How AI Is Reshaping Business, with Justin Collery

How this transcript was made

Transcribed by machine from the published audio, then corrected by hand. It is not a verbatim record: the changes below were made, and are counted.

This episode

45 speaker turns, 226 changes.

• closing CTA identified as CTA 2 - Subscribe (v4) (match 100%), replaced with the authored script and attributed to Max; the 17 words spoken before it stayed with Justin Collery rather than being swallowed by the substitution

What was changed in these transcripts, and what was not

Generated: 2026-09-20 from 2026-09-20-named/, itself transcribed from the audio that is live on the feed. Every change below is mechanical and counted. No sentence was reworded, no claim altered, nothing removed for being awkward.

The editorial decision

Owner's choice, 2026-09-20: light clean. "Um", "uh" and stammered word repeats ("and, and") are removed, because nobody intends to say them. "You know", "I mean", "kind of" and "sort of" are KEPT - they run to roughly 2,300 instances across the catalogue and stripping them would make the page read as somebody else.

Counts across the whole set

Change	Count
stammered repeat removed	667
um / uh removed	654
space before punctuation	241
split contraction	132
guest turn named (Joe Casabona)	101
organization -> organisation	71
diarisation fragment returned to the host	62
ai -> AI	57
empty turn dropped (was filler only)	36
synthetic outro restored from script	21
organizations -> organisations	18
hyphen followed by space	14
realize -> realise	14
A.I. -> AI	10
recognized -> recognised	9
gray -> grey	9
summarize -> summarise	9
color -> colour	9
behavior -> behaviour	8
synthetic intro restored from script	8
analyze -> analyse	7
behavioral -> behavioural	7
recognize -> recognise	6
percent -> %	5
personalized -> personalised	5
skeptical -> sceptical	5
analyzing -> analysing	5
center -> centre	5
optimization -> optimisation	4
realized -> realised	4
favorite -> favourite	4
realizing -> realising	4
canceled -> cancelled	4
recognizing -> recognising	4
announcer turn relabelled from Guest	4
summarizing -> summarising	4
labor -> labour	4
behaviors -> behaviours	4
colors -> colours	4
organizational -> organisational	4
customized -> customised	3
analog -> analogue	3
optimize -> optimise	3
honor -> honour	2
optimizing -> optimising	2
summarizes -> summarises	2
personalize -> personalise	1
prioritize -> prioritise	1
prioritized -> prioritised	1
organize -> organise	1
rumor -> rumour	1
organized -> organised	1
capitalize -> capitalise	1
centered -> centred	1
organizes -> organises	1
emphasize -> emphasise	1
split name repaired (Schaefer)	1
humor -> humour	1
traveling -> travelling	1
defense -> defence	1
maximize -> maximise	1
Organizations -> Organisations	1
apologize -> apologise	1
digitized -> digitised	1
monetize -> monetise	1
analyzed -> analysed	1
narrator opener re-attributed to Max, words left as transcribed	1
theater -> theatre	1
specialized -> specialised	1
split name repaired (Wadsworth)	1
aluminum -> aluminium	1
specializes -> specialises	1
skeptic -> sceptic	1
customize -> customise	1
personalization -> personalisation	1
optimized -> optimised	1
visualize -> visualise	1
spoken domain -> humanspark.ai	0

What was deliberately left alone

• "Program". A computer program keeps that spelling in UK English; only a broadcast or a scheme is a "programme", and only a reader can tell which was meant.
• "License" and "practice". Both have a noun/verb spelling split in UK English that a rule cannot decide.
• Capitalised words mid-sentence that would otherwise have been respelled: 4 of them, most often part of a name. Left as spoken.
• Retired show and brand names. Spoken by the host and guests. A transcript that removes what was said is not a transcript.
• "Co-pilot". Sometimes Microsoft's product, sometimes the metaphor. A rule would corrupt the honest ones.

The synthetic segments

The opening voiceover on the archive episodes and the closing call to action are read by Max, the show's AI narrator, from a script written in advance. Transcribing them means guessing at words already on file, and the guesses were wrong in places - one opener came back as "an AI that boned him up" where the script says "phoned him up". Those passages carry the authored text instead, and are attributed to Max rather than to the guest: the diariser groups Max with the non-host speaker, which would otherwise put a real person's name on words they never said.

Alastair McDermottSo today we're going to be talking about how AI is going to change everything and I think this is a topic that we're not seeing a lot of and I see people downplaying the impact of AI and I

kind of think that people are missing the big picture here because this is very very significant like sci -fi movie significant and we're going to talk a little bit about that with my guests

and today my guest is just Justin Collery. Justin you're welcome to the show. Hi Alastair, great to speak with you.

So I love your show with my friend Frank, you guys have a show called The AI Argument and I listen to it regularly and so I'm going to have Frank on this show as well but I know that you

have a lot of great opinions about AI so I want to talk to you about it.

Justin CollerySure I would like to and I'm great to hear your intro because I totally agree with you that AI is I mean I cannot think of a single business and or you know a single venture that isn't going

to be completely transformed in the next five or ten years I'm

Alastair McDermottKind of shocked to be honest when I see conversations about AI I'm just a little bit shocked at how people are kind of continuing on as if you know nothing has changed where this this huge

thing is coming and it's going to change every level of our businesses every level of our society like it's going to transform the human race I guess like I don't want to I don't want to

overplay it either but like it's pretty big

Justin CollerySometimes Alastair I feel like Leonardo DiCaprio in Don't Look Up and I'm like I'm pointing at the sky and I'm saying can you not see this thing it's coming it's right there it's absolutely

inevitable that this thing is going to come and we need to start planning for it right now and I wonder why it is that organisations don't seem to grasp it.

