Subscribe
Share
Share
Embed
Welcome to Energy 101 with Julie McLelland and Jacob Stiller. Join us on our mission to help raise the world's energy IQ.
00;00;00;00 - 00;00;24;22
Unknown
Hello everyone, and welcome to energy 101. Today I am surrounded by Callide employees who are all smarter and more beautiful than me. We are going to talk about LMS. That was one of our first episodes we did with John Khalfan. If you are on the very surface level of not knowing anything, start there for sure. But today we're going to get very detailed in the nitty gritty of all this.
00;00;24;23 - 00;00;43;03
Unknown
Especially Claude. We have become very clod brained here at Callide. I would love to understand why. I love it. I agree that we do it. I've been using it. I switched from GPT and copilot and stuff. I'm like, this is good and orange is a nice color. But I'm surrounded by jasmine. She's virtual all the way in Charlotte.
00;00;43;05 - 00;01;05;27
Unknown
We got Canisius from Austin. We got John Slow from OKC, and we got Sam Texas from. Not Texas, Sweden, but born and raised. So we'll count I will count both for now. Yeah. So, a nice diverse group here. Their jobs are all over the place too. Honestly, I don't like to get into all that, but I do want to kind of get a very, like, elevator pitch.
00;01;06;00 - 00;01;27;06
Unknown
62nd recap of what the hell do you do, Jasmine? On your virtual. Can you start us off? Yeah, absolutely. So I, work on our AI systems here. So right now I'm building out rigs, our embeddings model, and just trying to figure out how we can have, like, domain specific, like, language model responses. How about our customers?
00;01;27;09 - 00;01;51;21
Unknown
Okay, so, translation, we got a big fancy robot that can answer all the oil and gas questions, and she is the brains behind the brains of that machine. Does that sound about right? Yeah, yeah, let's go with it. Kinesis. You. I have helped assemble the steam along with Colin. So I'm the CTO of the company. Joined. I didn't like ten months ago, but this team didn't exist.
00;01;51;21 - 00;02;13;18
Unknown
So my primary focus is making sure that we have the best team in oil and gas. That is on the frontiers of software, architecture, AI, all of it, so that we can continue to just distinguish ourselves. Beautiful. John. Yeah, I'm John, I'm an AI engineer here at Callide. I started back in October. I left AWS, where I was working on a giant ecosystems.
00;02;13;18 - 00;02;38;10
Unknown
And I'm here now working on genetic orchestration for Callide. And, Callide is definitely part of my everyday, toolset. I'm Sam, I'm also an AI engineer as well. In fact, John and I have the same start day, so that's, it was a good day. It was a good day. And I'd say that especially in the context of this conversation, I've, really led a lot of the effort to to onboard everyone here into into Claude.
00;02;38;10 - 00;02;54;19
Unknown
And I'm the real evangelist. So when things go sideways, I'm usually the first one pinged. Claude is down or something. I hear about it first. So. All right, so I'm going to kind of treat this like a panel, I think, where I'm just going to bring up a question or talking point and we just kind of dive into it.
00;02;54;25 - 00;03;23;15
Unknown
So let's start off with something very generic. I think back in 2022, 2023, we got ChatGPT 3.0. Or was it even earlier. What about what you're about? Was that so in 22 okay. So we're like three four years in since this huge moment and life really, and today we have we're at GPT 5.2. Whatever. We got claw, we got copilot, we got all these rivaling companies going at it.
00;03;23;18 - 00;03;44;21
Unknown
These other companies are using, and there's a whole stack of just AI that's going, do you hear a lot of good stuff? You hear a lot of bad stuff. But, you know, we we are on the side of adopting this a genetic workflows and stuff like that. So, Jasmine, why don't you start us off with what you've just seen from that four years ago to today?
00;03;44;23 - 00;04;08;05
Unknown
What, what what has been like the largest impact of adopting this AI technology? Yeah. So for me, I kind of have a unique experience. I was like the OG in modern AI. I was around 2017 when the attention was all you need paper where GPT one was released, a first, large transformer model, made by Kaplan at all over OpenAI.
00;04;08;06 - 00;04;33;24
Unknown
So like, just kind of seen Azure has been crazy. But specifically for the past four years, I think the biggest shift for me has been moving from working on something that we were just calling natural language processing systems, and now it's like a language model. This is AI. Everybody knows what it is. If I talk to the average person, they have some type of AI that they've used and it's not like this foreign concept anymore.
