Hugh McKee: Test Driven Prompt Programming: How AI is Changing the Way We Code

This talk explores the transformative impact of artificial intelligence on the coding landscape, particularly focusing on test-driven prompt programming. The speaker delves into the rapid advancements in AI technologies, emphasizing the significance of AI coding assistants such as GitHub Copilot and various other tools. They discuss their experiences as a Java developer utilizing these AI tools to enhance productivity and efficiency in coding tasks. Furthermore, the speaker introduces the concept of micromine services, showcasing a biologically inspired event-driven programming model that leverages AI's capabilities. The talk culminates in proposing a novel approach termed test-driven prompt programming, where code is viewed as prompts that define component interactions, signaling a shift in software development practices. Throughout the session, the speaker conveys an optimistic outlook on the evolving role of developers in an AI-enhanced future.

[Music] ladies and Gentlemen please welcome our next speaker humi presenting the topic test driven prompt programming how AI is changing the way we code all right hello everyone welcome to this talk I don't I guess I don't have to repeat the title again because they had this nice announcer to the title for me um but in this talk I'm going to uh try and challenge you a

bit we're going to look at some of the things that are going on around us in the IT world I'm a developer so really from a developer perspective and we're gonna I'm G to take a chance so it's my risk but hopefully your benefit if it turns out this way we're going to push bit a bit into the future so um as we know and and there's

a great great PE group of people here AI is is just been crazy last year was year that AI was Unleashed from the labs out to the real world and we F and all of us finally got our hands on it and last year was crazy there was just massive U series of it seemed it's never ending announcements and it's continuing into this year we're halfway into

24 and just like last week for example there were some pretty big announcements from both open AI on chat gbt 40 and and they always do this just before a Google event the Google event was the next day and then Google announced some pretty significant uh changes and improvements to Gemini so it just keeps going and going and it's just crazy so this all started late in

22 in November in 22 when open AI um Unleashed like I said on the world chat GPT and this is really after 70 years of development I mean we there's been people dreaming and thinking and working on AI in one form or another since the 50s since Computing pretty much started and it's been in the lab though for the most part up until you know this this

last year or so the uh the in the interesting thing though is that we we can see this pace of change happening and there's this anticipation that this pace is going to keep going uh so I all these diagrams that I have in my slides are AI generated mid journey is for the most part and I um so I I try and do a picture that's relevant

to what the the thing I'm talking about in the slide so I kind of like the way this came out even though it's pretty abstract but there's some graphs that kind of go exponential straight up but then some graphs that may be plateau and if you watch what people are saying and I've been watching this space a lot all last year I've been kind of obsessed with

AI and and things like that you know some people are saying we're going to have artificial general intelligence maybe this year maybe next year some people say in 2030 some people say well maybe you know we're seeing this kind of very exponential growth but maybe it'll Plateau who knows you know it's all over the map what's going on but we're just def definitely just getting started but

for us in it we know things are changing and I love this picture again this is I think it was Deli um that came out and this is kind of conveys the way I think a lot of us feel I know it's the way I feel was like you I feel like a s sand sculpture and the winds blowing some of this you know diffusing me and

what's all this debris that's coming off of this figure I feel like this is my skills and uh capabilities and experiences that I've had over Decades of being a developer maybe you know evaporating going away changing and really um think about this is that we're going to remember this decade this is the decade when everything changed is what many many people are predicting most of the world

doesn't realize this yet I think all of us here do to some you know varying degree but I think we'll look back in the future as the early 2020s the 2020s is when the world completely changed we went through a massive inflection point we've gone through these before but everybody that's really um and many people have said this is that this is a big one this is

a is a really big inflection so I know I was experiencing this and I I know a lot of other people are maybe you are as well as like okay as you start to internalize what's going on you get this feeling like wow okay am I G to have a job you know you know I get I write code am I going to be writing code next

year you know a few years from now what's going to go on what's this going to do to people that that uh in my family you my friends all these types of things so there's this kind of um Melancholy is a good word and you know this weird feeling but I'm an optimist so there's also this like wow this AI stuff man is so cool I I

just can't believe what we can do so I have a lot of uh hope for the future as well that we'll figure out how how to deal with this this feeling actually has a name it's called vespin the three QR codes down here are for three videos the the the person that coined this name is a YouTuber his name is David Shapiro um and I have a