I think there's a lot of plumbing and there's a lot of you know infrastructure that needs to build up around the AIs.

for you know let's say the person on the street to really start to feel it as being real.

I do worry for people that don't recognise this that it's going to come as a shock to them when changes really start to happen but I think it's incumbent on our leaders right and the people

who are in charge of businesses to really come to terms with what that change is going to mean and then to set out a vision for you know what is my business going to look like in five years

time and how is this technology going to change it and what should I be doing today in order to plan and to take advantage of the opportunities that this change is going to bring.

Alastair McDermottBut I was talking to somebody who tried ChatGPT and it wasn't very good so AI really isn't all that good.

Justin ColleryYes so the thing you have to remember here and I'm guessing that that person tried ChatGPT maybe a number of months ago when they were using ChatGPT 3.5 and so now they've made ChatGPT 4.0

freely available to everyone you get a limited number of goes and I have to do a huge shout out to Anthropic who just yesterday released Claude 3.5 which is the best model the best model on

the market and I would really strongly say to that person go and try it again and use it.

So there's that sort of level of interaction and there's a beautiful graph to think about you really need to you know get your hands around or get your arms around is if you look at the

state of the art back in 2019 five years ago that was when there was you know ChatGPT 3 and that was terrible you know but it was it was terrible and yet already the researchers were saying

this model is so powerful that we're not going to release it openly because we're worried that it can fake emails and fake human interaction too well so they were already seeing it we then

had ChatGPT 4 and that was only you know maybe 15 16 months ago okay and that was where people like me really started to get excited and get interested in the technology and go okay this is

a game changer and then yesterday you've got Claude 3.5 so you can see that there's a progression of capabilities there's a progression of speed and there's a huge reduction in cost and it

doesn't appear there's nothing to say at the moment that that's going to change and certainly there's an awful lot of money like enormous resources are being put behind the development of

the models the training of larger models the deployment of the hardware and all of this is going to lead to a reduction in the price and an increase in the performance so you have to

consider what was the performance five years ago with GPT-3 what was the intermediary steps that we got to to get to today and do you think that that's just going to stop or do you think

that you know we're going to get a GPT-5 and a GPT-6 over the next five years and what are the capabilities likely to be for those models

Alastair McDermottI think that so there was a there was a document released by a guy called leopold ashenbrenner called situational awareness he published it it was like i i think maybe it was you that

described it like a it's like a phd document it's about 200 pages and it's describing the near future development of the technologies of AI and he's deeply deeply in that world he's a very

smart guy and he's just talking about how the there are no barriers to the capabilities to the development of capabilities i mean the only real barrier that i can see that we could really

face is lack of power like literally electricity everything else can pretty much be solved and so what what he's talking about is how the capabilities in terms of the reasoning the effective

iq of the system its ability to comprehend and understand and being trained on more training data that they will be getting better and better and better and so when i'm talking to people and

i i hear them say you know i tried it and wasn't great i find it a bit frustrating because it's literally right now the worst it's ever going to be and right now it's pretty damn good you

know when we look at the new release from claude and claude even before 3.5 was really really good definitely my favourite for writing text but now with the jump to 3.5 i think that it's

probably leapfrogged with my very basic testing i did this morning i think it's probably leapfrogged chat GPT-4 so or chat GPT-4 oh so these systems are continuously improving and and that

that pace of improvement is not going to stop it's going to continue like this for the next year for the next two years for the next five years and that's what i think is going to change

everything because it's going to change it's going to change the way that we do business it's going to change the way that our society runs because we're simply not going to need as many

employees and like that's going to have a massive impact on how our entire society

Justin ColleryFunctions yeah and it's i mean there's a whole look there's a couple of things that i'd say to you around that so yeah that paper it was amazing and i would i totally agree with you the

power is the biggest impediment to progress and there's a couple of other papers that came out recently as well that would sort of say the similar thing so there was another paper where they

got they got an AI it wasn't it wasn't a ChatGPT type AI but it was they would train the AI to play chess and so in chess you've got this idea of elo so that's how good you are are playing

chess and they only trained us on chess players they trained it using human chess players and they only trained it on chess players up to 1000 elo and then they got the program to play chess

itself right and what they found was that it would get achieve a standard of 1500 elo so even though the training data only went up to a thousand the actual AI afterwards went far beyond

what was in the training data and so what it was doing was it was basically if you think it was the wisdom of crowds so it had all this data from you know various chess players and obviously

they made bad moves every now and then but it was able to take out the bad moves and supplement it with a good move and that increased its overall capabilities now i think when you apply

that same idea to all of the internet knowledge so if you think right of all the different fields so let's say i'm an expert in computers and you're an expert in marketing and somebody else

might be an expert in some other fields right the ability for us to progress the state of current knowledge is limited by our own fields right we only know what we know ourselves it's very

hard for us to put bits of information together but a large language model can and so i would expect that using a similar thing where you've got like a chess AI that can achieve 1.5 times

the capability of the data that it was trained on because it's averaging out that data or it's taking learnings from different people a very similar type of process will happen with just

general knowledge and and so on so that's that's that part the other thing i'd say to you that's that's complementary to what you spoke about there is ilia stutzker so he just in the last