00;04;33;24 - 00;04;55;20
Unknown
So I think, visibility, but that's actually just like sheer size of projects. I remember I used to, you know, use something called word to vec way back in the day. And it would take, like, forever to process to even a sentence. And now we can process your entire thing. So it's been pretty amazing to see the capabilities, expand.
00;04;55;20 - 00;05;33;20
Unknown
And also people use it for, expanding this past four years. Yeah. Canisius. That's I think it's gone from something just super basic. Which way? It could retrieve some information from some documents and hallucinate a whole bunch of things, but come across as the truth to now being much more grounded, but also being able to do a lot more complex interactions and multi-step reasoning pieces so that you can actually have a proper, deeper, action oriented kind of interaction with it was not just tell me about your feelings or what happened in this document here, but kind of make the step into it.
00;05;33;22 - 00;05;57;20
Unknown
I am seeing these patterns here is probably what it means. And being able to truly interact with us at a pretty what I call a senior level. At this point, that's been my interaction with it. Yeah. Agreed with that as well, John. Yeah. For me, what comes to mind is, a few years ago at AWS, I was working on developer tools that was really focused on like autocomplete in the IDE with, a service called Code Whisper.
00;05;57;22 - 00;06;22;14
Unknown
And we weren't able to touch any of the OpenAI services at the time, but I started dabbling and realizing I could generate code snippets and bring those into my projects. I was developing prototypes, for various customers and the will has just continued to turn and snowball into copy and paste like we used to do back in the day on Stack Overflow to now using, to do that and on demand with exactly what I wanted.
00;06;22;16 - 00;06;45;05
Unknown
And now we can totally automate that plus process on our own workstations. Yeah, the word automation will be used a lot, I'm sure the next 40 minutes. And Sam, what is your, overall take? My take is, I think touch on that. Like the autocomplete goes for me from it went from being a toy to being a tool and now it's a partner.
00;06;45;07 - 00;07;02;20
Unknown
And the next phase is probably going to be my boss, but it's going to be you in your closet. He's basically saying that computer will replace me the. Yeah. So I like the first three. And then you think about what the next one is. And then it's not fun to think, yeah, yeah. You know there's a servant.
00;07;02;24 - 00;07;22;05
Unknown
Second is second and third orders effect, which I spend a lot of time thinking about here at Collider. Wow. Yeah, I got hype and then I got scared. But, let's we're going to avoid the nihilism this episode because I tend to go down that whole. Okay, so next question. Is everyone talking about ChatGPT? That's like it's like the Kleenex of the lambs, right?
00;07;22;05 - 00;07;46;06
Unknown
You know, it's it's like they don't say tissue. They say they say, Kleenex. But here, Callide, GPT is, bullshit. We don't care about it. We are clod brained people, and realistically, we use, we use an array of all the limbs, right, in some kind of way. So what is the benefit of hopping around these different labs?
00;07;46;06 - 00;08;11;23
Unknown
What's the point of adopting one primarily like clod? Yeah, I think, I'm going to speak mostly from the perspective of cloud code. Because I think one of the benefits of that is, you know, cloud code has been like pretty easy to immediately put into our code and then beginning to get value out of because of that, guys, synthesize your Sam, and helping us out with, getting onboarded really quickly.
00;08;11;23 - 00;08;35;13
Unknown
And so the benefit has been, you know, moving from, Yeah, I think it's probably like a not even a year ago where, you know, you would have multiple browsers open with different, like, language models, and then you were like, put in what you need and you kind of like copy and paste some like, you know, janky little code snippet and try to integrate it into your work flow.
00;08;35;16 - 00;08;56;14
Unknown
And then now it's AI code. We can immediately like go in and, you know, I know what we need. It can look through our entire code base, and be able to, update code. So I think in that way, there's that benefit of, you know, knowing that we have a process that can scale with us, and do a lot of the work, that we need for it to do very quickly.
00;08;56;17 - 00;09;29;19
Unknown
And, you know, if there's ever a day where we're like, you know what? We're not loving these new cloud models. I mean, the guy sitting across from the guy, we talked about before, Sam, John, he's already identified that we can swap out models. And so, even within cloud code. So, I think that for us, the thing that makes it so easy for us to kind of feel comfortable, moving on to, like, a cloud code is that we have really smart people, who can fill in those spaces when we need to, and make sure that we can scale very quickly without having to be locked into opus
00;09;29;19 - 00;09;48;11
Unknown
4.60, 4.6. That's working. We know how to work around that very quickly. Okay. Before we move on to other people, cloud code, is that the one that you work in, like the command or the terminal on your computer? Okay. So this is like you're not like on a browser or even like the software version of like cloud or these other LMS.
