0:00 Alright welcome back to a little bit of a special episode today on Energy bites We've Got Bella Colada Khaleel Good Yeah good nailed it the events manager over at midnight marketing with US Day Thanks
0:15 Thank you Guys for for letting me come in and yap now I think It'll be really fun not every day we have you down from montreal so and we're Gonna we've got my Trusty co -host Bobby neal here was good
0:26 How's it going made the trek up from Southwest Houston
0:32 are like a lot of cows out this morning I didn't see any over here by the office Traffic Yeah
0:38 not nearly as many over here by the the DW office for sure but now today's came about because you made a post you're getting your is is it a class or a certification it's a specialization course right
0:52 now that I'm taking online I'm on petroleum engineering with AI applications I have yet to get into the AI Applications portion of IT I am about like a month and a half in and I've just been trying to
1:04 get as much information on it as possible without having a background in oil and gas at all I'M just doing my due diligence to understand a lot more which is what you'd knock out two birds one stone
1:14 oak yes
1:16 what AM I going to cause when I don't want to spoiler alert about some of the questions but it's like you Gotta know the the subject matter before you could even get into like a data science to know
1:26 for sure and so she made a post about it and I was like Hey I know a lot of people have a lot of questions around this so why not just jump on and do an episode about it so it was all about learning
1:36 anyways right we to take some stuff away so I think with her grilling us with the questions Yeah Yeah we're flipping the table a little bit today bell has got a very long list of questions that we're
1:45 going to roll through and hopefully Bobby and I can sufficiently answer those for you and everyone else listening to just to to learn a little bit more about both petroleum and AI and how they kind of
1:57 interact and overlap there so I am very excited I mean I I grabbed a lot of the questions from the collide post that I originally made posted about it twice on Linkedin posted about it on slack asking
2:10 everyone at work if they have any questions and also how to Harass a few people to ask me questions as well because I know for a fact no one feels comfortable just like typing it out in the comments
2:20 and obviously I won't name names I did tell them I would I would keep this anonymous but I had someone who is in there like in their in their forties and message me said Listen I've been following you
2:32 for a little bit and have never really reached out before but I'm nervous to actually ask the questions in the comments because I've been here for so long but there are just some aspects that I don't
2:42 know about the I want I want to learn more about but I feel like because of who I am the industry if I were to do that people would look at me differently and I'm like Listen I'm still somewhat new in
2:53 in the industry so put it on Me I'll ask all the embarrassing questions and to you guys well that's fair too but I mean I've said a few times but when we started this podcast I was more of the
3:03 Technical guy and John was more of the color commentator in a way but mean especially with a with the eyes of it's really kind of been flipped almost cause he's been like deep in the weeds on the AI
3:13 in the L M Stuff and I've been more like high level understanding what's going on haven't you know until recently probably devon is nearly as deep as as they have so so in Layman terms what would you
3:24 say your your expertise is right now based off of the questions that were going to go over Yeah so I would say the way we kind of split this is one AI and ML LLM incorporate AI and ML But AI in EM all
3:39 by themselves also are stainless things and so now there's a distinct kind of difference in some of these things machine learning traditionally came about specifically for numerical data whereas
3:50 language models are literally language models so thereon language they're not on are built for numbers other generally not very good at doing much with numbers in most cases it's passing off any math
4:03 to a python script that the language model writes itself and then does the math in Python and then brings data back but somewhat lakes kind of like what you were showing me earlier as well exactly so
4:13 Yeah my My kind of role here at a client is on the data side I lovingly call myself a data plumber but that's that's that's really where a lot of my background has been over the last I dunno God seven
4:27 years now five seven years something like that but it all came about because it was it's a means to an end for me of getting answers right away as an engineer i just want to solve problems and so data
4:37 is a really good way to do that and then I feel like bobby you come at it more from you're definitely a practitioner on the data side but you're coming at it from more of the traditional Yeah I mean i
4:48 mean you think there's only then there's like descriptive analytics predictive analytics which would be more like forecasting types of prescriptive where you're making recommendations and then Nothing
4:58 now you're getting into like More cognitive analytics almost right like is where it is and I I've live probably more in the first two yea side of it which will be more like just data movement and like
5:10 now reporting B I reporting but then also getting into the data science side of it and I've done a little bit say with tensorflow and some of the deep learning I was DIP my toes but even still with
5:21 everything going on like that that foundation you have to build you have to have good data to get into it and you know I hear all these buzzwords and then I get into an operator that you know doing
5:30 something like consulting now it's like dude we just want to see this chart updated every day event that does that then I'm good and like okay but like I don't need you to tell me all the time just
5:40 does it do it do what I want it to do I honestly think that's like a very valid takeaway here for anybody and it's like just because there's there's going to be twenty different ways to do a lot of
5:52 things but understand what the actual problem is that you're trying to solve yeah not all the features that Someone wants to have or the way that people don't care how the sausage is made if it solves
6:03 their problem right if it tastes good and so that's ultimately what this boils down to is it's like do you have to use AI for everything no you shouldn't in all honesty right like those are very
6:12 they're tools in a toolbox right like you wouldn't use a hammer to screw a neon or a screw into a wall like doesn't make sense and so that's how I kind of hit it really hard and it may be but that's
6:23 kind of like right i guess you could but it's going to be harder and probably not as clean and slower than the pros and cons of where like black boxers is more explainable when all that kind of stuff
6:34 but Yeah Man I think the year before I get into like you're not as far behind as you think you are yet my Gosh you you Listening You're not as far as you hear all these things and you feel like you're
6:44 drowning you're not as far as behind as you think if one is constantly changing to wreck that's the hardest part of the fun part too is like here's A is a solution and here's a as an amplifier like
6:55 and Yeah how does it just make me more productive is it I think a big thing big thing right now before we get into it let's get who are you real quick God how how do we how did we end up here real
7:07 fast how did we end up here so I Gosh I started at midnight maybe like two and a half years ago and I know that midnight and what was DW at the time nike now collide we were always riding in the same
7:20 circles obviously tim knows and loves you guys a whole bunch and then said Listen Bella if You're You're new to the industry cause I've never worked in oil and gas before have never even had a
7:31 background in it at all let alone studied it he's like if you're going to learn from anybody sends me the Youtube paid for to vertical wall cavities like wash these listen to their podcasts and just
7:40 go for it so energy Wanna One and before it took a Hiatus and now that it's back that one was just constantly playing in the background all the podcasts or the gas startups like that was that was My
7:52 main hub for learning and then as I got further and further into the industry people would obviously I'M I went into business development I'm getting into sales and people are like so explain to me
8:02 how you would create a campaign around an x y Z product and I'm like Yeah what Pardon I don't understand any of those words I don't think any of that was English and so obviously because there was
8:13 this quote unquote language gap I wanted to I wanted to close that barrier and close that gap and started taking a few courses so UH maybe about a year and a half ago I took courses on like
8:23 introduction to oil and gas just have just a better understand it posted a really simple like picture of my thumb in front of the certificate when it was done and I think that's still my best
8:32 performing posts to this day and I was super exactly I was like I wanted to make more sense and people were very excited to know that I'm making the effort to learn and and so I wanted to take it one
8:45 step further and go into petroleum engineering with AI applications and again I have only just recently started this course so I have my notes in front of me like everything that I've been learning
8:55 from like oil and oil and gas fuel life cycle all the way down to like offshore analytics like seismic analytics and whatnot and and I figured I may as well just continue to do my due diligence to the
9:07 industry continue to learn not only to help us I like to to develop business better and to sell better but also to understand the industry and the way that it's changing like and to your point earlier
9:19 John like AI is changing everything constantly and nothing will what we're learning right now will not be the same as it is in the next six months but at the very least we can start with the basics
9:30 and the understandings of it which is why I have about all of the big questions a small questions the dumb questions and but again I'm happy to be the one to ask these so at the very least it's just
9:42 like surge of of knowledge that we pass off to everyone moving there to those like nothing people from males I don't think of it like oil and gas a super are inclusive community and as long as you are
9:54 making the effort to learn and be part of it he said im from montreal very almost Anti oil and gas charite and then you say before we Europe Theater or so I was and I graduated in cinema and