number of days has set up a new company called safe super intelligence ssi for short and what's interesting to me about that is his original goal in open AI was to create agi artificial

general intelligence and his new company which will have no products and is not going to release anything to the public until they've achieved safe super intelligence is he's moved beyond

agi it's no far no longer artificial general intelligence he's now focused on artificial super intelligence so that's intelligence beyond that of a human and in that paper a situational

awareness i mean in in the interview that that leopold gave he said it was it was quite clear the path to agi like that he knew and a number of other people knew what would need to be done

technically in order to achieve that so so that's where we are today and then we come back to what you're talking about you know about employees and stuff and there's an awful lot we can

unpack there right but the first thing i would say is if we go back to your example of the person who said oh i've used ChatGPT and i don't think it's very good well the first thing is that

where this makes an impact is not a chat interface right that's not where it's really going to make a difference it's going to make a difference in the background the best AI that you will

ever use is the AI that you don't even know is there so for instance if you are i'll give you a use case actually i think we talked about this before so a real practical use case is if there

was let's say a service where i can just send it an email saying here i need my utility bills are are up to be renewed can you just here's all my bills can you just go and find me the best

deal and in the background it uses an agent to go and it looks at all the websites looks at your bills and finds out you know which is the best deal and then just goes and switches you right

and actually does it automatically every time you know it just constantly monitors and goes oh i see energia have released reduced their prices i'm just going to switch you so you do nothing

right it's not a chat interface it's not you saying i want you to do something it's an agent in the background that i've kicked off to say look can you just make sure i always get the best

deal and it does and that type of thinking right where you have an agent in the background if you apply that to your organisation that's where enormous efficiencies can be gained so if you

think about your organisation and you know whatever i don't care what it is you do and this is why it's going to change everything right it doesn't matter what your organisation does but

just look at every step every piece of friction every place where somebody has to do a piece of work and break them down into little bits and then you just start to automate them using

agents and the ais that we have today the ais that exist now ais ais like claude 3.5 are absolutely capable of automating not everything but definite parts and if you follow that path

forward you know to next year and the year afterwards i mean there was an interview with the cto of open AI just yesterday where she said that in the next 18 months they will release an AI

that is the same intelligence as a phd student now i don't know any organisation that employs only phd students but in 18 months that's what you'll be able to do so again i come back to the

point as leaders you know as business leaders you really need to start thinking about how is it i'm going to manage that transition from where i am today to having phd students on call that

i can just fire up to any tasks that i want them to get them to do

Alastair McDermottYeah and i so i i think like we're talking about some of the really practical stuff here and i love talking about this well because i you know there's there is the big picture there's the

utopian potential future there's the dystopian potential future but when it comes down to it like what we need right now is we need to start implementing this day-to-day in business and the

way i see it we have some pretty powerful tools already at our disposal but the glue between those tools that the the the bits that make them work together are very clunky you know me as as

a programmer as a coder i've been trying to set up some of these tools here locally and it's just it like frankly it's a pain on the ass to set them up and and that's me coming with a with a

tech background i can't imagine somebody who doesn't have a tech background setting up you know setting up current agents in the state that they're in currently but yet i know that agents

will probably be i don't know if they'll be more important but they'd be as important in in in the AI sense as the large language models because they're the bits that like they're the hands

that will allow the large language models with all their reasoning to actually go and take actions by you know visiting websites and doing things so how do you feel about

Justin ColleryWhere agents are at and where they're going i agree with you there's there is an enormous opportunity for somebody somewhere to make you know a piece of software which makes firing up agents

easy right and and repeatable and i mean to me that's going to look like a sort of a a box a gooey where you more or less draw a process and you know it's got certain inputs it's got certain

outputs and you can probably chain them together i mean and lang chain doesn't do that right that's a sort of a technical product it doesn't do it and there's there's other products i

haven't seen anything yet that does it but there are practical steps that i think every leader should be taking right now in anticipation of a product like that becoming aware so if i look

at and this i think is especially true of small organisations rather than big organisations right so the promise of agents and the promise of AI is that with a 50 person a company i can

actually do the work of 500 or 5 000 people right that's that's really the dream right how do you make your developer a 10x developer or how do you make your salesperson a 100x salesperson

right and you you need to anticipate that that's going to happen right that's going to be possible in the near

Alastair McDermottFuture and and by the way not just in the near future like we've already seen a case study from open AI in conjunction with moderna the company who make vaccines and they've been using it

deeply over the last year 18 months and they the ceo or the cto of moderna said you know we're going to be able to have the capabilities of a company with a hundred thousand employees with

you know less than ten thousand so like that's that's the way they see it now in that case study it's really interesting to see all of their employees are on average having i think was 120

conversations per week i i i did the math and it was something like they each employee talks to ChatGPT once every 20 minutes in their workday which is insane like and that's the level that

they have it integrated and what that allows them to do is have this massive impact on productivity not doing grunt work employees are no longer doing grunt works they're happy about the

work that they're doing and they're producing better work okay and

Justin ColleryThat's a really good i mean moderna are doing a stellar job they are a poster child of how a leadership within a company you know should start to implement AI within their company so they've

just given chat gp to everybody and in that case study what was very interesting to me was that the legal department used it 100 the legal department and i think in a lot of bigger

organisations the legal department can be an impediment to adoption of this type of technology so that was that was amazing to me right but the next step from that right so just giving gpt

in the chat interface right that's yes you're going to get a big bump in productivity from doing that but the next step is automating tasks that's how you get that you know 100x performance