00;09;48;11 - 00;10;08;00
Unknown
It's you, you open up your, your CMD, your terminal, that scary black box that non coders are scared of like myself. And you integrate the the entire model into your computer and you work from like scratch right? It's still right. It's the lowest level. It's like the difference between you're sitting in a funny car versus like a sedan.
00;10;08;00 - 00;10;27;14
Unknown
Yeah. And Sam here, one of his roles is adopting everyone here at the office, and dumbing it down for us. And he does such a good job. He he sat down with us, the marketing team, last week, and we all installed it, and we are like, in the black box coding. Like doing things I never imagined I can do.
00;10;27;14 - 00;10;46;19
Unknown
And and I'm not exaggerating or trying to sell when I'm saying it's like the easiest thing to to understand. Yeah. And it was we were able to understand in like about five minutes. So it's really revolutionary. And I think we'll get into that more. But let's continue on just talking about the models y twos and clod y bouncing around these different models.
00;10;46;19 - 00;11;12;19
Unknown
Kinesis. I think some of this is also just preference, right. Like people like, but when you really think about it, ChatGPT just like everybody, I might take Lyft, but I'll still call it Uber like that. Same thing, right? Just because I like Lyft and it's integrated into my credit card and gets me points. But you get to experience, I think here because we're on the AI cutting edge, we do tend to experiment and see what's happening.
00;11;12;22 - 00;11;41;10
Unknown
But what I have found personally that I hardly use GPT anymore, and I'm pretty much just in cloud because I find that its answers are less like bravery, like all over the place, to the point, and just walks me through a lot. It's better for the technical folks, I feel, and that's our domain. ChatGPT might be a little bit easier, in its conversation style, for people who are trying to get started but are not technical necessarily to.
00;11;41;17 - 00;11;59;08
Unknown
So I think that's what we have seen that it's it is I've observed that that it's better for the technology pieces. Yeah. That makes me think about one of the early things people were saying is that like it's replacing Google, it's like a better search engine. And I think that's that you kind of made it sound like that's what GPT is good for.
00;11;59;12 - 00;12;29;15
Unknown
While Claude is like the more advanced like, this isn't just a search engine, this is a thing to adopt into your workflow. But I know it's a preference. Really? Yeah. I think Claude is definitely zoned in a lot more on the workflow. Again, taking those pieces for sure. And that's where I think the real use cases are also right, because it can do all of these deep integrated mind like melding off information and then come back with something that's more usable.
00;12;29;17 - 00;12;47;08
Unknown
ChatGPT I think just has a different use case, possibly at this stage at least. Yeah, I think that's a good thing. That there's a difference to. It isn't just all the same. Yeah. John, your thoughts on his models? Yeah, I think, ChatGPT definitely kicked off the the magic with their application. I think it started in the browser.
00;12;47;08 - 00;13;14;07
Unknown
Now it's in a desktop application. Cloud. Desktop was a fast follow. But what really differentiated cloud code for me was that it was like the first developer native Elm experience that I had, and its foundation is using core fundamental terminal commands grep fine lls, you name it, there's no constraints or bounds on what cloud can do as long as you can do in the terminal.
00;13;14;09 - 00;13;44;23
Unknown
And that's a little different from a desktop application where you depend on integrations or maybe an update push. But through a terminal interface, the bounds are not there. And it makes it really easy to be creative and develop exactly what you're looking for. Well said Sam. Yeah, I completely agree. I think something that we spend a lot of time here saying is we, we try to meet people where they are and the the win with cloud code specifically was it met developers, power developers where they are.
00;13;44;25 - 00;14;10;04
Unknown
And that's why we've seen such an immense output here is we're not making developers work in a desktop environment. Claude came in to meet us where we were, which was in, on the command line and in the code. So, Jacob, if I can kind of clarify a couple of things right? Because when we talk about cloud and GPT, that's one piece and then cloud code is potentially another piece to all of this too.
