinion right but this is like the truth I preach which is we still

[00:09:54] our experts do the fine-tuning of the models to write the prompts our clients don't really care

[00:10:00] they just want to click a button it works so we do all the building we fine-tune the models we

[00:10:04] develop everything and then deploy it to their environments each client will have their own

[00:10:08] one with all their bespoke prompts and models and all the rest of it that's kind of where I'm

[00:10:12] going well where I want to go with it we've gone 60% of the way we can now do the demo

[00:10:16] environment piece and next year is going to be exciting does that mean look if I simplify

[00:10:22] radically yeah like I've often in the past to be in the pitch you know and then it comes to the PR

[00:10:29] agency that you know and Darren here will be your guy okay and then Laura here that's your team

[00:10:35] right can you can you envisage a time whereby it's essentially here is Paul your assistant

[00:10:41] right you know or Linda your your your your your employee or your digital employee yeah but

[00:10:49] then you'd have a support human support around that yeah I know answer right you're not right

[00:10:57] answer you know Linda will be emailing the client prompting them now that that's that's really like

[00:11:04] very very interesting very exciting we're not that's what we're already talking about I mean

[00:11:09] there's very obvious use cases a billion use cases for an hour right that's the most exciting

[00:11:13] thing about it but take a very practical example right so we use a there's plenty of these services

[00:11:18] around but we use something called response source which is where do you have exactly that how we were

[00:11:23] connected right yeah cool right okay so what if we have for example an assistant that is connected

[00:11:29] to our inbox yes and it has the file search bit turned on it's got all the client's data all the

[00:11:34] subject matter that's stop for a minute sorry I should have said to explain explain response

[00:11:40] just for those sorry yes yes because it's very important for this bit yeah yeah journalists

[00:11:45] send out inquiries they'll ask the request information via something called response source

[00:11:49] there's other ones called health report out Harrow if you just go onto Twitter and use the hashtag

[00:11:53] journal request you'll see lots of requests from journalists looking for information case studies

[00:11:58] all those bits and pieces so we get them directly into our inbox now we have a situation

[00:12:03] where we very soon may have a an assistant that will see the email come in from that address

[00:12:10] it will identify using its conversational abilities whether this particular client has something to

[00:12:17] say on it based on the files it has associated with it it will draft the response it will

[00:12:21] automatically email it to the client then it will send chase emails because it'll know when

[00:12:26] the deadline is because the deadline is on the response source initial email it will chase

[00:12:30] them and then it'll come back to our team and say the client is signing this off sending the journalist

[00:12:33] or we remove the humans all together okay maybe maybe in two years time is your your digital agent

[00:12:42] is talking to the client's digital agent we are yeah it's really cool really cool okay now the

[00:12:50] the human dimension here look how are you exploring that you're explaining this to your

[00:12:59] your pr teams your the professionals the communications professionals doing the front

[00:13:05] line day-to-day how have you begun that training or that that or that met you that methodology

[00:13:12] it's it's difficult man like bear in mind so initially it was a case of doing like kind of

[00:13:21] what is AI basic training with everybody in the company then it was a case of showing them how

[00:13:28] to use the proprietary solution we developed then it was a case of showing them how to use it

[00:13:32] better i.e. doing basic prompt engineering training with them showing them the basic

[00:13:35] tips and tricks and techniques they can use to get more out of the model or more out the models

[00:13:40] they're using now we're moving into a phase of we're currently undergoing undertaking a piece of work

[00:13:45] where we move between each team in each three offices and identify pain points and areas they're

[00:13:51] struggling with and then put a more kind of concrete plan in place i think the the issue

[00:13:57] will be that at the moment the AI team is me and two other people on a part-time basis

[00:14:04] plus a development team and and once again this is my opinion and the uh my company will

[00:14:09] not be surprised to hear me say this but i think we will quickly reach a point where we need more

[00:14:13] resource to chuck at this because what we want to be is a central function that supports all our

[00:14:19] teams at the moment we need to we're doing a deeper dive on understanding how we can do that

[00:14:23] i mean we've had other things in place as well like we have a prompt request form so anybody

[00:14:28] in the company can request a prompt our prompt engineering team will write on it upload it

[00:14:31] to that bespoke prompt repo test it let them know it's there they don't need to know how

[00:14:35] to write the prompt it's already there for them kind of thing we've got an assistant request form

[00:14:39] so they can request an assistant to do their qualitative analysis for example so the other

[00:14:44] day i was writing an assistant to help somebody move through free tech survey data and we the assistant

[00:14:49] looks for exact match names of a brand but also phonetic variations of it because it's that