QR code at the on the last last slide for the whole slide deck as well but um the I just added a third QR code because just this week he put out another video on this and he talks about this feeling and and he's a really interesting person that talks about Ai and the impact on humanity and all these different things so I highly recommend you check

it out but it's not just about the machines it's not I'm not going to roll over and just say oh yeah okay I'm out of a job I'm going to have Universal basic income and you know just be in Perpetual retirement it's like I don't think so so it's but it's really about how we can go forward taking advantage of this technology and adapt to it so

that's really my strategy and and the the the big motivation for this talk so in this talk there's three parts part one is the future is now ai coding assistance so we'll take a quick look at that and then I go I start to push out a little bit but this is actually on something I'm I've been doing for over years now and um I but I

call it and this is where I'm kind of pushing things I call it micromine services and I'll give you kind of an introduction to that and how it fits in with with everything and then I here's the third part I really go out in limb and and this is where I start talking about um this concept of of some form of test driven prompt programming so part

one the feature is now a coding assistant so I'm a Java developer python JavaScript typescript that type of thing Prim primarily Java though but I've I've been using vs code for a while and and one of the things I really like about it is the extensions in vs code and I just in preparing this talk I just took a quick tour of the extensions in vs code

that are related to AI coding assistance and there's quite a few so you can see there's codium and one thing to look at is the the number of downloads like this has around 600,000 downloads um tab 9 is a bit more popular been around for a while almost 7 million downloads Amazon has one a couple million downloads it's okay but there's new ones coming along all the

time so there's no lack of these these AR programming assistant that are coming out the big one is co-pilot you know GitHub co-pilot I've been using that since it came out um really enjoy it it's you know 14 million downloads and then later um after copilot first came out then they added a chat feature which is also really nice it's all built into the IDE um various

IDs you know like vs code intell things like that it has 8 million downloads uh here's a fork of vs code is cursor and the you know it's the AI first coding editor but it's it's familiar if you're fam with vs code this one's was really interesting came out a few months back be became quickly a bit controversial it's called Devon and they didn't say AI coding

assistant they said AI software engineer and people freaked out I remember when this this first came out and they released a bunch of videos talking about it I don't know how many of you have seen it but um it's the videos were pretty interesting you know things like the AI sets up the project the AI does all you know the AI is at the command line the

ASI set you know setting up projects it's setting up source code it's getting you know doing a lot of work um but since then people have been kind of calling calling them out on it it it really isn't available yet and then there's the the large language models you know and this is where I started when early last year there's Claud which is uh very good and

from anthropic and of course there's Google Gemini and and open AI chpt which is probably the most famous ones all three of these things the all three and all these other tools were kind of in this race the capabilities of for coding in particular but the just the general capabilities of these AI just keep incrementally getting better and better and better at substantially rapid rates uh it's

it's really interesting to watch and to play with so here's an I want to show you an example um so I'm in vs code and this something I did a while back but um I'm in the chat on the side here and I just said um um using uh using the currently open file as a reference create a new class named uh shipping orders View the class

contains two queries one query that selects by customer ID and another query that selects shopping orders from um by a from and to range of ordered at uh and then both queries do paging and then yaa comes back and says sure okay here here's your code so here's the code that it that generated on the left um I go thank you very much because it's like there's

some tedious code in here you know these annotations the sequel maybe it's not exactly right but bam you know it's like okay I got a lot of code so I I don't know how many of you have been using AI coding assistance can I see a a a show of hands okay it looks about 50% um so it it's it's pretty wild um so that's one example

here's another one though and hopefully it'll work I'm in some code and yeah it's going to work so I I just you know I I just hit enter and you if it's you can kind of see this kind of grade out code they called ghost code and it said oh I think you want to write this I go okay yeah that's that's right and it goes oh