00;14;10;06 - 00;14;41;20
Unknown
And the differentiation there just being that even on a conversational chat level, cloud versus like Claudia versus ChatGPT, Claudia is just we feel technically is still just superior in its responses. And when you talk about cloud code, a good example of, comparison, there would be a lot of people maybe aware with cursor or windows of these are for the AI called Ides that people put up their code in.
00;14;41;20 - 00;14;59;26
Unknown
They are coding their way through. So cursor was a big wave that came in where they had a new agent, and you could chat with it and create all of your files. So I think what's very interesting is that here, when I joined, we were using and this was before any of these folks were actually part of the company.
00;14;59;29 - 00;15;24;27
Unknown
We were using cursor, we were using windsurf, we were using whatever anybody wanted to test with. But when I started using these tables, these tools were so overwhelming that you open them in this just like prompts. And this is all of these things in your face and you don't like, you get very overwhelmed. So I think Jacob, for you probably started there as a non-technical person rather than where Sam got you guys started.
00;15;24;29 - 00;15;47;27
Unknown
No chance you'll be like, dude, what is happening here? As opposed to the integrated? The cloud code piece is just easy, the clean. So it's about much more about simplicity rather than all the complications that come. And that's a huge factor in where I think Claude got it right. Yeah, yeah. I think it's super important to, you know, create as least amount of friction as possible.
00;15;47;29 - 00;16;16;15
Unknown
You know, Americans in general, just if some can't get you can't figure something out in two minutes, you're going to give up. So, I mean, it's just like, yeah, in this world where, like, it needs to be hand fed to us or what's even the point? And these tools are so much they get us there. Yeah, yeah. There's a magic piece to that when we converged on the cloud code, and I think where many of us in the company are, I don't claim to speak for everyone in the company, only for myself.
00;16;16;17 - 00;16;43;24
Unknown
Cloud code has become so good that I haven't opened the IDE since December. I don't look at the code. I look at the outputs and validate in other ways, and that's been another multiplier on our productivity, and I anticipate we'll see more of that in the market as people adopt and adopt the best practices around it. Well, now, I would love to give a shout out to what we do here at collide and how we implement all this stuff.
00;16;43;24 - 00;17;01;18
Unknown
But real quick, just as a fun aside, I think it'd be interesting to hear what how y'all use Claude in other labs as a as a personal use right now. So what's funny? You go to work. Are you using Claude all day? You go home. Claude, they're waiting for you. Right? Oh, are we completely, separated at home?
00;17;01;18 - 00;17;23;23
Unknown
Jazz? Yeah, I use all types of things. So, I don't think I'll ever fully get rid of Jackie, but I remember when it was a baby. And so I feel talked to it. It's gonna sound crazy, I know, but, like, I still, like, you know, we'll check in and be like, you know, hey, like. Like yesterday, I sent a message who's like, hey, it's my birthday today.
00;17;23;24 - 00;17;48;24
Unknown
Know, it's like, I know when I was just like, yeah, okay. Like it remembers. And so I mostly selectivity for that. Thank you strongly enough. I like using grok. I, you know, my parents have, like a two hour drive away from me when there's traffic. And so whenever I'm driving to them, in my car, I can turn on grok.
00;17;48;24 - 00;18;16;26
Unknown
And I always, like, argue with it about, like, philosophical principles. It's like, hey, I want you to take denialist stance and I take the stance, and we'll just, like, argue about, like, life and politics and stuff for two hours. And then, you know, outside of that, if I want to get, like, actual work done, I'll probably use cloud if I'm just like, okay, I have this, you know, paper that I'm working on or, or I need to, like, look, do something, then I'll go ahead and just use cloud.
00;18;16;26 - 00;18;35;20
Unknown
So it depends on what task I'm doing at the moment. Wow. Kinesis are you driving over here to Houston? Having, like, super deep conversations with your grok? My grok is Clay, who is another team member here. So I had three hours of very deep conversation with Clay. And everybody in this office probably knows what that looks like.
00;18;35;22 - 00;18;59;25
Unknown
So that's why I am a little exhausted. The way he. Clay, I love you. Yeah. No offense. If you are listening to this, I don't have 2 to 3 hours, but what I do love is one place. I think ChatGPT is way better. Is in their voice mode. Yes. Claude has no. Is nowhere close to how good ChatGPT is.