gain now then as a leader you have to ask well what is you know what is it my business does right how do i serve my clients and what is it that i know right and the hardest but yet most

valuable information in any organisation is tacit knowledge knowledge that's in your head and it's knowledge that's in my head and so leaders i think should be thinking about ways to capture

tacit knowledge so we're having a conversation here and as we talk about things you know there's lots of questions being asked and answered right and in doing so actually what i'm doing is

i'm giving away some of my tacit knowledge easiest thing in the world to do would just be to record every meeting just record it and store it and keep it there so let's put it in the context

of i'll give you a practical application now right let's say we're developing a software product together and you're the business person and i'm the technical person and if we're using an

agile way to develop software there's going to be a lot of over and back between you and me where you say here justin i want you to do this and i'm going to say well why do you want to do

that would you not do it a slightly different way and you'll explain to me the business logic behind why you do something a particular way and eventually we'll agree to okay this is how

we're going to implement it now what i've actually got there is i've discussed the technical details and the reasoning right and you've discussed the business logic behind it that's

incredibly valuable information and often what happens is how does that get captured today in an organisation right you might have a couple of bullet points from you know meeting minutes and

maybe there's a couple of jira tickets that get you know created but all of the really important knowledge right your business knowledge and my technical knowledge is totally lost and then

six months from now right there's a terrible accident and both of us get run over by buses and two new people come in a new business person and a new technical person how are they supposed

to know why we did things in a particular way because that tacit knowledge has been lost so the most valuable thing you could do is just record every meeting and just store the transcripts

and over time what you'll find is you'll build a huge body of knowledge that's going to be incredibly valuable for when it gets to the point that you want to start automating your processes

because then i know right so at the moment i can use it as a as a look back to it so i can say why did we do that thing like why do we make a decision to do x y and z and i'll be able to

know what was the reasoning behind it and so long but in future what you'll be able to do is say this function does this but now i need to augment it to be able to do something more and the

AI will be able to go back and look at the tacit knowledge and understand why something was done in a particular way which is going to make it more able to take the next step and this is

kind of like a flywheel where you know the more you know the better you are doing things faster so that's the first thing right that's super important that i just think is is kind of lost

it's like such an easy thing why aren't people doing it now and the second thing is you have to reimagine your your company and your business from the client's point of view what points of

friction are there and you know because if you don't do this somebody else is going to do it and they're going to and i'll give you a practical example in an insurance setting so and some

companies in america have this already but let's say you're insured and you've got car insurance and i have a crash right i should be able to take a couple of photos fill in a couple of

forms you know that can all be checked by AI it can look at the photos it can check the forms and it can check your policy and it can it can pay you straight away it can just i mean if there

was an insurance company out there that said if you submit all your paperwork it's all done electronically and we guarantee to pay you within an hour well that's a huge point of friction

gone for me right that's a big win for me as a client and so all of us in our businesses have to look at what we do and how we do it and reimagine what those businesses are going to look

like in a world where AI is totally pervasive and the person answering your phone call is a phd

Alastair McDermottStudent i love that as a podcaster so i've been podcast ing for four or five years now and i i took my other podcast recognised authority and i took the transcript so i have 175 episodes of

that podcast and all

Justin ColleryOf

Alastair McDermottThose are in-depth conversations of about 35 to 45 maybe even an hour long with people who are absolute experts in their field and you know people like you know best-selling authors you know

really top class people and i was able to take that transcribe all of those and then get AI to distill and pull out all of the key insights and so i was able to then take that create a

spreadsheet with all of the key insights referencing the amount of times that each insight was mentioned and then i was able to create like a league table of all of the top pieces of advice

from all of these people and it's so powerful when you have transcripts another use case just for transcripts and this would would apply in a business setting this that where i used it was

in a in a sports club we run a charity run which is a funding fundraiser for the club for the the rugby club here in sligo and so what i did was with agreement of the other people at the

meeting during the planning meetings i would record the entire meeting and i was then able to take the transcript turn that into notes and then turn that into a field manual for running the

entire because when you're when you're running a charity run there's a lot of moving parts about you know getting medical getting guard approval changing the bus schedule because we're going

to be blocking off the bus there's so many different things how many marsh als do we need where are they going to be lots and lots of little details that you kind of forget year on year we

were able to take the meeting notes from three or four of those meetings and now we have a manual where it's printed out and it's like this is everything you need to do here's when you need

to do it and we don't need to remember that anymore you know it makes life so much easier and that's a great

Justin ColleryUse case right and so let's take it to the next step right so you have all your manual then you can say look can you come up with an action plan please tell me what is it that we need to be

doing and when do we need to do it right so that's your co-pilot right for your event and then every right or every week you can just go okay what is it that we need to do this week and it's

there just guiding you through the process and as issues come up say oh look at this thing happened what should we do about it i mean think of how easier it makes just to organise a simple

charity event to do things like that yeah but that's totally applicable in every scenario there is nothing in business that i can't imagine that would not benefit from a similar type of

Alastair McDermottSetup yeah yeah absolutely now so then we say okay so now we're we agree on the value of transcribing calls transcribing meetings we're talking now about a societal change where everything

that we say at work is being recorded all the time and transcribed that like that's a lot for people to wrap their heads