and I think you need this block of code all right so if you haven't used these things before it's like oh holy crap you know and then not only is this a block of code but it's a block of code that's in my style of how I want this code to be written it's it's like you know when this happens to me it's like you're in my

head you're reading my mind type of thing it doesn't always do this but it it does do it quite often I'm done right this this block of code is written now I kind of bootstrapped it because I wrote a few of these you know uh it's kind of a pattern uh that's just a quick example of how these uh these AI coding systems work it's it's pretty

pretty significant so that was very quick introduction to Ada coding assistance so let's jump into part two the micromine services what do I mean by this so first what I I want to say is what I don't mean by this because I use the word mine but it's this isn't about arml and I want to point out that the the type of programming that I do I'm

a back primarily a backend developer but it doesn't matter if you're a front- end developer or backend developer the type of programming that we've been traditionally doing and the type of programming that they're doing for the AIS they're they're on a completely different it's like they're it's a dimension how that code works the kind of data that it uses things like that is utterly different than everything

I've been doing for decades as a as a developer and I think for for most of us but what I'm going to try and show here is that we can do some things that start to take advantage of what we're learning about uh how we can build intelligent software so what it this is based on is event driven an event driven type approach but bear with me

because I'm going to link it up to what's happening and and some of the AI stuff because it's fully event driven and and I'll I'll show you what that means right here so this is a design diagram of a it's not a pretty picture this is a design diagram of a real system so I'm a developer Advocate now so I don't have to write production code anymore

but I wrote tons and tons and tons of production codes so when I write demo code I like that demo code to be at least somewhat non-trivial you know at least interesting so this is a a demo application but this is real this and it's you know this is my design I'm just kind of showing it in this fancy way but thing I want to show here

though is that these diagonal boxes represent small focused services like a small Focus microservice all right when it lights up that means it had it got a request and performed some kind of an operation because it's event driven that operation emits events now those events may be interesting of interest to other components Downstream so you can see the the highlighted line here at the top left where

the the highlighted um service on the very left is you know lit up has a lit up line that is um going to another service so that triggers that service to react to it so an event comes out which triggers a command going into another service that service reacts to it and another service and another service and another service so there's this cascading sequence of events and

what I'm showing you is the processing flow in this application I use this as I'm writing the code for this application because it's event driven and in order to kind of keep everything together I I need some kind of visualization of of flow but it's a really interesting and very different way of of coding what's even more interesting is that in this particular case in this diagram

there's only three shapes and they're called components and there's only three types of components so the first that diagonal box I said is like a small Focus service it's called an entity and if you know entity is kind of a clue an entity is like a unit of state right a shopping cart and order an iot device a person whatever you know those types of things but

then the other shape on the bottom is that v-shaped thing that's a view so in a in a vent driven you often hear of U cqs which is command query responsibility segregation it's kind of a fancy acronym but basically what it means is that commands that perform State changing operations happen in the entities the queries the reading happens in the views because event data isn't very queriable

so you need to take event data and transform it into queriable Data but they're SE separated from each other that's the segregation piece and then the third component which is very interesting it's called in this particular tool is called an action but the action is uh its most common use cases it's subscribing to event stream an event stream in in uh most cases so as events come

in it will take those events and transform them into commands that they it sends Downstream but the this this whole system was built just using this pattern of these three components now here's where I kind of pushed things um as I was getting deeper into this and I was you like I said I'm very interested in what's happening with AI and I'm very interested in in neural

networks and I've been very interested in biological neuron networks as you know as just as a a citizen you know I'm not a I'm not in that space but there's tons of material out there on it and I find the neural networks both artificial and biological extremely interesting and it occurred to me it's like wait a minute this coding style here is kind of has this feel

that these entities are are like neur neurons and these actions are like synapses synap what a synapse is is the junction of signals coming out of one neuron going into another neuron so it it has this flavor or this this feel to it that that does this but back to reality a little bit you know this the system just works on Cas processing cascading sequences of events

and that's what what this microm mind is so I want to show you a little movie that I did from real data of this application running from I just took log data and I projected it I never get to watch it on theen scen but uh so what's happening here is the dots represent units of State the lines are events coming out that transformed into commands and