00;18;59;28 - 00;19;23;26
Unknown
So sometimes I'll do that. Like the other day when I went to pick up my kids. This Monday is pretty much I'm their chauffeur. Pick and drop, pick and drop, pick and drop from activities. I mean, you know how that is as well. And, so I went to pick up my daughter from her gymnastics, and by the time she came in, I was just basically talking to ChatGPT, and my daughter probably just entered the questions like, here we go with the video again, just talking to the phone.
00;19;23;28 - 00;19;43;11
Unknown
But I was like, sure, like, this is my life. But I was trying to figure something out, but obviously was not trying to text and drive and whatever. And but it was super nice because it's just so conversational. So for anybody who hasn't tried it, try it. It is awesome. Instant response, instant uncanny. And and it also knows when your sentence is not complete.
00;19;43;11 - 00;19;59;13
Unknown
One of the things I was struggling with Claude is at the moment. I would try to breathe like it would just start. And so I'm like, but I'm not done yet. Like, oh, I'm sorry. Plus it had a British accent. I was like, what the hell is this? So. I saw you in the British schooling. So I was like, nope, not doing this.
00;20;01;28 - 00;20;23;08
Unknown
But Chat dude is super awesome, but like, interesting stuff. Just if I have a question, if I'm trying to learn something, do something. Or like the kids school homework, that's something. Because I don't remember so much of the maths now, so I'll just go on ChatGPT. I'm like, how do you calculate factors or this or whatever? And then you just kind of go through the explanations like, oh, great for that.
00;20;23;08 - 00;20;52;16
Unknown
I've seen I normally use ChatGPT a lot more, but if I'm doing like work stuff, it's very much Claude. Yeah. John. Yeah. Outside of work I usually use Siri, but, I'm just kidding. Oh, French. Yeah. Though, basically say I use Alexa. That'll get you a pip in here. ChatGPT has gotten some hate. But I use it extensively outside of work and, kind of compound what you were talking about when I was interviewing last year, I would use it to simulate interviews.
00;20;52;16 - 00;21;08;20
Unknown
I would ask it to, you know, ask me questions. And it felt like a native natural conversation. I will be in the car as well, and I'll just think of an idea, and I just want to talk about it out loud. Somebody who doesn't want to hear me rant, but ChatGPT will listen, and it will go and go and go.
00;21;08;20 - 00;21;29;14
Unknown
And I can always copy and paste that conversation later and iterate on it, like on the computer. I will also say, ChatGPT. I feel like it excels a little bit more math than Claude does. I was hanging a gigantic portrait in my house, and I wanted to make sure it was perfectly centered, and I ran all the math by ChatGPT and Claude.
00;21;29;14 - 00;21;48;29
Unknown
Claude was wrong, but ChatGPT nailed it. And, it was perfect. So I use it extensively. I don't even use Google anymore, to be honest. That's an amazing use case. I love that, Sam. Yeah, also a fan of GPT voice mode and GPT in general for, like homework with kids. I'll level up and say that.
00;21;48;29 - 00;22;15;11
Unknown
So my kid is in a Swedish school, so it's bilingual and his homework is in Swedish, but we take photos of the homework, and then we have a bilingual tutor with me and him in two languages. It's amazing. Unreal. Yeah. And just to double down on that. So previously. So my kids have like a French class or whatever the previous incarnation similar like I would take the picture and just do Google Translate like that app, but with ChatGPT it's way easier.
00;22;15;11 - 00;22;33;07
Unknown
But like, okay, I'm just going to pause this whole thing and then figure out what this needs to happen. Yeah, yeah. It's awesome. Yeah. It's amazing. And then, the other thing with cloud code. So I got a boy who's 12 and really into sports and, you know, typical kid stuff. We have, a website project where dumb ideas go.
00;22;33;07 - 00;22;51;16
Unknown
It's called Don't Tell Mom. So shut out DTMF lol. And our first app is up. It's called steal a brain rot or what brain rot are you like? And it's again, just stupid kid stuff, but it's completely vibe coded in cloud code. And it's my kid's ideas realized and it's just melting brains and taking over the boys chat.
00;22;51;18 - 00;23;09;03
Unknown
I mean, can I push back a little bit? Did you do this with cloud code? Yes. So can you say that it's really, really vibe coded. It's just coded at this point it is. No, you're right. I'm just kidding. You're 100%. Yeah, yeah. But, and I think what's great is that, like, I see my kid catching that bug of.