Justin ColleryAround like do we really want to do that yeah i think well you see i think one i think that there is okay there's a societal change you're totally correct right so and i think a positive

societal change that's my personal view and i think what happens is you get a very strong delineation between on the record and off the record conversations so when we're having a serious

conversation it's going to be recorded and you understand that it's going to be recorded and everybody knows right in order to facilitate a good meeting there's always the 10 or 15 minutes

beforehand where you're catching up with people and you're just sort of you know having banter and maybe just sort of everybody's getting relaxed and you're not going to record that part

right and then you're going to hit the big bread button you're going to record and then you've got your business meeting and then you end it and you have another conversation and off you go

so that's that's one change i think that will happen but the positive thing that i see that one of the things that professionally frustrates me is that it cuts out a lot of politicking for

instance within large companies and you know it's because again i imagine in the future and not the distant future i imagine these AI's to be active participants in the meetings so if

there's a disagreement or something about why a decision was made or was this discussed previously or you know why have we done a thing this way you could just turn around to the AI and say

here you know give it a name alfred hey alfred can you tell me when was this discussed before and what was the decision and that just sort shortcuts a whole load of arguments you know we're

not it's not like being down in the pub with your mates having a couple of points where it's fun not to have your phone and to argue over who won the world cup in 1966 right it's you know

it's it's a different type of conversation in a professional setting and to me like what is what is the purpose of a business meeting right what are you trying to do right you're trying to

get to the truth in the shortest fastest time that you can you want everybody to be heard but you don't want to be going around in circles so i think this is a great aid to that and i but i

do think it changes the culture but not in a bad way i look forward to it the okay so

Alastair McDermottTranscripts clearly a great place to start and i know like when when we got on our last call we were on zoom we're not on zoom right now we're in a recording studio but we were on zoom last

time and i brought my note taker in it recorded the whole meeting and it showed a little message that came up and said hey Alastair is recording this meeting for note taking purposes and you

you saw it and you just carried on because you know that's that's how it works right yeah that's going to become ubiquitous that's going to be something that just everybody does in fact i i

see it now where i'm having zoom meetings and somebody's note taker will join before they do you know

Justin ColleryIsn't it isn't it interesting that it isn't it interesting to sort of a serendipity or synchronicity that covet happened and we all became so familiar and so easy with having meetings over

zoom where it's pre-covet you know if somebody joined over zoom it was kind of oh would you not make an effort to be there or whatever covet has trained us all to be totally comfortable with

having a conversation like this and not in person which is just a happenstance of history i i love it because

Alastair McDermottI've been doing this for a very very long time one of my first jobs out of college i work for a multinational called sun microsystems and my boss was living in kansas i was in dublin one of

my teammates was in london in the uk and we had teammates in the east coast of the us we had teammates in austin texas and in san francisco so that was back in 2001 and and we worked

remotely all the time i'm really glad the rest of the world caught caught on and caught up with that because it was it's just such a great way to work i i like the fact that i can work from

a beautiful place in the west of ireland and still work with people all over the world but it it certainly has paved the way for this kind of remote interaction it's also paved the way

potentially for rather than people to appear themselves it's potentially your avatar could be joining a call at some point in the future and so your avatar is an AI who looks like you and

has access so you're you're alfred so it could be it could be just an alfred and they join the call and say look sorry justin can't make it today but i'm here to answer any questions that

you might have

Justin ColleryFor him you know again that's a really good use case right so we're all really busy right i'm sure that it certainly frustrates me when you know people stomp on your calendar and you've got

five different meetings at the same time and you can't go to all of them so you can have your avatar turn up for that meeting exactly as you say it the avatar is there to answer questions

but the avatar is also there to hear things from your perspective and from your ear so your avatar knows what you're concerned about and you know it knows everything about you and so at the

end of the meeting you can imagine you know think about your routine today right the first thing that i do certainly is i come in check my messages check my emails see what's lined up for

the day and then get working you can imagine that you're going to have a similar routine at the end of the day where you have a conversation with your avatar and to say okay all the meetings

that i missed is there anything that's really important what do i need to know and it sort of gives you that information so it's again it's a great way of enabling silos in big organisations

is a big issue and it's a great way of breaking down silos because now i can be in five different places at the same time and then just get the information i'm not getting meeting minutes

which are kind of bland i'm not having to listen to an entire recording my avatar knows what i'm concerned about it knows what i'm interested in and it knows how i like to consume that's

another great use by the use case of AI which is you know you could take your content as a podcaster but also in a professional setting different people consume information in different ways

i like bullet points other people like stories some people you know like they like the information presented to them in a different way you could take your podcast and you could disseminate

that tailored to the individual consumers desire on how they want to receive that information and that and not of course in the language so even if i'm in a multi-language multilingual

organisation and there are people and you know they're all speaking spanish or whatever my avatar can listen in interpret the entire meeting and then give it back to me in english which

again is a great boon for productivity and just being able to allow people to interact with each other

Alastair McDermottAnd your avatar can speak every language fluently yes that's that's another big one so like i know that a lot of this kind of sounds like sci-fi to to people watching or listening but a lot

of the sci-fi is very close or is actually reality already you know i can already take a video recording of say this this podcast and i can put it up and upload it into a tool which will

then translate it and have me speaking with my mouth synchronizing with the words correctly in fluent german or japanese or spanish or any language you can think of and that like that's

that's all possible with today's technology there's nothing there now i'm not saying it's easy or it's quick and and so some of these things can be a little bit slow and clunky and you know

i know that one of the frustrating things that people find is they see some amazing stuff that's been created with AI particularly some of the AI generated videos or the AI generated images

and then they go to use the tools and they find that what's getting output it doesn't look anything like what they thought was going to get output and that's because people like frank

prendergast who's our friend who who will also be on the show at some point people like him learn how to use these tools and it it takes time and also they iterate they they it's a trial and

error right now with these systems but that they are genuinely producing this output and these systems are going to get better and better and quicker and quicker