here's what's ha what's happening right now is the system is processing around 200 orders and what the system does is it's not just taking in orders but it's allocating stock to orders the behavior of this application is it's doing this stock allocation and it's not just decrementing counters it's actually tracking physical units of stock it's remembering where every single physical shable unit of stock goes in this

system so the on the right is stock on the left is orders the red represents orders are in a back order State because there's insufficient stock so another behavior of the system is that when it runs low on stock it orders more stock and when the stock comes in it it goes looking for orders that need stock so there's this really interesting Behavior that's happening here and

when everything turns green all the orders have been processed but I wanted to visualize it this way and this is where the whole micromine thing is if you if may have seen some videos there's really interesting videos of like the neural activity in an insect brain that has you have 20 30,000 synap or neurons in it and it looks very much like this but this was generated

from real data of this that D design D diagram I showed you um doing real work and then the code I wrote to do this by the way I use chat gbt to write the code because I the the um the the code I I I use blender to do this blender is an open source tool for 3D videos and visuals and stuff like that has a

very elaborate API I'd never used the API before but I was able to write about 800 lines of python in a couple days to consume the data from my log file and create this movie it was a blast it was blast so the the the concept here with this microm mind is I feel like it's got this biologically inspired event driven process processing flow biologically meaning you

know they there's this kind of neural processing feel to it because with biological and artificial neurons the way they work is a signal comes in and it may or may not trigger the the neuron to send a signal out what's really crazy here is that this mechanical process of neural networks it has this elegant Simplicity but this proud profound cognitive complexity and just these stunningly powerful capabilities

the so think about this is that we showed up humans showed up with this massive biological neuron Network between our ears we have about 86 billion neurons is the number most people say that we have in our in a human brain and there's trillions hundreds of trillions of of connections like a neuron typically is connected to 10,000 other neurons through 10,000 other synapses so there's trillions of

but this neural network this biological neural network showed up into planet Earth and we took the we took over the planet now the same processing model it's not exactly the same but it's fundamentally the same these artificial neuron networks are coming along and they're kicking our butts right they're changing the world on us so the point is that this this kind of neural processing pattern is really

powerful really powerful and using the Rola ly non um AI type of programming style this neural kind of programming style has I've been really shocked by how simple it is to implement it and how powerful it can be with this just relatively simple flow commands come in events come out events get transformed into commands going Downstream triggers another event uh the event triggers a command and so

on that's the pattern here these entities they're just objects like here's a shopping cart as you I'm just showing an adjacent it's just an object right it's kind of a distributed durable State object but nevertheless it's an it's an object so how's this all fit into uh AI well like I said it's modeled after the power of of the processing pattern that um we're learning about much

very rapidly we're get all getting schooled on this very quickly on the power of neurons both biological and and icial but the other thing is that um this system and the systems built in this way is composed of relatively simple components software components that don't get big don't have hundreds and hundreds of lines of code they're typically very focused and very small and AI programming assistants AI

code generation loves this this kind of stuff it it's almost a form of prompt programming in a way because um the the the reason why that that quick demo I showed you where it deduced the the next method to be written it was looking at all the code it was there and it was deducing oh I think that the next thing that needs to be written is

this so be because it was following the patterns of code so these these AI assistants really they can they they pick up on patterns and reusable code patterns very quickly and they they can repeat them over and over again so here's here's some source code you can see this is the definition of an API one of these entities and you can see it's just this is like

you know spring annotations that are here if you're if you're familiar with springs but you can see there's three apis defined this all basically a recurring pattern and this pattern istic approach for writing the code works well but here's where we fit in I think is that okay fine the AIS can write the um the big picture is much harder to do as an example we've been

trying to get that last 10% of self-driving cars for a long time we're not quite there yet you know the cars can drive pretty good but the that last part of full autonomy for the the vehicle is is very very hard so there's this I think I don't know how long this will last it may last a long time it may last a short time but I

think where we fit in real well is we can take very ambiguous this is what we do as developers and Architects and designers of software system is that we take in start with like very ambiguous requirements specifications and we drive down into very specific specifications ultimately into the code itself the code is the ultimate specification um but we have the big picture we can have the the