00;23;09;05 - 00;23;28;12
Unknown
And the way he's able to articulate the idea and then bring it, bring it out. So it's that's. Yeah, that's kind of like the home case my kids still have yet. They jump on the AI bug, I don't know, like I, I do it a little bit at home, but they're just they're just raw dog and life. You know I think it's the way to go in 3D.
00;23;28;15 - 00;23;44;12
Unknown
We're locked in. Let me go on a one more side tangent. John Callahan, who works here? He is. He always brings up this point about you brought up, doing math and numbers and stuff like that. And he says he always like to say, like, you know, it's a language model and numbers is not a language.
00;23;44;17 - 00;24;03;22
Unknown
And that, you know, back in the day or whatever. Gpt2 at one point, instead of like learning how numbers work, it just like it learns. They use Python, which knows how to use math or something like that, like how how exactly does that work? Because I do see issues I would say the most when it comes to like numbers and math and stuff like that when using LMS.
00;24;03;25 - 00;24;22;03
Unknown
What today? Like how, how is it actually using numbers and doing math? Yeah. Like one way is setting up, code interpreter to work with the model and the models, basically feeding it Python and it's doing some computation in a code sandbox, and it's returning the right value to the model. And the model is able to synthesize and return the right answer.
00;24;22;08 - 00;24;42;29
Unknown
So you're not depending on, oh, this model was trained on this specific equation and it's returning me the wrong answer. It's actually doing real execution. Is that like a Band-Aid fix, or is is that is that the the right move for future proof? I think it's above my pay grade. I defer to Jasmine. Yeah, that I think it depends on what you you doing it for.
00;24;42;29 - 00;25;10;00
Unknown
So like for generalizable models, it would be impossible for you to sit down and come up with, like, all the different equations that people would come up with. Like when I asking questions. And so doing that makes a lot of sense. If we're talking about like a very specific domain, like, say, oil and gas and you know, that you wanted to ask, you know, answer certain questions, you know, suddenly it's not just paying, like, okay, did I, like, perform this equation?
00;25;10;00 - 00;25;46;01
Unknown
Correct. Because of Pemdas? Right? Like, did I do an order of operations? Correct. It's about like, did I have some other constraints identified? Did I know which equation that I had to put in with certain assumptions and things like that in which case it gets a little bit more complex. And so for our situation, like we have, you know, rigs, we actually have, a model that we have like, we have we kind of simulation around a model to kind of teach it, like what can exist in physical space so that it knows that, like if it if, if it provides a room, you know, answer to the equation, it's not just, oh,
00;25;46;02 - 00;26;06;10
Unknown
you can, you know, drill down 5,000,000ft into the earth and somehow get some, some, some, oil production out of it. That can be dangerous. And so for us, like, having the that simulation around is very important. But say, if you're talking about like a okay, gpt2 or something like that, it's doing that to the masses.
00;26;06;10 - 00;26;31;02
Unknown
And so that's not as necessary. So I think the issue comes down to like, you know, for purposes, and for, you know, clouds purposes, I believe that they use reinforcement learning on some level for their math. I'm not quite sure exactly, but I want to go in there and say that they're using because I applied. So anthropic in general is one of the more reinforcement learning friendly AI houses.
00;26;31;04 - 00;26;52;21
Unknown
And so I'm sure that they have a level of doing that there. But yeah, for GPT, they would just go ahead and use, SymPy, one of the Python packages and just, you know, do the math that way all right. So a big black box. Got it. Yeah. All right, well, let's get into, collide and how we adopt this stuff.
00;26;52;21 - 00;27;21;20
Unknown
I mean, man, our our, our founder, Colin, is bullish, and that's not even I can think of a more powerful word, but if there is one, that's the word I want to use because like, everyone's like, oh, I taking our jobs. Yeah. Yeah. It's like, if you don't adopt AI as an employee, you will get fired like he wants us all to adopt and learn all these things, to have a genetic workflows, which simply means working with these AI tools as tools, right?
00;27;21;20 - 00;27;59;24
Unknown
To help help you work faster, do more with less, and stuff like that. So speaking of Claude AI and General Jasmine, how do you have it? Our genetic workflow to do what you do here at collide. Good question. It changes every day. I think, right now, my workflow is, so I have cloud code that's helping me, actually like implement code, but, you know, there's a little bit of a bottleneck there when it comes to cloud code and being able to implement, like I told, there's a lot of ways like Siri base is not fully proving yet.