Justin ColleryAnd answer we've been here before right we've been here before many many times so if you're in you know anywhere in 1998 and you're trying to buy a book online right it will be very hard for

many different reasons you would be probably using a dial-up connection which was very slow and didn't work and was expensive your computer monitor was you know not very you know good and

took up half your sitting room and the websites probably just had a picture of a you know a toaster flying across the screen or something like that the infrastructure wasn't there and it

would have been very difficult for you but yet five years later in 2003 amazon existed and you know the the process of of being able to buy books and do commerce online had moved forward and

again i come back to you know you say you're shocked at how little people understand just how much this is going to change their lives and i absolutely agree with you i am constantly shocked

and it seems to me that we're what we're lacking at the the moment is a vision of the future and you know do leaders do business leaders have a clear vision of how this is going to change

the way that they interact with their clients and so again i go back to you know to to that you know that dot-com type period where you had barnes and noble versus amazon right so barnes and

noble sold books and you know they thought the best way to serve their clients was to have beautiful bookstores that you could go in and whatever maybe even buy coffee and stuff like that

and amazon didn't they had a totally different vision of the future which was that a book is a commodity and the cheapest easiest way to buy a book is on the internet with one click and they

realised that actually they're in the logistics business which is that as soon as you click something has to happen really quickly and you know nowadays they've got it down to they will

deliver to you within the same day or if you look at you know the difference between blockbuster and netflix the classic example so again i'm sure block blockbuster had a website but they

didn't you know take it to the next level and say well actually the easiest way for our clients to interact with our content is just to stream it straight to their computer people are

inherently lazy at the end of the day they want to take the easiest route to get something done and so from a leader's point of view how does this revolutionize the way that i interact with

my clients how has this happened in the past and and then just to your point about you know there's lots of companies and you know it's hard it's clunky because again it's like the internet

back in the 90s and there's an awful lot of money which has been plowed into trying this is this is the really creative phase of the technology cycle and i'm absolutely sure that we'll go

through a dot well dot AI bust because what happens is you get enormous amounts of capital being thrown at thousands of companies and they're all just trying to figure out out what the best

way is to interact with this technology how do you integrate integrate it into our processes and it's hard and 90 of them will fail but the 10 that succeed will be truly successful companies

and so a visionary leader today you know will be thinking deeply about how do they put

Alastair McDermottThemselves into that 10 so one of the things like it is cl unky right now and like i remember you know i was building websites i think you were building websites three or four years before i

was i was building websites in around 1995 96 and like that was the the early days that that was the the dawn i think you were you were in there pre-dawn building websites 93 was there yes

well like it was hard to build websites back then and the only people who could do it were people with a software development background because the tools didn't exist we literally had to

hand code things and i kind of feel like that's where we're at right now with things particularly like agents which are going to be massive but are really clunky to use right now and only

people like software developers can build them right now but that's going to change so the the question i have based on that is why why why should a business leader not just wait like why

why should they start to take action on AI right now when it's going to be much easier to use it in 6 12 18

Justin ColleryMonths time so i mean that's a great question because sometimes doing nothing is the right thing to do okay but but but we've already discussed right that even if you don't make technical

moves to take advantage of the technology there are practical moves like transcribing all your meetings right capturing the information distilling it those are practical things you can do

today my view of any organisation is that you should take you know you should be giving the technology to your employees so even if you're not going to you know change the way that you

operate today give them ChatGPT or give them claude you're going to make your employees more productive just by doing that and what i think is you know there's going to be a push and a pull

within an organisation so you're going to have from the top down right you're going to have leaders are developing a vision of okay this can really change our business and this is how we're

going to do it and that's going to be the pull right but i think if you give tools like ChatGPT to everybody in the organisation there's also a push where employees are going to you know

will recognise quickly which are the repeatable parts of the processes that they're doing and then you go cool let's write a piece of software that just automates that one piece of the

process because you're just doing the same thing and your employees will tell you which bits those are and they don't want to be doing boring monotonous stuff over and over again they want

to be doing interesting work so there's going to be a push and the pull so there's there are practical things that you can do today knowing that the tools of tomorrow are going to automate a

lot of that work

Alastair McDermottThere's two things i want to follow up on that the first is that you know making these tools available to your employees so microsoft found in a study with linkedin recently that many many

people are already using the tools whether you give them or not they may be using those tools themselves and this the they call it bring your own AI to work and it's this is a problem

because you know if you're in eu or you have eu clients you're probably violating gdp or if they're putting in client information in there we already saw a thing where samsung employees put

meeting reading notes and transcriptions into ChatGPT and it had proprietary i think had client information and it had proprietary intellectual property and basically they lost

Justin ColleryThe

Alastair McDermottThe rights to that because it was in the terms of service for for ChatGPT that they were using that they could train the model on it and and so these are the risks that people are they're

probably already using them and you know that's and and that's a risk so your employees are probably already using them if and even if they're not putting those tools in their hands is a

good idea but also having some level of control over it putting guidelines processes in place you know all like all of the really boring kind of governance stuff that's actually really

crucial because your employees are probably if you're not like your best employees are probably using AI already because that's the way they think you know they're they're good people and

they want to use the best tools available to them so that's that's my warning to business owners like they're probably using it already if you haven't made it available for sure right and