big context window you know they talk about context windows and AIS and they keep getting bigger but we have a huge Conta Contex window in in our brains and so this is where I think we fit in as the big design um you know the con conceptualization down into those components then let the AI help you components so that's what I'm getting it here is that um

the a can help you write the components but the design is something we do the overall design of the system using patterns though at this level so I want to just very quickly go through it if you heard of the Saga pattern um The Saga pattern is kind of a step-by-step type of a process with contingency plans for what what happens so there's all kinds of different

patterns there I'm going to just go through it quickly but um the you can see that there's 13 of these entities and most of them are implemented using some form of pattern this uh this last one this reduction tree there's like I said there's 13 four of them use this reduction tree pattern it is a tree a data you know data structure of a tree where the

leaves have detailed information and that detailed information get gets reduced down to the trunk of the tree so it has patterns so it's it's kind of wild um here's another design I'm just kind of flipping back and forth completely different application but the same patterns are used in in designing this type of a system and here's kind of a side by side real quick of something that's

a little bit more complicated where this is where stocks getting out at orders or in the other application where um deposits are being consumed by by withdrawals but there's a symmetrical side to it where the the flow can be reversed that that stock can go looking for orders instead of orders looking for stock and deposits can go looking for withdrawals instead of withdrawals looking for um deposits

so same pattern used in both applications so it's you know solid design patterns it's uh and that it has the characteristics of like do one thing do it well loose coupling asynchronous communication State isolation we've been talking about these kind of characteristics for microservices for year for years ever since microservices came out another one though is this is really good for distributed types of systems where these

different operations can happen in different space and time by different space and time meaning on machines something happens here and uh it something else can happen there and they're not locked together and some kind of a synchronous proc processing flow so let me just get into the the test driven prompt programming part part so um right now the general um view of AI programming assistant is they

help us write we prompt them and they help us write our source code and I'm I'm going to just push that it's like okay let's flip it imagine that this source code is prompts the code that gets generated we care all right the the analogy is that uh that first off you you know many of you might be reacting Oh no I got to see the source

code but we've been through before um if you look back in the history of programming we programmed in binary first and then we had introduced the assembl language and the binary program is go oh man no way not going to this Assembly Language stuff you guys are nuts then Assembly Language took over so the Assembly Language settled in before we could catch our breath in that one

Along Comes things like C and Fortran and the assimil program is oh man no way I don't I need to have the control of the machine I'm not going to those new languages so there's always these big changes that occur and there's always this kind of inclination to resist those changes so what I'm I want to push you here say think about it your source code is

your prompts you don't look at the generated code we don't care that's our new the Java or JavaScript or python or whatever it is that's our that's the new Language the the the idea is that what we're building is human and machine readable prompts you know we understand them because it's in our language machines understand it because they understand our language but there's a technique to it

and I this I mean this is just a this isn't something real this is just me kind of hallucinating myself uh but I think the style is results oriented test orent that's why I said test driven prompt programming So you you're writing the prompts with the results in mind what you know what do you you know kind of tell me what this code should produce given a

certain you know different kinds scenarios and it's iteratively developed until the prompt is sufficiently unambiguous but it's not like you're prompting you're writing one prompt for an complete application you're writing prompts for parts of the application right so here's an example um the this is a prompt so what I did was and I've been kind of refining my prompt technique because and I use um all the

different AIS all the time I've got subscriptions to them all and I just keep going back and forth and trying different things but one one thing I do is I I call it warming it up so we're get into a conversation so I had a conversation before this and I I was talking to chat gbt and I was explaining this idea of test driven promp programming we

went back and forth for a while until I kind of got sufficiently warmed up on the topic and then I said all right I said please promp uh please provide an example test driven prompt such as the process of a customer adding an item to an order such as adding an item to a fast food order and it goes all right here you go and the prompt

that came back that it wrote for me based on that very short definition was prompt functionality add items to to order user story so created user story input specification the expected Behavior so it has some expected Behavior now at the very bottom you see system recalculates the total price of the order that one's probably too ambiguous I probably need to provide more detail like okay exactly how