10:09 communications in college and then I didn't go to university i just went straight into sales from there as a manager after that I was helping out building out this company that was doing door to door
10:19 sales and then obviously to better understand what it is that I'm selling for recruiting for I went on the field and so I started doing door to door sales so I'm almost implementing that so though I'm
10:30 not necessarily running this company I aM part of an integral department business development and the only way for me to learn is to be like be in everyone else's shoes without physically being able
10:43 to go off onto like rick sites and like get my hands Dirty I know that's to Bobby's point that's the industry appreciates effort right like that and that's one of my favorite parts about it you don't
10:55 have to have a PHD to to go work your way up into you know important role in the industry people will help you if you show that you care and you're trying one hundred percent it's pretty amazing I was
11:06 a high school math teacher and a coach and had no really yeah nobody was getting a job thanks John had just
11:14 kind of go phillips but you get in and you just dive in and rework work hard and you know try to learn as much as you can and people will lift you up I think the big thing too is I had a career at one
11:24 point in time on the sales side as well and it's very hard to sell something you don't understand and so like I think that has bitten a lot of service and just companies in general quite a bit where
11:36 it's like you know they hire a marketing department or they have one internally or they hire somebody but they don't know anything about the industry so that they're just tried to use kind of generic
11:45 sales techniques that don't necessarily work if you don't know what you're selling and so your your approach here I think is spot on you I mean as with most most things you can learn as much as you
11:56 Want A book but until you go out and do it you don't really know and you know and so I Applaud your efforts on on trying to do that and educate yourself both you know academically but also by you know
12:09 the real world experience for those of you listening she'll be ready to come in as a rather engineer six weeks give me give me five and a half just wrap my head around it and then half a week just
12:21 like fully prepared myself okay like I I want to I want to get myself the full harder actually I have a hard hat at home shiny white one not shiny good it's better that way even even better has a
12:33 single sticker on there by the end of my by the end of my Journey I will have not a single patch of white left on there perfect is going to stick them all on there that I'll be ready every safety
12:43 person's nightmare Yeah sorry Guys used to be such an integral part of the field and now it's like frowned upon because it could be potentially hunting a crack in the heart and so it's like The more
12:56 you learn Okay then I'll I'll keep it white stickers are still like oil field bartering currency for sure so they're very valuable in the in the field side of things I love that that's perfect I Kick
13:10 Us off Let's Get You Guys ready to be drilled was doing alright let's start off early now alright it's coming to a tee shirt or you soon I still have to talk to my team about that I guess by the time
13:25 this comes out they'll already be aware and the T -shirt will be in production Hope Y'all already available on the midnight marketing website it comes to your merch I'm alright guys so let's you Real
13:35 Quick let's start as Dumb as we possibly can what do AI ML and ML LLM stand for I'm just glad he didn't ask what GB T stand for thanks to come which I'll never forget so what you want to take Yeah I
13:49 mean AI is Artificial intelligence AI ML is machine Learning and then L AMs as large language models so I mean I think the lines get blurred really quickly yes as to what people are calling Me AI has
14:02 been a concept for sixty seventy forties fifties Yeah it's like basic science fiction alright robots can do things and talk and eat whatever I mean aren't our first guess ever to laugh sorry mom he
14:15 was working on an AI computer vision thing at U of H in the nineteen nineties and I think this is one of the things we've known what we want to do but we haven't had the compute power capacity and and
14:26 now we have the entire internet and somehow we can consume that all into massive supercomputers and do all this cool shit so Yeah over the last like year have come to this realization that like just a
14:37 convergence of technology right so it's like you can't have a machine learning model unless you have a bunch of data to train the model on and so then it's like and you need a lot of compute to train
14:47 the model on the flip side of it and so over time we've been getting more and more data and finally we're getting better and better data and then on the other side you've got the convergence of
14:55 cheaper and cheaper hardware store better hardware way more storage and so it's like okay well now we're at the point where we have so much data that humans literally can't promptly hinde it and so
15:06 how do we fix that while we let a we train a computer to do it and that's kind of the gist of a lot of this stuff is just it's it's like an evolutionary process in a sense at least on the technology
15:18 side of like okay well we initially had a bunch of numerical data that we couldn't make sense of and now we built machine learning models to help do that and now we have a bunch of written data that
15:27 we can't make sense of and that's where language models kind of come in but a lot of them have like intertwined pieces or components that one uses and another doesn't and all these things so it it's
15:38 weird cause a lot of people use them very interchangeably but they do sitting in a while as data science machine learning that there's a big data and via big data but like mean I think Where I would
15:49 may draw the line is like machine learning you've got more of your clustering algorithms you've got regression Yeah and somebody's like even random forests like tree type models and those just
16:01 constantly iterating and you know learning you know as you add more and more data and then you can kind of redo the model then push it out like do those kind of things finale AI five I think what
16:13 people are more ubiquitous e L EMS are kind of more what people are talking about there but that's where it's been he can speak speak more to what's going on there relative to the Yeah there's a lot
16:22 going on on the AI side or the language model side and as everyone's aware there's basically a new model it seems like every week and they all keep adding new things and but Yeah it's it's like I said
16:35 it's this evolutionary thing of like okay what we Have I mean within a single oil and gas company there's probably millions of documents right and so it's like how the hell are you was defined the
16:46 information that you need when you have that volume of of data meant by this clause written well yeah and it's or even like and that's the part of the problem is it's like I know this document exists
16:57 I know it's in this contract I don't know where the hell the contract is or there's twenty versions of the contract which is the one that I'm supposed to be using and so like the amount of time I've
17:08 heard everything as far as thirty percent on the low end up to seventy percent on the high end of engineers time in the energy business is spent looking for data and so it's like you're paying people
17:19 very large salaries to describe all this and half of their time is just look they just need the information to now are to do their job right right they're not doing their job they're looking for the
17:29 data so that they could do their job and so the the promise of this is to allow people to do their jobs way more efficiently way more effectively and like the long term goal in my mind of language
17:40 models at least from a enterprise corporate perspective especially within the energy business is historically as you know oil prices rise and activity increases you need more physical bodies you need
17:53 more people whether in but that's in the field or in the back office or in engineering whatever the promise with language models is this ideally it will allow people to scale their company without
18:03 having to scale the amount of people as quickly or you as the engineer are empowered to do more because you're not spending half of your time looking for stuff you have the stuff it literally at the
18:15 the click of a button and then you can go to your job and so that's kind of a very simple way of putting it but
18:23 Yeah it's a lot I also think the really interesting part with language models is it kind of allows US convergence of data from all different sources there's a bunch of things that like MC PS that
18:37 allow you to pull data from a bunch of different places that might not be in your internal dataset like in a poi i do but without having to go and you know dump all of that into a database and
18:49 structure it and understand all of that you can just set up an MCP server and it will do it for you essentially and so now you can have okay well I've got my internal dataset but then I can
18:59 contextualize it further with public data that I pull in from like I was showing earlier and and so now you you have this marriage of historically speaking engineers typically either have the raw data
19:11 or they have the report they don't have both and that's because raw data typically lives in a database and the reports live in a PDF which is we can go into how terrible the format that is but the the
19:23 promise of language models in my experience that I've seen is just it's really being able to unify all of that into one place so you don't have to go it's like the multiple tabs problem you're at tab
19:32 switching you're not going to different though I've got to get the data from this app and then I've Gotta go pull it from this database over here and then I've Gotta find the PDF in another system
19:40 that correlate to those it's a nightmare one more point on this question then I know got a lot to get to but is I would say what all EMs are AI but AI isn't necessarily limbs I mean there's a whole
19:54 other side of this Yeah Yeah achievement makes it fun it's it's much deeper than we think and people just having albums have done such a great job that people think this is the only type of AI or but
20:06 like there's Managua s now we have multi model multi -modal multimodal models like so now yet computer vision adult stuff too but you have computer vision you have text to speech or speech text you