[00:14:55] obviously it's conversation they are it can do that right like so we can look for like things

[00:14:59] like that as well so we've got request forms for different features of a bit sorry different

[00:15:03] assistants and prompts but yeah we're doing a deeper dive at the moment that will help us

[00:15:09] kind of inform our helping form our 2025 strategy when you're talking to clients

[00:15:14] i i would imagine that you would be blowing them away i mean is that is that fair

[00:15:22] is that a fair assessment because it sounds deeply impressive to me look i mean that sounds

[00:15:29] a little bit like i'm getting off on my own hype man like i don't say that i'd say i'd say that when

[00:15:36] you i'd say the biggest challenge and this is not news for anybody is understanding the use cases

[00:15:41] right is understanding how you can use it how it can do things for you and when you when you explain

[00:15:47] to a client when you get when i get excited about like so we integrated claw 2.1 right because

[00:15:53] it had a 200k contact window and the clients like nobody cares why right no this is wicked

[00:15:59] they're like why i'm like because look and i demonstrate and then we were like because

[00:16:03] we could include 15 award entries you've previously written in a prompt that now

[00:16:08] writes your award entries for you so when you when you help them realize the value when you show them

[00:16:13] it they're like all right that's why you're excited so like uh anthropic this week shipping

[00:16:20] an enterprise version of claw that's got 500k context window they were very keen to point out

[00:16:24] because obviously it's reams and reams of codes you're pushing in it if you're an enterprise type

[00:16:28] company like bringing it to life is what excites clients the things that excite me are not the

[00:16:34] same things that excite the clients the clients that get excited by the use cases by the fact

[00:16:37] that suddenly they're i mean in a sad way like they they don't need to use as many

[00:16:42] copyrights over here to do this thing because this can do it for example or

[00:16:46] they don't need to use as much translation agency resource because this can do it for example

[00:16:51] that's what it is is that so i hear you because that's a that's a sensitive question it's a

[00:16:57] sensitive issue but then um i would imagine there are there are emerging skills emerging

[00:17:05] types of discipline that you know you and the industry is going to need to hire

[00:17:12] yeah i think so i think i think a lot of i think people can i think people can i think a lot of

[00:17:22] people can reskill as well not just hire yes i think people who i work with internally think that

[00:17:29] there is a level of technical proficiency required to do things like prompt engineering and draft

[00:17:35] instructions for open ai's assistants and stuff like that and to a degree there is but it's not

[00:17:40] the bar's fairly low i mean if you understand that if you like xml tags you can write delimiters in

[00:17:45] prompts that and just basically at least and things and i think people are capable of way

[00:17:50] more than they think they are yes and they think they're going to therefore like struggle with it

[00:17:54] and actually i think a lot of it's going to come down to reskilling on our end at least um i don't

[00:18:01] know about other agencies or internally app clients i mean i've read like yeah i was reading a

[00:18:09] press release from zeta labs i think they're called the creator day autonomous ai agent called

[00:18:15] jakes i was reading one of their press releases from june yesterday and it just developed a

[00:18:21] mass tutoring company it built the product when it found the first customer and started billing

[00:18:26] autonomously so we're all in trouble um yeah i think ultimately there's a difference between

[00:18:33] me prompting a diffusion model for an image and Picasso doing it and that's the point

[00:18:37] i try and make internally which is the expertise doesn't we still need experts clients will still

[00:18:42] want to know that we've got experts who are overseeing these things who are fine-tuning

[00:18:46] the models with silent stuff off yes like i can't give a diffusion model the lighting

[00:18:53] composition camera angle all the rest of it that somebody who spent the last 15 years

[00:18:58] directing video shoots can for example so yeah it's a combination of reskilling but also

[00:19:03] understanding that like the human expertise won't will still be required yes i mean do you do you

[00:19:11] see a situation or is there a reality whereby you are constantly reminding team members that this is

[00:19:17] assisting you you know you will read the output for example right you know before it goes out

[00:19:24] before it goes to clients or how do you approach that yeah there is i'll be honest with you

[00:19:33] internally so we're about just don't think about 100 people big 110 people like um i spend so i

[00:19:42] we we very early on put things it what's the best way to go like we kind of put the uh

[00:19:48] the safeguarding in place the guard rails in place early on by making the point during our training

[00:19:53] that you wouldn't just send the press release or a guide from to sign off without getting the

[00:19:57] account director to sign off first nothing changes from that perspective gotcha gotcha yes

[00:20:02] so that's that's still the reality yeah it's just you're making things more efficient now

[00:20:06] right when you've got one AI that can check another AI's work then it ceases to become the