do we calculate the price how do we do taxes but it you know the the point is that I'm not you're not we aren't writing the prompt ourselves we're actually using the AI as a partner in writing these prompts that will develop things so it comes back with constraints and then finally it came back with the test cases it's like okay this is this is pretty cool

it wrote this for me and now I've got a document this is my test my prompt I can go back and refine this with the AI I could tweak it and whatever and then then have it generate code and generate tests and um until the the results are what are expected so if the results aren't what are expected well maybe I don't have sufficiently well- defined and

unambiguous Pro U test specifications or something like that so a lot of people ask well do you trust the code and it's like so my quick response to that is that um say you're a team leader and you're delegating all these you know these responsibilities for writing code to me and to these guys here and you know whatever right so what's the difference between you delegating responsibility

for writing a code and ATT test to other people and expecting them to come out with the good results versus with an AI you know we have processes to make sure that the systems works well and I don't see that that huge of a difference between this so the idea here is that back to this design is that there would be prompts for all these different components

not a big prompt for the whole system but each you know little prompt writes a little bit of code right and so my source code is a bunch of prompts that Define how to write all these different components and how everything's wired together something along those lines the the AI starts to learn these patterns it can start to maybe on a single prompt to generate U multiple

things working together it's producing the source code I don't care I don't look at it anymore so the idea is that that you have systems that if you have systems that are composed of tightly focused Loosely coupled components they can build s uh systems with surprisingly powerful behaviors so you know is test driven prom programming the future I think some form of this is what's coming and

I um the the main thing I want to leave with you is that just dis flip this stop thinking about I mean today use AIS to generate our code tomorrow that's probably not going to be the case that's that's my m main point here so I covered parts uh the future is now the AI coding assistance the point I want to make here I know I've talked

a I've talked to a lot of people about this a lot of companies are the ones that are holding back on adopting this for all kinds of good reasons but you you have to figure this out if you're in a company or in a place where you haven't started to use this stuff and get a feel for how it fits into your style your organization style of

developing software you you've got to do this the the longer you procrastinate on this the further behind you're going to you're going to get this micromine style this isn't uh this is more concrete I've been programming this style for a long time and um it I'm I love it it's it's been it's been a lot of fun and then this test driven promp programming that I'm I'm

just kind of speculating on on the future but I I The more I've been thinking about it the more I've been talking about it the the more I think that this is a direction that we're heading so I want to leave you with this that you I told you about that word vesence which is kind of a bit of a downer type of thing but I wanted

to I tried to come up with an madeup word and I couldn't find one but I adopted this one word called Inova and the re the my definition of Anova in this in this talk is that there's this reaction when the AI does an AI does something for you and go man that's it yeah thank you very much a bunch of work if I had to do

this on myself I would had to power through it and just kind of tedious work and boom it's done and this happens over and over and over again and it's like man okay cool you're reading my mind you know sometimes I say get out of my head but um just real quick AIS will not replace you people using AIS will this is getting a bit old but

I think it's important and then this one I think is really important you have to be able to um adopt your learning style you have you're going to have to give up things that you've worked very hard to learn and open up your mind to learn new things so I think this this uh uh quote from Alvin tofler was a you know a great futurist is is

very interesting so that's that's that's the talk the QR code for the slides is on the top right um the code that I uh showed you is something that's for my company is called kaix that's why I'm wearing this kaix t-shirt there's a QR code for that below um give it a try you can try for free anonymously all that type of stuff uh it but everything