20:18 have all kinds of different flavors of it so to speak and I get but it has a robotics center thinking about the robotics and all that kind of stuff it's just that now that like like I said it or not
20:27 all powered by LLM is around but it is AI in a sense yeah but it's just like marketing right like at the surface people are like oh marketing right like you you do social media and it's like a cotton
20:38 make posts know there is so much more to marketing or development right it's like Oh you write code it's like okay well there's a
20:46 production isin code managing servers DUI front end like it's a lot more to it than just that and actually interfacing with people turning their idea into Saturday's and the way that you have
20:57 obviously you have different platforms that you're using in the marketing world to kind of like dive down into like E into the science of it what platforms would you suggest starting on if you're
21:07 getting into real like real time experience because obviously once you've once you've learned all of those once you've learned about AI and ML Then LLM obviously want to start implementing it before
21:18 you can go into the hole before we before after sorry that was in English before we get into the whole petroleum engineering part what would what platforms would you propose like clicking around on
21:29 and learning on before you can really implemented properly you have some from the I mean I'm Gonna go out and say that Python is probably the best starting place just because it's everything has a a P
21:41 I basically and most most of the ML and language model stuff is pretty much yeah the main river became like our was a contender there for awhile and I love our M M shriek and still do a good bit of
21:53 this with their printer they have pipes tons of tour but Yummy Python have all the different data science models machine learning yet tensorflow I think you put in there and those are all things
22:04 leading up to the L M's now Open AI and Claude all of them have an interface through Python much all directly supported at least through now you get really meta because now you can write your code
22:16 using Claude right but you know maybe don't know python but you can write Python using claude and now have that access is private open AI like there's all these things you could do are there any free
22:26 vs paid models that you're using right now that you would prefer other over others and I would say from the free side Geminis Pretty Damn Good I mean they all have free tiers rate but how good is it
22:41 actually get Yeah I mean some of how often he can use it to open AI how often or whatever chanting Anti because I was using it and I ran up on the daily Limit and I'm like Life Pretty nice Gotta run
22:52 anyways and twenty Bucks a month or whatever it was it was like and that's the thing I'm going to get value out of is it's like if you're using these on a even a few times a week basis twenty bucks a
23:02 month is literally nothing but think about like the I mean I only know you I've used to help write some linkedin posts I'm sure like even just to get for I brainstorming stuff like that but this is
23:12 iteration that you wouldn't be able to get maybe without another person you know so again now we start getting into the how much augments just know for sure and and you know if you really want to do
23:21 it for free there's a library called Ulama that allows you to install open source models directly on your laptop and that is again of course very contingent on the Specs of your laptop but I mean even
23:35 like most Mac books I think are pretty pretty good at running these and they have Want ties versions of them which is a smaller lighter weight ones that you can run on a lighter horsepower hardware
23:47 and stuff but I would say those are probably the two easiest to day open routers another really interesting one and it's literally a router so you can go in there and just pick whatever model you want
23:58 and if you have a pa keys you can pose them in there but they have a bunch of free tiers and stuff and they host all kinds of open source models and free kind of preview models and stuff like that but
24:07 they let you use for free so it's kind of cool because you put in one prompt and then however many models you picked they all respond at the same time so you can compare answers to okay it's pretty
24:17 pretty neat but even I mean we didn't bring up say cursor figure and look on the outside of the free development environment you can just download and was built off of vs code that's what you are
24:26 looking at swimming again now you have access to that now you can write your python in there but you can have it help you write your Python code you can do your prompting to the side and it basically
24:34 build everything for you Yeah on the coding side I would say cursor in Windsor for the two big contenders right now if they're just integrated their IDS with LLM integrated into them and so like I was
24:47 shown you earlier like it makes it so easy to iterate on a script or to have it write a script for you and just implement it right away as in for me especially if I was a more private operator I know
24:57 the public's have a lot more some data security and privacy things that they have to adhere to but like why would not give my best kind of like code first kind of determined nears like cursors I go
25:08 while yep like while that and I mean even their first paid tier includes private like they're all private and so it's yeah it that I saw a post the other day and it was like if you give a junior Dev
25:21 cursor it's going to create junior def code but if you get a senior Dev cursor it's going to create really really good code and so like I was telling you it's like A if you just understand the basics
25:32 of Python and how to code and you know the logic and how to do loops and if you need a P eyes to get the data all of that stuff it you can get very very far especially with just like scripting right
25:47 like not production eyes code but like I need to go pull data from here and do this to riot which is a lot of what a lot of people in our industry do or need to do and so it's a yeah it it makes it so
26:00 fast so it looks like I'm taking out all of my notes from all my coding classes in college then there's another thing I want to learn is going to is probably going to be tough for Me I will but
26:11 learning it by like I learn by doing right and but one of the big things that I will say is people don't use language models enough for helping understand say you can explain this code to me right
26:22 have it write the code for you then take the code dump it in your favorite language model and say explain what this code is doing to me step by step and it will do a very good job of that in ninety
26:31 nine percent of the cases and so like that's a really good way to learn it right because for me like I've always I always been in the dataset of things I've always wanted to learn python but I can't
26:42 learn something unless I'm doing it right like you teaching me about classes doesn't mean shit to me unless I go do something with that you're physically implementing it and then seeing know why I'm
26:53 in I need to know otherwise why AM I using this exact code what is it doing and then what's the end result absolutely leaving like I for one of my clients I wrote a SQL query to pull himself down and
27:03 I sent them like the the the output of it but I also sent them the code it was like well they're not going to say and the code was like plopped it into claude was like I'd explain this to an
27:14 accountant whose primary XL first so he went through in his like will this join is kind of like a you look up or x look at me and Blah Blah and I get busy put it in their terms and they will share
27:23 that across a they can understood what it was doing at every step in their own with their brain works he also says no a question on that actually cause I keep hearing people mention Claude and this is
27:32 probably one of the only platforms I haven't heard on heard of yet what what exactly is it and how does it differ I like the Chachi PT at this point it's another foundational model it's just different
27:43 different than what the company okay their methodologies for how they train it and what they're doing it's a competing product to chat JPG or Gemini it's nice at least to this point is between it and
27:56 gemini really for the best Coating Yeah so chronically I think the Debate Yeah for sure and that changes to the New Gemini six months ago but it wasn't that way that is and so it's yeah then there's a
28:08 bunch of open source coding specific models to that it's like I Haven't even tested those but they could be better who the hell knows so for someone for someone like myself that is just getting into
28:19 petroleum engineering with AI applications where would I start first would it be more like an LLM time series models something else would you start doing what you're doing start with the actual term
28:30 engineering because there may be an answer that's already been fleshed out as well and that doesn't even require a like a torpoint doesn't mean sorry that well the you we were talking about how quick
28:42 things change but I think the thing that doesn't change is the foundation right like if you don't understand what a well bore is or you know where the data is coming from or why the data looks
28:52 different and it's like Oh well that data came from Halliburton and this data came from Summer Day so it's going to look different even though it's the same shit those like there's a lot of nuance to
29:01 that and across the industry and so understanding like how things work and why things are the way they are before you get into it is in my opinion a much better like I can I can teach we can you can
29:17 can teach anyone to code it's much harder to have the SMB piece of the industry than it is to put it and you don't even have to teach them code now you just put in the language model i guess to that
29:29 ammo if we're going to assume you have a base knowledge and patrol engineering yep and or geology or whatever
29:39 I mean again I think he innocent need to understand the fundamentals that lead up to that also like even just linear regression and basic stuff basic stats and stuff like that because that's what's
29:48 going onto the hood in these things and I can see both side because someone could actually not know any of those jumper I told LM get a good answer and like if they waited to learn those things to get
29:60 to that answer it's a it's a there's a balance but if you really want to be an expert I think you need to build that foundation so speaking of foundation before you start laying off layering on the