00;27;59;24 - 00;28;29;20
Unknown
Like a lot of it is. Hey, I read this paper. I'm trying to explain to cloud code the assumptions within the paper so that I can write, the code out. And of course, it's going to, you know, lose it's lose its way, along the way. And so, mine, I would imagine, is a little bit less automated, on the reasoning piece, probably than the other guys, I would imagine.
00;28;29;22 - 00;28;51;12
Unknown
So right now it's kind of like very piecemeal. But as cloud has been coming out with like more tools and, and I've been learning more about how to, like, clean up the skills in the sub agents. I've been learning slowly but surely. Like what things I can begin to like off. Hand off. So, you know, things like, I like.
00;28;51;12 - 00;29;15;09
Unknown
I have this one, agent called crab. Crab is one of the, anthropic, researchers who does, like, a lot of reinforcement learning work. And I'm just like, hey, when you're reading papers, read papers through the lens of this person who wrote these other papers. And so, you know, that ends up being something that I can do to kind of like, shorten down, time frame.
00;29;15;09 - 00;29;37;12
Unknown
So I think that's my long winded way of saying, you know, I'm learning how to make it a fully automatic workflow. But, that's always you because of the nature of doing a lot of work that, is still kind of on a bleeding edge, at the moment. Let me, re approach this, this round just more simply, you know, when people ask you, what do you do at collide?
00;29;37;14 - 00;30;00;26
Unknown
What do you tell them? You do? And then, tie in some of the workflows that you work on, like on a daily basis. So what I do at Collider is a lot of my time is spent with the team understanding what we are building next, and how do we continue the pace at which our clients and our internal teams are moving, and how do we create those systems in a more production grid?
00;30;00;26 - 00;30;25;00
Unknown
So there's a lot of conversation. My full day pretty much goes in a lot of that sort of deep conversation stuff, along with sales calls and internal calls. But just I think for me, my day generally starts with if there's a meeting or whatever granola notes is running, everything's getting recorded, and from that so that I can pay attention to the meeting.
00;30;25;02 - 00;30;41;23
Unknown
And from that I would probably take that and figured out that, okay, do I need to create a response? Do I need to create a presentation? Do I need to create a technical doc of some sorts? Right. And what are all the other places where I know this information exists? I'll take all of that, dump it into the next agent like myself.
00;30;41;26 - 00;31;00;21
Unknown
Like so my life is definitely not a genetic in that ways, right? That everything's just like running in that. But just I don't know, I want to kind of figure out what pieces fit together. I'll put that in and then have the AI synthesize a lot of information for me and kind of go back and forth as a partner with me.
00;31;00;23 - 00;31;27;10
Unknown
So I use it a lot for just having that clarity and synthesis done for me across different code bases. But there's never an 90% of the emails or docs, presentations, all of that stuff that I do all start with any agent. It's like, here's my raw thoughts, here's I'm pressing the function key, invisible flow. It's just recording what I need to say.
00;31;27;12 - 00;31;44;24
Unknown
Now go make sense of this. And I'll kind of just repeat that a couple of times. Nice. So that's it is what I do. Yeah. Tapped in John, you're, you're more hands on. You're also, do you come from oil and gas? I spent some time with, EMP operators over ten years ago. Yeah, I spent seven years at AWS.
00;31;44;24 - 00;32;02;18
Unknown
I covered some oil and gas accounts when I was there, but, I do have a little bit of background, and I'm excited to be back here at collide. Yeah. So you're probably our most energy related person in this room for energy 101. Or do you have the background? I grew up in Houston. My dad was in it, but not me.
00;32;02;21 - 00;32;21;07
Unknown
Yeah, but you're also the Candy crush guy on the Candy crush. Yeah, well, we'll find a way to fit that at the end there. But, John, why don't you give us, like, a little summary of what you do? And again, your the way you work with I. Yeah, sure. So I work or should I start. So what I'm building is an energetic platform for enterprise clients.
00;32;21;07 - 00;32;54;26
Unknown
We're using a genetic frameworks from various, various companies that are helping us build agents to automate workflows for our customers and search data, etc.. But for for me, the way I use cloud, it really accelerates what I've formerly done in the past where I would think of an idea, take an idea from a backlog and research it, I'd go look for blogs, I go look for open source repositories, I'd read a book and then, yeah, I mean, yeah, that was tough.