Justin ColleryThose are valid concerns right and i mean i would i would put on two hats right i'm going to put on my corporate hat and i'm going to put on my entrepreneur hat so on my corporate hat right

absolutely valid concerns you need to be really concerned about that sort of stuff and there are you know microsoft give you and open AI have enterprise licenses where they guarantee that

they will not train their models on their data they are medical compliant they are you know they will they will co-sign agreements with you to make sure that that data is treated in an

appropriate way and that they don't store the data or save it and amazon will do exactly the same for you as well so there are ways for you to mitigate those risks and i would say you just

have to you know it's imperative for your organisation to do that that's my corporate hat on my entrepreneur hat i would say to you there are enormous opportunities because you should know

that every really big organisation is going to be totally concerned about security privacy and you know all of these things which are going to slow down innovation in those really big

companies so if you're a small company you have none of those shackles you can innovate at a light speed compared to a big organisation so if you can spot an opportunity your chances of

selling a service or a product to a big organisation are never going to be better than in the next two or three years because your speed of development is going to be light years ahead of

the speed of development in a big organisation so there's huge entrepreneurial opportunities there also

Alastair McDermottYeah and and to to speak to the the corporate hat version of you i i completely agree and and like the all of the AI providers know that they won't get widespread adoption if they don't take

the the privacy question seriously like they know that that just it won't happen so they have to take it seriously but that only works if you were signing up as a business and you're signing

up to one of the to one of the licenses where they they offer that if you look at all of the personal level you know where you sign up as an individual even if you're paying usually there's

something in there that says we're going to train on your data unless you tell us otherwise and the risk factor for you as a business leader is if one of your employees is using one of these

personal level accounts because you haven't you haven't put the the the the professional level tools in front of them that's that's where i see the risk and that's why i think it's really

important for organisations that i like at the enterprise that were even like small businesses who have a number of employees that's why i think you have to put put the tools in their hands

and have control over them because and have that guidance and governance

Justin CollerySure i mean and there's a number of ways you can do it now i would push back a little bit and say that anthropic in particular have guaranteed that they will not train on your data unless

you specifically opt in to have have your data used to train their models so you know if you're concerned i'd certainly would take a close look at anthropic and of course you can always

block those websites at a corporate level say we're not going to give you any access and you can provide you know through apis you can you know now i would argue that's not a good way to go

because i think you know again what what makes again if i come back to the the the what's going to happen to many many if not all industries which is there's going to be a you know a

rechange a changing of the guard and what's the difference between a big company and a small company so if you've got moderna doing the work of a hundred thousand people with ten thousand

people and then you've got pfizer who actually have the hundred thousand people well what's the difference right why can't moderna kick pfizer's ass with their ten thousand people and the

reason is history right it's it's all of that knowledge right it's all of that you know again tacit institutional knowledge that moderna just don't have because they don't have that number

of people so if if i am moderna i don't want to be rewriting chatbots right and and rewriting stuff which i can get from free from microsoft or from open AI or from from anthropic i need to

be focusing on where does this deliver value so the idea i mean i could control access using an API i could block access to all of these tools and say you only are allowed to use it through

this API i don't think that's the right way to go though i think you just have to say what is the best protection i can get using the tools out of the box that i can get my hands on and once

i've done that then how do i actually operationalize the automation of tasks within my organisation because that's where the value really is there's one last topic

Alastair McDermottThat i want to discuss because i i think this is like discussing the the kind of the the one of the most important things is operationalizing i don't know is all of this information all of

the documents all of the transcripts so you've already got all of your like let's say you've you've got your marketing department has a bunch of documentation about marketing you've got

product spec sheets you've got internal documents about your products your services all of these documents are in different places and need to be brought together into some sort of single

repository of of or maybe not maybe you keep them siloed because you you don't want you know you want to keep them separate because you want to give different levels of access because maybe

you've got you know different levels of security within your organisation like you don't want to put your all your employee salaries for example in there because you don't be able to pull

that kind of stuff out so i think that document management and updating this internal repository is going to become very very important the way those documents are sorted named laid out

internally within the documents so that an AI can read it and understand the context and the priorities can you tell me like like have you thought about this how do you see that is this a

problem

Justin ColleryIt it is a problem it is a problem as in the concern that you've voiced is an absolutely valid concern right so but it's a problem that's been solved already and we just need to piggyback on

that solution so i i'm a strong believer that you're the problem is that there's all this data and it's in different places but there's some data that you've got access to and there's some

data that i have access to well that's already been solved as an issue right so we have sharepoint we've got ad groups we've got you know security profiles there's all sorts of different

ways where we manage people's access to data day to day already you use exactly the same mechanisms again so what i would imagine that the technical solution to this is that i will have this

thing called a vector database and we don't need to go into it but it's it's you know a key component of being able to look up documents and and pull out relevant information to help the AI

answer questions so there will be a vector database which sits on my machine and that is a vector database or you know a knowledge store of everything that's on my machine and if you've got

a share point site there's a vector database on the sharepoint site and that contains all the knowledge that's on that sharepoint site but you know only within let's say that particular

because sharepoint sites are split up into different areas with different areas levels of access so if i have access to that share point site then i also have access to that vector database

and when may my when i ask a question of my AI it knows which vector databases it can communicate with in order to get the answers back so this is a free it's a gibby right there's no extra