30:09 machine learning what are those principles and what does that look like Yeah I think I mean I would say it's basic data science stuff Yeah that sounds even like data modeling because like you've got
30:18 to put the data in a format so that you can put in there in minutes then then the proverbial eighty twenty like Eighty percent of your time is spent you know getting the data ready with actual data
30:28 science in the modeling is not I mean literary what what you just what we were walking through before we got in here that is literally the eighty per cent of what makes We're doing a clyde different
30:40 it's not just dumping PDFS into a model we're doing a lot a lot more around all of the data's side of what goes into the model and then like the model piece it's like the models we just pick which
30:53 model we're going to use and it's there we don't have to do any maintenance or things on it and Yeah some of my leaving in outliers kicking them out you know but then again it didn't that to that
31:03 anytime I like make time series himself I mean there's a ton of time series data so I think having an understanding of just time series data in General and I think some of the time series models I
31:13 think but it say it say in oil and gas like a lot of these things that you'll pull off the shelf or more seasonal models in the oil and gas have rarely has a seasonal component like at least in the
31:24 this more science side of me like shirt sales and stuff there's season and seasonality as baked in but may be already have like you should understand the AARP's equation because Yeah but you could use
31:34 AI or some other model I'll Look I have for gas and oil Yeah and it's not nearly as good as what people were doing and ARPs for years I think the data foundation is really a good point here because
31:46 with even even if you're a front end developer you're getting data from somewhere and you're displaying data like data is the root of most most things that you're going to be doing on especially now
31:59 that eyesight or anything but even like say you know I think it just dropped today but we were talking to Jeff criminal a week or two ago but for him it's like I we've got four different sources of
32:08 data and this one calls revenue this and this one calls it that so if you don't even understand like that at the eyes not going to understand how that contexts unless you give it that context you have
32:19 to clarify it's like okay these these synonyms need they need to mean the same thing we need to make sure that we are calling at the same otherwise they just start popping up with different names
32:28 they're like wait so is this anonymous with no Yes maybe no I mean we're we're extracting PERF tables from a poster of reports at the moment and Jimmy one of My ER New Devs was like wait a minute on
32:41 this wine you know it's got the clusters broking out but there's no shots per foot there's just a little side table that says the number of shots per foot and the spacing and the phasing and stuff and
32:50 then in the Halliburton table it's got every single detail written out and all this stuff and I'm like Yeah this is exactly the problem that's the I think I've I know have mentioned a million times on
32:60 the show but like that's another thing coming in that a lot of people don't realize is that the operators based don't generate a lot of their own data until the well goes into production and so
33:08 because of that all of the data that they're getting could be from is and it's not could be is from dozens of different different vendors in different formats with different terminology but it all
33:19 ultimately means the same thing to the end user but if you're doing stuff with that data you as the SMB have to understand that you know well head pressure and treatment pressure are the same thing on
33:31 you know but they're calling them two different things because it's two different companies and so there is that the data Foundation PS I really do think is a lot of will go a long way in helping a
33:42 lot of this stuff but even like with your background you know like you don't even realize how much data stuff you're doing on the marketing side because that's all I mean you've got the front end of
33:51 marketing which is the the events and the advertising and all of that and the person all of that stuff but then there's
34:00 where it's like why are you doing these things it's because you want to get data so you can optimize for the best result which is data driven and so it's like if Alien is all I don't want to
34:10 overcapitalized one hundred percent you know asio all those kind of things like I'm in the middle right now of kind of comparing the work that I'm doing on Linkedin with the work that I'm attempting
34:19 at doing with these courses right now and it's like I've I've been able to fine tune the content that I'm posting knowing what date sorry what days post it went to post it what to post who to target
34:32 it to We are actually talking about this over coffee very briefly but like what to and what not to do and when supposedly coding Yeah on Linkedin if you wanted to perform it's like you don't add links
34:43 you don't add Emojis and and you don't copy paste from Chachi BT directly because Linkedin knows and what were you mentioning earlier protip one there was a there was like a security thing that came
34:57 out about judging beauty even on the private versions when you click the copy button that triggers an event on their back end that stores whatever you are copying because they're using that to
35:09 reinforce this was good cause they want to write like if you're copying it that that basically says that your intent is that this is good but am taking it so you're taking it somewhere else to do
35:19 something with it therefore that was a good response system that back to the model to keep the model training but the other thing that a lot of people don't know is within that text there is like
35:28 hidden unicode characters that basically allow other platforms now to recognize whether this was human generated or not and I I know there's going to be a lot of work around that because especially
35:41 with the video stuff like the image and video stuff now it's it's kind of terrifying at all outlets of that but those are Yeah it's it's it's there's there's even websites now that you can just paste
35:55 the copy into and it will remove those texts the hidden characters from it and then you can go use it somewhere but you know it's it's always going to be this back and forth of you can never you
36:06 you'll never at some point you'll never be able to tell if it isn't originally written post or if it's all AI and half the time people can figure out if it is AI because people use EM dashes and I and
36:17 I spoke about this the last time that I was here and I was like urging people to stop using Em dashes they don't understand the concept of it because sometimes he post it and it just looks like you
36:28 got that hold copy the whole tax off a chat GB t and so now when I see EM Dashes and peoples and peoples post my Oh Yeah sure Geez and I thought the venue was either on Twitter or Linkedin but like I
36:41 think colin put something about that too but I was legit starting to use them every email sparingly like and and work back for even five years ago now I was like I won't touch it like forget the EM
36:50 dashes a regular dash
36:53 or we're not
36:56 exactly an area it's actually unlikely that each model type your sentence he will stop get Outta here with your grammar Bobby now but each model also tends to use like there's a shortlist of different
37:08 words that the Buzzword pretty much always include in them too which is another interesting kind of tell of these things but that that already that's a whole other topic but Lincoln Buzzwords and
37:18 chats literally tell Chad TBT Don't use EM dashes and the Response I remove them all and it still leaves him and like I can't help myself yet again these things are not sentient yet so they don't
37:29 understand The the reasoning and thinking part of them is not they say that they do it there was actually a paper or article that came out I think this week or last week and somebody went through and
37:40 tested like they qualified the different layers of reasoning or whatever but they were basically said outside of just like light reasoning once it gets past a certain level it just completely
37:51 collapses on itself as far as the reasoning side of it goes and but that's important right like if you don't understand the intent of what the user is asking and then how to go take that intent and
38:01 break it down into steps of things it should do then yeah it doesn't and but that's the whole point is these are language these are models and the idea is that the model was trained to do a certain
38:12 thing it doesn't think it is instructed to think no one really knows what that means but you know they don't have comprehension is probably the easiest way to say that in my experience at least the
38:26 factory engine guess not yet is the Keyword we're almost there they say the asker but That's I mean I dunno I'm very on the fence about that because everyone is saying it's getting closer and closer I
38:36 tend to believe that it's probably not as close as they think it is but then I'm curious on do you think that we're getting the best version of models or is there stuff going on behind the scenes like
38:47 the CIA or like thousand government as well as what should all things technology Yeah it it is ten to twenty I guarantee you language models derived from the fact that we are trying to process as a
38:59 government on the federal side we're trying to monitor people and were trying to process millions and millions of records of emails texts like all the the written shit that people have and so Yeah I
39:12 Guarantee you this has been operating and government agencies for probably a decade now In the digital footprint is so real and is so scary Yeah it's terrifying I mean i fully know that I have so many
39:25 videos out there where people can probably replicate my voice oh Yeah that's but we're also getting to a point now where AI AI made videos you can tell that it's been A I made like I Dunno why but on
39:36 Tiktok the other Day I saw this like talking Gorilla actually hilarious don't know why and it looks very real but you can tell that it's you can tell that it's not so again to your point it's like
39:46 they're not sentient yet or scarily UI or image video stuff it's just like video games were like me playing Golden Eye on Nintendo sixty four now looks comical how terrible it looks but whenever