I showed you codewise and and design wise is based on KX uh product and uh let me see if we have questions room six okay okay can you talk more about warming up the AI oh good question so it by warming up the AI it's like um I think it's called One Shot did a lot of people they go to an AI and they they just ask

a question it's just one one question and they get the response back and um the the idea is that no get it get it you and it into the conversation like and another so another approach is people say well you're an expert in blah blah blah you know and expecting it to be all right you know the all seeing expert in a particular area I think it's

more than that sometimes you know that like if I'm um especially like if I'm writing a document or something or uh or generating an image um I like I say I'll I'll write out some description of what I want to do and maybe and the really cool thing is when you say do you have any questions and a lot of times the ALS come back and say

oh yeah this is really fascinating and I do have some questions here they are and you ask answer the questions so you can get into really complex discussions with the AI and get some really interesting dialogue going with the AI when you have this back and forth kind of conversation so that's what I mean by warming it up talk to it ask you you know tell ask

it questions and ask it to ask you questions that thing uh the I coding support for at the fingertips isn't that the risk of engineering higher um I if I take it where it's like well we're worried about the code that it writes for um then yeah you know it's like it's not the fully optimized code but we like I said we went through this before when

especially like when we went from a L language to C and Fortran those types of things that was a big resistance because the Assembly Language programmers felt like they had their their fingers on the metal right they knew they were talk you know they were controlling registers and they could optimize the actual you know flow of of instructions at the Machine level and then all of a

sudden that was gone but then we've you know we don't worry about it you know the computers got faster for one um so I think it's the same thing that we're we're this will have to mature over time we're just have to level trust that the code that gets written but you know it's like say you do develop some code you put it into production and you

you know you end up with the inevitable hotpots the performance hotpots the low hanging fruit where can I where is it running slow where is it spending a lot of time go back to that and fix it but now the process would be tweak your prompts and go to the AI and say Hey you know we we have a problem here with the performance let's let's fix

it um if you're familiar with recent situation uh res resignations due to security concerns yeah that one we're going to this a lot I think so open AI is under the spotlight right now because it the question is there was a lot of people that left open AI last week um especially the security team and because there's concern about the what's AI going to do you know

how can uh aibu yesterday there was a talk with on security with AI so the concern is okay now people have you know malicious people have these AIS that can do nasty intelligent things to try and break into a system but I think we can counter that with um we have AIS that are also fighting back right very smart very adaptive those types of things but this

whole yeah this security thing we're just we're watching it play out real time and um there's a lot of soul searching going on though as well with the companies about this I think they're cutely aware of uh would you recommend and I can't read it find oh thank you find over chat um it every week it changes you know there the you know like for a while

I was I stopped using chat gbt because I Gemini was really picking up the slack before that I really like uh Claude but now as of like last week when gp24 came out and I start I play with again it's like wow okay this is much better so you really this is where I um it it it's an advantage to have access to multiple AIS but they're

they're definitely going to be uh one-upping each other all the time so so one day somebody one is ahead and the next day they're behind that that's just the way things are happening right now um let's see if we stop looking in the code and as the time goes by the knowledge of the system also goes away uh who do you think will fix the system okay

good question the knowledge of the system at the level of the code like Java JavaScript or something like that yeah our you know you can imagine our skill sets will go away same analogy um we none of us have any skill set with Assembly Language program right and we fix it with our higher level of of abstraction source code think of the this prompt it's just another

level of abstraction we will adapt to um fixing systems that don't behave well At a next level higher level of of abstraction we keep going up these abstraction levels over and over over I mean think about computer systems they're just massive layer you know look at a stack trace this massive layer abstraction after abstraction after abstraction so yeah I think of your the initial reaction is it's

a legitimate reaction it's like oh man I'm going to forget how to write code it's like yeah you are but but your new skill is going to be how to to articulate unambiguous prompts Drive the behavior of systems in very precise ways and there's going to be techniques and styles and expertise at that and we'll be kind of up another level then we'll go up another level

up beyond that so I think we have one more time for one more how how safe is AI addends in terms of code leakage so I I by code leakage I I guess you you mean that it's U leaking out to other people it's like if you want to look at my code even though it's proprietary have a ball it's my feeling it's like um you know

how hard it is to understand your own code when you go back into it months later right so we're expecting people to steal our you know our code I I think it's more involved in that um so it that what that the the company knowledge the company jewels are at a higher level as well it's it's the bigger functionality not the like the the lowlevel functionality I

think although that is a legitimate concern so one thing that can be done is you can have your own private AIS on premise nobody gets to it but people in your company and nobody can steal your intellectual uh property so we're out of time thank you very much for uh attending the talk I really appreciate it uh