[00:20:10] reality it's going to get interesting isn't it well that's what's that's that's the whole

[00:20:16] that's the whole super alignment isn't it the the uh the over premise there is the super alignment

[00:20:20] team that open AI that disappeared isn't it they're weak AI to control the super AI because humans

[00:20:25] aren't capable of controlling it okay what are you excited by i mean we've had some hints

[00:20:30] there but if i was to ask you to itemize you know excitement what are you excited by

[00:20:38] from most to least the things that's like me the most um i think the thing i mean

[00:20:44] what day of the week is it i mean what yesterday we had ilia raising a billion dollars for a product

[00:20:50] that he doesn't intend to ship this is the former co-founder of open AI

[00:20:58] i'm a chief scientist yeah open AI yeah who's gone and started his own

[00:21:03] agi company to create safe artificial general intelligence his super intelligence yeah in fact

[00:21:08] he can click his fingers say i'm not going to ship anything for two years i want a billion dollars

[00:21:14] to sit in a room with my team for two years and work it out i find that phenomenally exciting

[00:21:18] yes i find it's very hard not to be excited by the rapid advancements in the

[00:21:24] autonomous AI world i mean this week i've seen everything from all terrors minecraft village

[00:21:31] that's populated with a thousand AI agents inventing religion and reward systems and all the rest of it

[00:21:36] and what excites me about that and i'll talk about this with my development team this morning

[00:21:41] there's a lot of these companies are doing it using existing models these are not the result

[00:21:47] of new advanced models this is the result of mixture of expert architectures where new where

[00:21:52] models are being blended together for greater they're greater than some of their parts right

[00:21:56] we have the japanese CEO open AI this week saying that gpt next will be 100 times more

[00:22:01] powerful than gpt 4 but i haven't used more compute to achieve that they've just re-engineered the

[00:22:06] architecture like it's so early doors now i would i should as a word of warning i have

[00:22:11] fully drunk the kool aid right i am i think the media i laugh when i see headlines about

[00:22:18] bubbles bursting and things because i think the media in general treat uh AI like the commerce

[00:22:24] revolution or the invention of social media yes i think they think it is a tech product right i don't

[00:22:30] think it's a tech product i am fully in the camp of this is a paradigm shift in society i think

[00:22:36] the reason governments take this as seriously as they do and we've seen like Singapore financing

[00:22:41] re-education for those over the age of 40 and the nsa in america taking very close

[00:22:46] notice of open ai's announcements this week yes because this fundamentally changes everything

[00:22:52] yes um so i am i am fully bought in i am 110% kool aid uh saturated um so yeah that's not for the

[00:23:01] fun of it though if i was to you know if i was to support that look it's not it's not you know

[00:23:05] it's not kool aid i do mean a lot of people that are very enthusiastic almost dare i say

[00:23:10] religious when it comes to ai and especially gen ai but you know i would imagine you're seeing

[00:23:16] the output benefit of it daily yeah your teams are seeing right it's not right so we can go oh

[00:23:21] that was a bubble we'll keep we'll keep that thank you because it's still it doesn't matter what yeah

[00:23:26] but also it doesn't matter what any of us sale do because this is never going to not be a thing

[00:23:31] right yes like my mate was saying to me this morning um who he sold his software company

[00:23:36] at the same time we got quiet so we chat about these kind of stuff and he's like aren't they

[00:23:40] isn't it true that open ai and google and Microsoft only done 40 billion sales but invested

[00:23:45] 500 billion in building stuff i'm like yeah it is but look how quickly ilia raised a billion

[00:23:49] by taking these things and not shipping anything yes and are you going you know uh is your agency

[00:23:55] when are you going to stop paying money to these companies yeah and like and if the private

[00:24:01] center investment dried up if the if the p went away the vcs went away governments would finance

[00:24:07] yeah yeah the american the u.s administration would happily fund it just like the chinese

[00:24:14] government is currently doing no this is not this is not a bubble this is this is this is something

[00:24:19] different um so that excites me the the kind of um the the advanced abilities with the current

[00:24:26] models and then the other obviously the big thing that's incredibly exciting because it's

[00:24:31] terrifying is the is the autonomous piece is the agents piece yes which is seems to be developing

[00:24:37] a rate of knots we've got research out recently showing that uh we have now have agents that can

[00:24:44] beat the average human web navigation for example where do you see opportunities

[00:24:50] in terms of conversationally i managing that that engagement that some of your clients are

[00:24:54] using your front end aspect yeah yeah sure i think i think the uh the reason so on my uh

[00:25:01] on my LinkedIn profile like my little tag that i came up with myself was uh uh don't look and sound