40:00 whenever the sixty four came out that was mine bought like we didn't have three D everything was flat and so it's the same thing they're like now they're pretty Damn good they have these weird things
40:11 that they do with like fingers and appendages that you know it just
40:15 it just keeps getting better right and so like at some point it will be an unrecognizable which is terrifying a
40:25 match but Yeah it's Yeah That's a whole other can of worms there one a week or two month ago it was like I think All AI generated but it was like within a conference that looked like it was like a
40:39 boat show or something like that and people were talking oh yea they they're now they're now doing really good videos of like fake interviews like one on one interviews of people though like like you
40:49 Guys do where you like walk up with the mike it's completely fake from a conference or whatever it's like well this is terrifying phenomenon I feel so much better about the future of my work now that
41:01 just means you get to you get to generate one hundred different versions of that and throw it on social and then look at the stats and see which one did better and just hope for the best Damn Good I
41:10 think at some level that that to your point I think people are not always the meal on a case by case basis but like I think original content it's not like people will it will shine through and people
41:20 will desire that more I complete always at all as there's not a single time where actually no that's a lie I was going to say there's not a single time where I've posted and or if posted content with
41:32 ad spend behind it where it hasn't done well obviously it's meant to perform better because you're forcing it in front of other people but still I'd say like the top ten of what I post right now is
41:42 I'd say seven of those pieces of content are all organic yeah that's for another topic you're seeing it with like I think I saw Mr Beast Ad for something he was doing with lows this week and I Texted
41:54 Call and I was like Uh I'm calling top on MR Beast Right now advertising for lows but it's the same thing like you saw with that right like when MR Beast got really popular all the social media went
42:05 to like quick cuts and like now Youtube videos like everybody started mimicking him and so there was like a year or two where it seemed like every thing was these every you didn't have thirty seconds
42:16 go by without a cut and then like quick action type stuff and then people start leaving the the like tags up at the front like You're You're not Gonna Wanna Miss this make sure you check out stay to
42:28 the end or like whatever and seven like it's they're just copying that format but are there call to action has been pasted at the Friday here and there and it's just something completely wild it's
42:37 like someone someone bungee jumping off of God knows what and then the craziest thumbnail and all this stuff but the beginning Yeah and now but now more organic like less produced are either highly
42:50 well produced not in that social media realm or completely unproduced just like pop your phone out selfie style those seem to be doing better than others and it's because it's beep at the end of the
43:04 day it's a human you know you're targeting a human humans want to connect with other humans not with companies are which is why we're getting into I feel like we're getting into a world now where
43:16 topics as as information heavy and as tech heavy as petroleum engineering is it needs to almost be organic because if you're just going at it like straight AI and district all your questions are AI
43:29 all of the all the footage is AI all of that it's it's simply just all going to perform well at all and and second of all it's so again it's incredibly tech heavy if you're not speaking to the truth
43:42 like again which is why we're doing this we're doing this in person rather than me just going off the Chachi between being like hey answer this for me cause I've I've given I've given Chachi BT
43:52 Questions before knowing what the answer is and it just straight up gives me the wrong answers I was telling someone a er tech this week I was comparing Chachi between Collide and I'd asked I don't
44:04 remember what the question was but I had acid and oil lagasse specific question to chat to BT knowing somewhat what the end answer would be but not fully like to it's full extent I knew that it gave
44:14 me the wrong answer So I shoved it into Clyde and Clyde gave me a much more detailed answer I was like you know what I'd much rather I'd much rather that but then again there are a whole bunch of
44:23 questions I know that we still have to get to here and that require organic answers and so do so to get into those organic answers going outside of like learning petroleum AI and like where to start
44:36 where to go and I want to talk about like the real applications and what's working and so
44:44 Yeah it was just another one to have like five in here that I want to hit all of them say so how is AI being used for due diligence Yeah so this this is one of the applications we've been approach
44:56 we've been approached with from a couple of different folks but there's this concept of data rooms in the energy business so someone is trying to sell an asset so they put together what they call the
45:07 data room which is you know their selection of The files and documents about what they're selling that they're allowing people to come look at so that they can make offers on on the property and stuff
45:19 but there's a lot of advantages that you can have if you can process that quickly and generally speaking there's each company has their specific way of kind of going through all of that data right so
45:33 you're looking at everything from you know the land and title the the you know the sales contracts as to how much you're getting paid and who's taking it away and how long and all that stuff to the
45:44 actual operations the reservoirs logs all that fun stuff and so it's it's just very data heavy and again we as humans can only process so much data it's at once or at a time and then even over time
45:59 humans become fatigued and you get worse at processing that data and so being able to have a tool like a language model go through a big giant set of data it's looking for very specific things Based
46:11 off what the operator is you know their methodology for evaluating these workflows and it's going through and basically just extracting and then allowing an aggregate view of you know he hears a
46:22 thousand documents here's a five page report on every thing that's in that document that is custom tailored to exactly how you would evaluate a normal deal and so there's additional things that you
46:34 can do on top of that as far as like a if there's database data or production data you can run that through economic models and you know do all kinds of different forecasts and stuff like that but
46:44 that's one of the big and er that's a very interesting application right now with a lot of language model suffer so it's Yeah cause I mean like with the data rooms available for a certain time but if
46:56 you can get the data in and process it quicker and maybe you can beat someone to the punch with a bid and maybe the bids but someone else won't yet but you can get a more confident answer faster
47:07 betting on other side of it too I will say you win the bid now used by the company or the rig now you Gotta get their data in and I was just talking to a company the other day out of Canada actually
47:17 and who they kind of do like
47:21 data AI things for landlines and all that kind of stuff but like they actually were able to hold an acquisition and I think they'll close in over the weekend they were able to get all the well files
47:31 and stuff migrated migrated over the weekend which as you speak a few weeks to a couple of months that process at your prior company that took months because I mean it's it's true that's a there are
47:43 so the amount of people we talked to you that have you know either data in servers or data in physical filing cabinets that they got through acquisition that they just literally don't know what all is
47:55 in there is insane but it's a very common thing in our industry right and so it's like Well first I can't deploy a language model into a filing cabinet so we've got to fix that problem but even then
48:06 it's like Okay You know You can Envision a world where it's like we point an agent at the folder of all that data and it goes through and categorize this stuff it indexes it it makes it all searchable
48:16 and discoverable and it gives you a big symbol like like yeah and you can set rules to split a file cause like I mean one of my clients was asking me hey do you know any good tools like new this
48:26 because it was like we've got our lease in here but then there's a bunch of extra junk in this PDF that isn't actually part of the lease but it can be me check stubs are just extra they got rolled in
48:35 but we only want this light and so like the yums of the platform as I can but I imagine Y'all could probably do it with right rules or understanding like you can actually split those into multiple
48:44 files and route them this one goes to this folder this one goes here yeah that's again it all boils down to the data right like at the end of the day the dating of hours or woman hours you know like
48:57 the more you give it the less time it takes to actually generate it I mean it would generally watson because you were saying something could take a week and it could take a week it could take months
49:05 what's the general life cycle of something like this if you were to do this
49:10 i was going to say if you're just going to raw dog I'd probably cut that out you if you were how long would it take if you were to do this without AI or months or even years already it never gets done
49:21 in some cases just like we'll just slam it all here and and if someone needs it you know that they'll Dig through we'll figure it out then I'm having even a Jimmy We I Say We Michelle who's on my Team
49:32 did a couple like almost one or two iterations of reorganizing the files like because we did an acquisition or two or divested and like well now's a good time to do it but she wrote Python scripts
49:41 that were pulling data from this folder and moving it over and categorizing now we want to categorize by call center now we want to do by you know DS you know like there's ways of thinking about it
49:51 but Yeah when that's like the aunt the real answer to that is obviously it depends but the you know the big thing with it is that there's so much nuance right like to that your point right Jimmy might