work to be done to control your information except me i said an interesting question a point for you to just do what you do already right just make sure that the users have the correct level

of access and then just have the AI only have access to the same information the concern that i hear in some corporates is that we don't know what people have access to so we're worried that

if we create in this instance a better search engine for internal documentation people will suddenly start to find out that they do have access to everybody's salary even though they didn't

know it because there's some loophole in the security and the issue there isn't the AI the issue there is your security so you've got to tackle the problems at their root and i i would say

by the way that i i did discuss this with an architect in one organisation and they were like you know a vector database on on machine because i'll give you another great use case for AI is

if you are let's say working in many different industries right you get it you get a document a form in and you have to take the information that that a user has either written or typed into

a form and you have to transcribe it onto your internal system right and that that's just a regular thing a lot of people do that type of work data entry type of work now an AI will just

read the form and just pop it in it's the easiest thing in the world to do right and it would be a simple plug-in and they're like oh AI running locally on your machine i can see lots of

issues around that and people you know it just feels wrong it's inertia is a huge issue to the adoption of AI but if you look at what open AI have done recently is they have released a you

know a local version of open AI for your mac where it can look at your screen and it can do stuff and they'll soon i'm sure have one for windows as well and it's a very logical next

progression that they will then have a store of information at this thing called a vector database locally on your machine and it will interact with that store so as you're chatting with

ChatGPT even though it's on the cloud it will also be able to interact with all of the documents that you have locally on your machine and then as an extension maybe every network drive or

every sharepoint or everything but it's based on your access and that's how you solve the security issue

Alastair McDermottYeah well i i think that this will be solved i think that it may change how we format documents so for example i think that we may now need to add more context to a document that's not there

for the humans reading it but it's there for the AI reading it so that it understands okay this is connected to this other thing and here's some meta information about that because i think

that this will get very complicated because there's such a vast number of documents you know like i as a as a single a single person in in my business have generated you know tens of

thousands of documents over a decade and if you multiply that out in an organisation how many documents people are creating all the duplic ation some of them will have incorrect information

in there you know there's a lot of different issues there that will that will impact on this so i think sorting the like the actual format like the file format versus the internal format

like you know do you use headings is there meta information all of that will impact on how how efficient and how effective the the AI is able to

Justin ColleryProcess those documents and AI can help you do that so i've run a a project already where i took thousands of legal documents and so court cases and i've gone through and i've classified the

court cases identified the area of law that the court case related to the arguments that were used and you know whether the arguments were successful or not so you can use ais as a

classifier for information to in order to generate exactly metadata and that metadata might be stuff like when the document was created or when it was last updated because more recent

documents are obviously going to be you know more likely to be correct than older documents and then you can you construct all sorts of other metadata in it as well and and then you use

those correctly i mean as a way to get to that that's i mean a key issue here and it's a very technical one though is getting to the right information so the context windows of LLMs are

getting bigger and bigger and bigger and some people believe that that's going to be a solution to identifying you know the right piece of information so gemini has a context window now two

million tokens which is just enormous you know and so maybe you can just retrieve all the documents throw them into the context window and let the AI decide which ones are the right ones or

not and then other people and i'm i'm not decided on what the right way is here other people are like well i'm sort of intellectually drawn to this a bit which is rather than giving it an

enormous amount of text right just give it what's really important and ask it to focus on those bits and the right answer isn't known yet right that's why this is an exciting time because

it's everybody's everybody's has an opportunity to make a real difference here because we don't know what the

Alastair McDermottRight answers are yeah i i think like the way i think of it is the old rumsfeld expression it's unknown unknowns and that's like that's the the field we're in at the moment is like we don't

know what we don't know and not only that but the the unknowns are changing exponentially you know so it's it's an interesting place to be and the go poster

Justin ColleryMovie every six months i mean the biggest thing you don't know is what are the capabilities of these models going to be 12 months from now all you know is they're going to be better but

they're going to be they're

Alastair McDermottGoing to be better they're going to be a lot better i i think where we're lagging right now is the glue the pieces that are tying things together i can see the i can see the the capabilities

in terms of intelligence and reasoning and writing and matching our voice and all that kind of stuff i can see those improving you know massively but right now where we're missing is is the

glue pieces to tie these things together to make them more practical and more useful but that will

Justin ColleryCome as well i totally agree and a lot of those glues are internal anyway right so you know and so it's the glue and the tooling are the two things right that we're missing right now but

they will come and we need to prepare now right we as i said we need a vision for how this is going to change our business because the change is happening it's happening around us now we

don't be a boiling frog we need to think about what we're going to do over the next five years to make best use of this

Alastair McDermottTechnology well justin i think that's a good place to leave it and thank you for giving me so much of your time where can people find you if they want to learn

Justin ColleryMore sure you can get me on linkedin just look for justin collery i'm also on twitter if you look for justin collery that's where i post my random thoughts and or you can reach out to me

justin.coll ery at wi-pipe.com and definitely check

Alastair McDermottOut the AI argument which is your podcast with frank brand egast because that's great crack as well

Justin ColleryYeah absolutely Alastair it's been a real pleasure talking to you and hope to talk again soon

MaxThanks for listening to Something About AI.

If you are new here, there is a back catalogue worth your time and a lot more on the way, so hit subscribe wherever you are listening and you will not miss it.

And if you want to go deeper, Alastair keeps a growing library of free resources over at humanspark.ai.