[00:25:07] the same right so like we have a language team at definition that is full of we have behavioral

[00:25:13] psychologists we've got linguists we've got some of the best language professionals in in the

[00:25:17] country right in this team and as such we do a lot of training with big companies around the

[00:25:22] world and we do a lot of tone of voice work and stuff like that now i think there's a big difference

[00:25:26] when it comes to conversation ii between setting up a chatbot and setting up a chatbot that speaks

[00:25:31] and talks in your tone of voice a tone of voice for example that has been shown empirically

[00:25:36] to reduce term rating customers or to uh reduce the amount of calls your customer service desk

[00:25:42] gets so a good example of that is uh and there's a case on our side that i can't remember

[00:25:46] exactly but i'm just on a tangent the beauty of the acquired is that i then get to work with

[00:25:51] these kind of companies so we got acquired we got put with another company called schwa that was acquired

[00:25:56] they're the team of language experts but they did some work for hsbc for example where they

[00:26:00] rewrote their customer health guides it reduced their calls to their customer health is their

[00:26:05] customer service help desk by x amount which reduced their cap their expenditure on it by

[00:26:09] six feet crazy yeah right now we are now in the same world with conversation a i where and there's

[00:26:16] a brilliant case study on google site about somebody using gemini companies in gemini for this but

[00:26:20] this is the kind of this is the future i think from a language and conversational and i perspective

[00:26:24] for us is doing using the tone of voice work we've done in the past using our kind of ability to

[00:26:30] simplify stuff and um and turn our tone of voice guides into prompts that can rewrite stuff but also

[00:26:37] kind of pushing our tone of voice into the conversation a i assistance that these companies

[00:26:41] use to communicate with various stakeholder groups it really can make a massive difference if you

[00:26:47] change your sentences from passive to active it can really like it your verbiage your sentence

[00:26:52] length your adjective use like all these things make palpable differences to bottom lines of companies

[00:26:59] i i just think i think so many people for obvious reasons aren't thinking about that aspect

[00:27:05] right you know they're thinking about the the nuts and bolts right especially a lot of

[00:27:09] the vendors i think it's fair to say a lot of the vendors that uh that i i interact with on a

[00:27:13] regular basis the conversationally i vendors that's perhaps not fun of mind you know but the

[00:27:18] importance of this i think is is missed at the moment but i wonder if that's a a new service

[00:27:26] because that is relations to public right it's its communication which some institutions i see

[00:27:34] have been very very very focused on this some banks they have a whole team you know

[00:27:38] the the rigorous about this but then that comes with it let's take a look at um UK banks you have

[00:27:44] Nat West their conversational AI Kora right Lloyds virtual assistant it's just called virtual

[00:27:54] assistant right no no name nothing it is Santander sandy right really interesting i think

[00:28:02] HSBC is just welcome to our uh oh it's got a name they use it once or twice but it's not

[00:28:08] standardized if i remember is it that kind of thing that you're you're looking at yeah there's

[00:28:15] yeah the the language it's using is all important i think the ultimately the nuts and bolts you

[00:28:21] described it the tech itself we're going to get to the point pretty quickly where everybody

[00:28:25] can build something that's pretty competent pretty capable right and it'll be as good as it

[00:28:29] can be and the tech is the barrier to the tech is lowering all the time that that that is it's

[00:28:35] becoming easier and easier to deploy these kind of things that are incredibly competent i think

[00:28:41] deploying the right conversation AI to talk to the right audiences using the right language

[00:28:45] and the right tone of voice is is is the bigger win in a way fine tuning that model with

[00:28:52] thousands of examples of before and after use split into adjective verbiage set this

[00:28:57] length all the rest of it is what's key and that's the kind of work we're now looking at with

[00:29:02] companies from a language perspective at least look for for those who are thinking goodness me

[00:29:09] we should be having a conversation uh what's the best way of reaching the company and then

[00:29:14] reaching you so our address is our web address is this is definition.com my name is luke

[00:29:21] budka bu d k a so you can email me luke.budka at thisisdefinition.com or find me on linkedin

[00:29:29] which is the the linkedin domain name forward slash the generative generation which is a

[00:29:35] prodigy shout out right it's a wrap yeah i'm fairly i'm fairly noisy on linkedin so you'll

[00:29:40] find me going on about stuff on there as well but yeah feel free to reach out through the

[00:29:44] website always keep the talk to brands who are looking to uh sound and look uh a little bit

[00:29:49] different when it comes to using conversational ai and generative ai in general fantastic look thank

[00:29:55] you so much for taking the time right thank you and uh yeah all the best cheers back bye