50:03 group their documents one way and then Devin wanted to group them a different way and so it's like if the data is all bare it's all the same but now I've got to reorganize every how that works and I
50:15 want to do that in some kind of automated fashion and normally that's by hand or writing scripts very specific scripts that do very specific things over time and then that's not even thinking about
50:26 the whole architecture of what you were using to what you're now you're going to use at the new company and like the different tools and the platforms and all that shit and so Yeah it's I mean it can
50:36 absolutely there's very justifiable reasons it can take months to years to do that but even I but I think with this because you're indexing it in certain ways like you don't even have to necessarily
50:47 think about do I want this folder charter or that folder structure I like because it is more searchable and things are related in certain ways and that now you just have this layer on top that can go
50:56 find this in the way that you're thinking about it Yeah I saw I saw a tool online Jane actually Yesterday that is explicitly for file management right and so it's like I can deploy it I can literally
51:08 use it on my system files wherever that system may be and it can just tell me what it knows the mapping of the entire file structure if I'm so it's yeah it's wild but then silly question how efficient
51:20 is to efficient if we start getting to the point where everyone is using AI to and to put all of their ideas together to quantify everything and to just make everything much quicker in in the world of
51:35 oil and gas if we're too quick do you feel like that would be detrimental or do you feel like if we're like if we're not if it's probably not the best way to put it let's just stick to how efficient
51:46 is to efficient and how detrimental is to efficient think of as efficient and trustable than that I don't think there's a problem with it but I think of as oH that was fast but then known people don't
51:57 really trust that it actually got done the right way or for any way to verify anyways know and that's that's exactly what I was going to say right like I don't in the short term as not replacing a lot
52:08 of people it's Magnifying what existing people can do is giving them time back instead of spending fifty percent of their time looking for the document they find the document and then they just so the
52:18 now they're just outputting more and you know over I don't know how many years yeah then you start automating you know you have agents I can start doing things automatically but even then it's the the
52:30 lowest then typically in my experience with technology the things that you automate first are the boring monotonous things that humans aren't good at anyway right it's like okay while you're taking
52:41 data from this invoice in your hand coding it into an accounting system no one likes to do that humans shouldn't be doing that they're not like you get tired you get lazy you fat finger something
52:52 there's no reason if you know what the mapping is and you can extract the data like why isn't AI doing that for you right and so that there's so many of those like low hanging fruit just like terrible
53:03 monotonous things that across every department that people have to do that I'm not like that's where we're going That's where this starts right and so it's like think summers is also who I was again
53:12 but about it before we came on but like distance from the Wellhead Yes mike mean the first thing that lobby will during with L M's was their open enrollment he has Alec were using AI you know like
53:23 unless there is a great use case in Hello im No Lo is high impact but low risk kind of thing but mean it's Gonna take a long time before we get to things running AI closer to the well or people were
53:35 really drives way too much efficiency and pulls humans off the site I mean like that hopefully does in some ways because it's safer it has over the years to write like when I first started fracking
53:45 and twenty ten you know I think our crew had twenty to thirty guys on it right like it I don't think there are any crews full you know twelve hour crews that have remotely close to that many people
53:57 now right and so Yeah it which you know still very torn on because it's like there's always going to have to be people at the the site like I don't I Don't or see US getting it if you met some of the
54:10 guys that were out there you might on a dichotomy of it though right cause it's like shit and AI is Gonna be a lot less you know prone to going out and drinking too much and coming in late then you
54:21 know if a guy in the field but you know the guy in the field can turn a wrench and he can physically interact because like at the end of it it all boils down to risk and if you Ain't having someone
54:31 out there as much as risky on one hand because it's a person out there but it's less risky from an operations perspective because at least if something goes wrong they can stop it bright and so Yeah
54:41 it's a tricky it's a weird thing but I agree with bobby that it's going to start with the office and and then move it's way towards the field and generally speaking again oh this one was specifically
54:55 this one was specifically for I guess the AI portion and that someone wanted to know if you Guys have any AI use cases that were on if that were that were deployed to production like what was it how
55:07 long did it take you want to do a traditional one of your own when weaken it on but Gimme your case studies having something that was we use azure I have a court document helens narrative call it but
55:23 then as more computer vision i guess that but it was just like taking like a sierra and LCR type stuff but taking the A B we had received a payout statements from ecuador that we had then we needed to
55:35 create a database of that so we were able to run that over all the power savings and pull out all the pertinent information cause we didn't have it stored in dinner or land system or anything like
55:43 that so we will pull that all out in which would take a lot of time and edit taken like a day once we got it all set up and pull that out I mean i wouldn't says huge AI literally but I mean it it made
55:54 it a lot more efficient that's the thing like there's so much so much of our industry data goes to die in a PDF and it's like PDF is the worst possible format for a lot of it but that's what everyone
56:07 uses and so you've got to have tools that and they've gotten so much better these days if you've ever tried to use tabula unlike some of these things and then you get like a printed PDF and be like
56:19 that les miserables or Yeah a printed PDF document that was originally digital gets printed so it could be signed then scan back in this is how literally pretty much all field and invoicing works and
56:33 then sent via email and so it's a scanned version of a PDF which is the absolute worst type of file that you can you can try and deal with and it sounds like that's so much more work and are so easy
56:46 though as it is I can buy a home using Docu sign but I can't get a thousand dollar field invoice signed without without going through that God forbid Oh Yeah then the Super secure stamp don't even get
56:59 yet on on the language mall side we just actually deployed one of our our first big kind of workflows around regulatory filing so Texas railroad commission has a bunch of if you operate oil and gas
57:13 wells in the state of Texas you have in pretty much every state you have all kinds of different regulatory permits and
57:19 by annual things that you've got to submit to them about a bunch of different stuff and so but a lot of that yet again that workflow starts with them sending the operator a PDF of a scanned document
57:34 that has a bunch of wells in it that they need to go get the data from and submit back to them while it's in a scanned PDF so the typical workflows someone takes that PDF and types out every single
57:46 well that is in that PDF and then they have to go find those what they use a V look up an excel workbook to go find those and then if they can't find the matches because it's not using any kind of
57:55 unique identifier just the lease name and the least number you've got to you have to then go back and
58:02 and so We've we've built a workflow that does it in less than a minute so it's pretty awesome shameless plug but shameless plug as you should wear it we're excited about it I feel like you guys need a
58:13 little theme song everytime you to shamelessly plug I'll Get Jacob to program some of those buttons on the board that we can start a little soundboard cook cook cook collide collide
58:27 mean the deepest voice possible I'm ok real quick cut are there any questions from the outside of like the real applications portion that you want to do you want to cover or do you want to just move
58:37 onto the next deck
58:41 cause I feel like we also somewhat Yeah we we touched a bit on this stuff and Yeah we can go to texts I got sick so outside of the real applications what's working the use cases and whatnot and let's
58:55 talk about tech stack and the tools cause I know that I know that tech now is constantly evolving just like the same way that we've just been talking about all the if everyone and their mother has has
59:07 a different version of Chachi BT that is doing something but obviously Chachi BT is not always the most reliable so when it comes to the platforms and tools what is best for building L M A O L L L M
59:23 Can't Speak L L M Applications in oil and gas Yeah so we mentioned cursor in windsurf those are coding tools to help you code there's a ton of Python libraries out there like you mentioned lang chain
59:38 on here Wang chain is a really good very popular very well supported LLM framework that lets you have a lot of flexibility and has a lot of tooling and stuff that makes things just a lot easier
59:51 there's haystacks another one that kind of lives in that realm pedantic as another one for python specific LLM frameworks but Yeah I would say you know and then The model of your choice right like
1:00:06 it's interesting because certain models as we've kind of already talked about are good or better at certain things than others right so like Gemini in my experience it tends to be better at
1:00:18 like this OSI are kind of extracting data than claude is I don't know why but it is and then but the flip side of that is like Claude and Gemini tend to be better at coding than GB T does and so it's
1:00:32 like but you don't learn that like that's purely anecdotal from me right like that's just me using these tools I could be completely wrong if based off of a benchmark but that's a whole nother can of
1:00:42 worms like every time a model comes out they run out against these benchmarks and a human being doesn't have any fucking clue what that means and so it's like well it it beat it in these benchmarks
1:00:52 and they want to use that from the publicity side of it but that just because it's better in one benchmark category that doesn't mean it's a better coding Assistant then another one or whatever and
1:01:03 speaking of benchmarks then bring up like snowflake and databricks and like they've got these things built into their platforms and like they're becoming like stuff that calls itself the data cloud
1:01:11 you're right and but me a lot of companies even switching on gas or using snowflake or databricks them or both in some cases where they've got all their company data especially the structure data
1:01:21 there and now you can use your stuff like as cortex and I know databricks has their own ella models within it and you're able to do stuff within like a one holistic environment rather than having to
1:01:33 cut a piecemeal together with some random python script you can do it within their platforms and an hour getting to like you know i say within the Database I thought he got these semantic layers now
1:01:44 which basically are creating context of how this table in this table are related to each other and now that's actually allowing the text to SQL or being well that just literally ask questions of your
1:01:55 data rather than needing to you know have your database guy go run a query for you you can literally ask questions of the data like how much oil do we pump on this well or within his D S U Yeah in the
1:02:06 last three months and it has the context of now to go actually write the queries behind the scenes from that information now that we've been playing with that quite a bit the last few months and it's
1:02:16 exactly for that right instead of our growth and marketing team coming to me and asking me for a specific SQL query or them trying to do it in Chachi BT themselves we've got a tool that I've connected
1:02:27 to our to the collide database and they can query it and ask all kinds of questions about it and it has been working very well but I think that kind of goes back to the point of like these tools allow
1:02:39 they democratize the data right so like whether you just need to find it faster or you need access to data that you wouldn't have previously been able to get to because it was behind a SQL query you
1:02:52 now have the ability to do those things and so because people have more access to more data theoretically at least you know they should be able to have make better decisions make them faster make them
1:03:03 you know more efficient another silly question then what if you don't have access to any of the data and you're still starting off and you still wanna to get into it and understand which platforms are
1:03:12 best to be using Yeah so I know Bobby's said this before but it's just start trying to use it for whatever applicant like whatever you're trying to do do on your daily basis right like there's not one
1:03:24 specific application that you would use if you have no data at all while some of it you've got you've got to have some data at some point or to give it but but like I got into the data side because of
1:03:36 this fantasy Football League a banana for the last fifteen years with my buddies from home right and it's like OH if I can use data to be better at that and like forecasts when my league have them at
1:03:48 two times back -to -back and anyway but it's that I mean Then I started realizing like oh there's a lot of really interesting things that I can do with this and then that just waterfalls in the rest
1:04:02 of it but I wouldn't have done that had I not had the need of like I want to win my fantasy Football League so like there's my problem statement all right how do I win my fantasy Football league
1:04:12 having any technology to learn or you have to figure out what's holding you back like me let's say a lot of people well I know excel so I'm going to do this in excel and sometimes you have to do that
1:04:19 just to get the job done cause it's you need an answer by four PM but if it's like I which I would usually do this in excel but now I want to try it in python like now I force yourself to use Python
1:04:28 python to solve that problem or now force yourself to try to use chatty with yours I mean like even with all this like there's so many ways that ivacy people post Oh I used GVT your clot or but a
1:04:39 Gemini in this way like Damn I never thought about that you know and now that it has all the context of your previous conversations and you can literally ask it questions about yourself if you want
1:04:50 the answer but like where where Am I if you want that answer The F B I One is another really good one to really get creeped out on was that it was like that someone wrote this big prompt about like
1:05:02 act as an F B I prowler and analyze the psychology of this person and AM analyze the threat level like all the shit it was is a lot better girls do for free when when one of their friends say OH I'm
1:05:14 dating this sunday this new person it's like who give me their first name that's all I need in thirty seconds you have their instagram you have their linkedin you know where they're where the mom
1:05:23 lives creepy but I mean that's I think that's where we're getting to now Yeah now again data is becoming democratised by the day more people have way more access to your point though if you want to
1:05:36 with oil and gas and want to say may you want to have access to like companies specific data but you've got folks like John Farrell and well database that have a free tier you can go pull down some
1:05:47 data out of well database and then try to solve a problem with it yeah that's a that's a really good one that we should throw out there for sure if you need access if you just need sample datasets or
1:05:58 you know let's see what that data looks like or anything like that like something to test out on your own without having to go an absolute lie all like when are they posted to a bunch of like the the
1:06:07 volume there's some there's some amazon source stuff alive or revolt like I don't remember the name evolved like database rare but like they post yup they've been actually very good historically about
1:06:16 can open sourcing their data or something you had some datasets too sounds right I think and so Yeah there's datasets out there I mean i recently had to scrape a bunch of Ohio data the Ohio public
1:06:27 data is pretty good although generally speaking and way easier to navigate than the Texas data but
1:06:34 yet again another nuance of the industry right like every single state has their own regulatory commission with their own regulatory filings with their own formatting and all of their own stuff and
1:06:44 then it becomes that whole thing of like while Texas calls it this or they all and then again we've talked about it numerous times but Texas reports oil production on a lease basis so it can take all
1:06:54 four wells maybe five wells whatever isn't smash them together say this how much oil produced so now you have to actually like the wells whereas like then Ohio or I know North Dakota or Colorado there
1:07:05 by well so like you have to know the nuances between the states you can you could you could go gravitate from Texas New Mexico and Colorado throw it all together and then like it would look totally
1:07:14 friggin wrong if you did know what you're talking about cause you wouldn't know that have actually all of the Wells Yeah why are all the texts as well as produces so much better than everywhere else
1:07:22 it's like oh there's six wells on that lease but I to Bobby's point I think fort like I literally don't know how long ago this was a year or two ago I made myself for a week try and solve all my
1:07:34 problems using chat GVT and that immediately hooked me to it cause I I did thing I tried to do things in it that I didn't have I would have never thought to try and use it for and then it did some of
1:07:46 those things way better than I ever thought that it would do and so like oh well let me keep going like you just keep going down the rabbit hole of course if you can't beat them or if you can't beat
1:07:55 eM join em yeah and so but like the easiest way to do that truly is you got to shift your mindset of like instead of doing this in excel I'm going to go do this in python or if instead of doing it
1:08:05 over here I'm going to ask chat Djibouti to do it if I need an idea for how to go about doing this as Chad Djibouti and see what it says it'll spawn like it it's really good for ideation or back and
1:08:16 forth if especially on a specific topic that you may not know someone that is an expert in to refer to so last set of questions I have about three of them for you this one is purely opinion based
1:08:29 you've got to pick one HM OK what do you think is holding back wider AI adoption and within oil and gas like why are people not super quick to just jump on it and either and take on take on a third
1:08:47 party that would like Clyde That'd be able to help them out with their well information and just like general knowledge
1:08:55 so
1:08:57 okay so the first the first piece of this is risk every oil and gas company the first thing they think of when they're evaluating any kind of decision is risk right and so the risk of AI it especially
1:09:12 in the operations side of things is it recommends that the driller do something and then the driller does it and they blow out the well and there's a catastrophic failure and people die and so it's
1:09:24 like it took me a very long time being on the tech side I got very jaded for awhile about how slow we were with technology and then I had someone break it down for me like that and I was like OK this
1:09:34 makes a lot more sense right like Facebook's risk of Facebook your picture not rendering properly on Facebook what is the actual risk to Facebook of that Pretty Much Nothing so Facebook has dozens of
1:09:45 different versions of the App running at All Different times all over the world and they can do that because the risk piece of it is not very high for them if you apply that to the oilfield the risk
1:09:57 is astronomical right like loss of human life environmental catastrophe financial things yes literally ruining companies and people and and so they're way more hesitant they're always look like we're
1:10:10 at the forefront of technology when we're looking at it but it just takes Us longer to adopt it because we want to be very responsible with how we integrate it and adopt it and make sure that we're
1:10:22 not setting people up to get injured hurt make bad decisions etc and stuff and so I think that's one of the big ones moving that despite ever that skepticism I mean like they've seen hype cycles come
1:10:34 through and I mean data science they are going to revolutionize us or companies coming from Silicon valley thinking they can solve problems and they don't even get close I was like alright I mean like
1:10:44 loving most people as a whole crossing the chasm thing right they're going to have to Do you have a lot of them are Gonna be laggards even if Nottingham Yeah I mean the whole industry is and this
1:10:54 isn't unique to the Oilfield hours per crossing the chasm great book and it tends to be fast followers are laggards right like they want someone else to D risk it for them and then once that person
1:11:07 does it then everyone will go at and you see it all the time right like it's kind of ridiculous but it also makes sense when you think about it holistically and stuff and so I I agree with bobby that
1:11:19 they explain ability piece also tAC is very good at over promising and under delivering in the short term but in the long term it does end up delivering what they say they do generally speaking and so
1:11:31 you go through these hype cycles of of technology and I think because we had just been through the whole big data data analytics cycle we which started off as all these promises of optimizing
1:11:44 everything and then every oil and gas company got into their data and realize all their data was shit and they couldn't run machine learning on it because they hadn't been preparing it properly or it
1:11:53 wasn't structured properly whatever will now we've finally got to that point they've been focusing on the data and getting it to that place and so now that you have good data implementing things like
1:12:02 machine learning or large language models actually makes it much easier to do because you're not spending a year going through and setting up all of that data and making sure it's clean structured and
1:12:12 all of that stuff and so I will say I've been shocked at how I open the industry has been to adopting things like collide over the last year but I think there's another element there is just the age
1:12:26 and element right like the average age and it will field is getting younger by the day and because of the whole great crew change thing and the younger generation expects it were digital natives right
1:12:37 like we expect things to be digital we want things to be automated we and and all of those things and so that has also helped quite a bit it's also very hard to sell a black box solution to a bunch of
1:12:48 engineers who just want to know the why and the how why I pick this right it's like if you can't explain that then it's not going to be worth it is not going to get traction and so like that's one of
1:12:56 the explicit reasons why on collide we cite all of the sources of the data like that's because I built it for myself I'm an engineer and I know that if you give me an answer and you don't give me a
1:13:06 citation of where that came from where I can at least go look at it and see like oh yeah that was right trust but verify rate it's not going to work very well and so but I think motes maybe all
1:13:17 different for L EMS are at least two is like people have had had access to it to go use it for in their day to day lives whereas like the data science machine learning were like oh we're going to
1:13:28 these nerds in the corner you know python and R and yeah go figure this out and giving an answer whereas if it wasn't nearly successful were like now the CEO of a company can go to Chatty B T or any
1:13:39 of these things and start doing something leasing themselves are Gonna see the value they can realize the value much quicker and easier Right Yeah now that's I that's very true rarely explain ability
1:13:49 of like oh we're going to build this predictive failure software it's like well how how's that going to save me money Yeah what does any of that part overhyped
1:13:59 which part like is there is there anyone that's like over hyping an AI system that's just not doing what it's supposed to OH I'm sure there Yeah I mean I think the problem is when a lot of people are
1:14:10 leading with AI like there's a lot of people just put up like
1:14:15 you know like all you're just jumping on the train trying to sell me something because there are people doing that within three months of you know coming out then all of a sudden he uses new AI
1:14:24 solution like
1:14:26 the way you came get we're ready for this or it's like the that they are AI solution as we we added our support docs to a RAg model and now we have AI driven support and it's like that's not that's
1:14:38 not the feature I wanted and my decline curve analysis is like that's not what I wanted at all but Yeah there's A The problem that I see with it is that the tech industry has gotten so good at making
1:14:51 things seem so simple that it trains the user to think OH well if they can like if they're showing it is that easy then I can do that right like I can go build my own rag yes you can using one of
1:15:05 these platforms off the shelf but it's not going to be that good once you get to a certain point like it all comes back to scaling and having true production level software and stuff but but Yeah it's
1:15:18 a conundrum because they make like their marketing and the way that they do stuff makes these things seem much easier and generally speaking they are but if you want to do something for a company like
1:15:29 at a company scale it's not nearly as simple as just dumping some things I can imagine not Yeah well it took you guys how long to to to build out collide five years and that we have been working on
1:15:41 collide for five years
1:15:43 probably two okay well then there you go this is in it well I mean it sounds like when I say only two years it sounds like that's quite quick but from the day that you that you came up with the
1:15:56 concept and you put it into motion and you got all this together while everyone and their mother is also coming out with their own or like trying to come up with their own and AI platform of sorts
1:16:06 it's like you really see how much time and effort it takes into building this out and you can't just feed information and hope for the best thing even when you're gone through multiple iterations of
1:16:18 many models of models like what it is today you may basically scrap what you doing as it started so it's like well that's the crazy part right like it's the wild west because new shit comes out every
1:16:29 day there's new models there's new methodologies people are trying an experiment and doing all these cool things which is awesome right like I love progress I love people being creative and trying to
1:16:40 solve these unique problems in different ways but because of that you as the user have to figure out like hey okay here is what we're trying to do and everything else outside of that is kind of noise
1:16:54 to an extent and like yes there might be some new shiny things but you don't have to be the first one to implement a technology you just have to be the one that does it the best right and so like we
1:17:04 don't have to keep pace with chat GB T WE just need to make sure that our clients are getting the value out of what our product that they expect and so as as technology evolves and that's the cool
1:17:17 part about being in tech in the oil and gas business is you don't have to be the first ones but if you can connect those dots and see okay GPS had this funky web search for why however long six months
1:17:29 it's like okay well we've got websearch added to collide as a backlog item and it's like every day that goes by that actually becomes easier for us to do because more and more people are building it
1:17:38 they're integrating it and so you don't necessarily want to be the first because you can let other people learn you can learn from other people
1:17:47 I mean those are really all the questions that I have I'm still processing half of it by the by if you see Me Go Blank for Thirty seconds I promise it's just the processing and if you listen close
1:17:57 enough you hear the years absolutely I mean and I'm going to be able to watch this back and just learn anything that I maybe didn't quite understand first time around which I'm sure is the majority of
1:18:09 people who are going to be watching this is like this will probably be one of those videos where and you will need to continue watching in order to understand the like the fine tuned version of that
1:18:19 and then to apply it in real time and I Am excited to kind of revisit this in a little while when I've finished my course it'll be interesting because this for me has really been like sitting town
1:18:33 sitting down with a tube to very cool university professors who just like let me let me spew questions at them back and forth until I finally understand this but being given a broader example of what
1:18:45 I'm going to be looking or two when I do start implementing this I will probably just make the learning portion better because just looking at my notes I've I've expand on them quite a lot so it's
1:18:57 going to look it's going to look very scared right now but it's like I have all of this and this is from part of module one yeah Yeah you are we're still go and I'm not done yet hang on like and I I
1:19:13 still have all of my really silly questions down there that I'm not going to show because I'm quite embarrassed of those ones that one that one's for a separate that was for a separate message when
1:19:20 we're not one for a separate call when we're not sure I'm not being filmed and but Yeah I want to I'd love to revisit this with you guys and you may or may not need to test me just not now and please
1:19:33 test me when I'm back in town we'll come back with a toast next time it'll be fun absolutely come back with a test or just come see me in montreal and you can you can grill me there there you Go Saga
1:19:43 wasn't down to go up there for Yeah Shuttle to the Saga Guys I'm going to try and get colin to let us to go up there for that but below thanks so much this has been so fun I hope everybody hopefully
1:19:54 learned something today through your and where can people find her Yeah where people find you oh Gosh where can you find me know you can find me on the trade show floor and every single oil and gas
1:20:05 trade show that has ever existed you can find me on linkedin you can find me on the and then instagram you can find me yapping away if you open up your ears will just hear me going at it from another
1:20:16 room and or if you happen to be in montreal you can stop by them in my Office and I'll Give You Guys a tour awesome there you go Guys we appreciate it bobby thanks again Man Yahoo see Thank you so
1:20:28 much for letting Me Yap at you or See You all next Time appreciated