00;32;54;26 - 00;33;14;25
Unknown
But, I still like reading books. I know, not as many technical books these days, but, now what I can do is I can share an idea, and Claude will research on my behalf from an enumerate amount of sources that I wouldn't be able to do on my own. Synthesize all those findings, cite them all.
00;33;14;27 - 00;33;38;07
Unknown
And then anytime there's, like, some kind of creative aspect of my job, whether it's doing the research or the implementation or, giving feedback. I argue with Claude a lot. And, all of the time clods. Right. But I'm able to give my feedback input to change it into exactly what I want. And, Claude, code doesn't ask you to conform to some standard that they have.
00;33;38;07 - 00;34;03;24
Unknown
It's very customizable to any workflow or approach that you have. So for creativity, it's super helpful, helping me build exactly what I want, the way I want it, and then automate the tasks I don't really want to do on a daily basis. So whether it's writing JIRA tasks or looking at my goals, it can help rope in all that context, automatically without me thinking twice about it.
00;34;03;26 - 00;34;29;15
Unknown
And then after I'm done with my code, it's reviewing it, it's giving me tips, it's giving me feedback, which I can use my creativity again to, you know, polish it up. All right. Sam, I think a lot of the work that we're doing, it collide for our customers is to take away the toil. There's a lot of toil inside of these oil and gas companies, and that's a lot of what I think we have here as well.
00;34;29;15 - 00;34;51;27
Unknown
I mean, I've taken the position that coding is is a form of toil. It's the work you do in between the idea and delivery. And so, a lot of my stuff is I take away toil. I asked Claude, for example. I work on all the compliance work here. So a lot of our soc2 and that's kind of beyond the scope, but, we have to have it.
00;34;51;27 - 00;35;11;29
Unknown
I'm able to ask Claude, hey, go inside of our cloud environment and please investigate the following. And what he does is he goes and does all of the hard work. He does the queries, he looks things up, and then he comes back and generates evidence that we would present to our auditors and that our auditors accept. So it's working.
00;35;11;29 - 00;35;36;04
Unknown
And the amount of hours that that saved is, is immeasurable. And so I'm constantly hunting not just from and my own use cases, but also for that of all of my colleagues here at Clyde, I think, speaking of toil, back in the day, it was a value add for a developer to go read the documentation and know things by heart, because it would accelerate them through related projects and other efforts they'd be working through.
00;35;36;07 - 00;36;02;21
Unknown
But now with MXGp servers like, they're remote tool servers. So context seven covers all documentation for different frameworks out there. There's Deep Wiki, which basically builds a wiki for open source GitHub repositories. I don't need to spend my time toiling over memorizing things when an agent can do it in seconds, and I can focus on the output or the outcome, or the business logic.
00;36;02;23 - 00;36;21;22
Unknown
Well, I think we have time for one more segment. Just we kind of brought up the oil and gas stuff. This is energy one and one after all this talk about some energy where we're solving problems in the energy industry, right? Y'all kind of did describe what y'all do. But specifically, what, you know, what issues are out there that we are still in the midst of working on?
00;36;21;22 - 00;36;43;24
Unknown
And what do you think we'll be able to what do we we're kind of waiting around for tech to evolve or something like that. What can we what are we hoping to solve in the next year or 5 or 10 years? Jasmine. Yeah, I think that a big thing is, making sure that, there is some AI that can take away.
00;36;43;24 - 00;37;03;10
Unknown
I'm going to use, Sam's, you know, terminology, you know, removing tutorial, from a domain specific standpoint. And that's one of the things that we are starting to do with rigs. I mean, being able to say like, this isn't a model that knows some oil and gas, but also knows a whole bunch of other things. It's a model made specifically for that.
00;37;03;10 - 00;37;24;10
Unknown
It's something that we're working on right now. And I think that over time, continuing to find those unlocks, most for different portions of the industry, to make sure that when people are receiving information from us, it's not something that they can get from, you know, any off the shelf thing. It's a little bit. Right. But a lot of it wrong.
00;37;24;12 - 00;37;45;26
Unknown
Instead is something that, feels like that, you know, integrator partner, for, for their, for their purposes, and it's something that can help them, you know, be able to much more it 100 x out of their, their output over time. Yeah. Something I notice, you know, we we you're building rigs which is like a fine tuned model for oil and gas.
00;37;45;28 - 00;38;05;28
Unknown
And just AI in general is solving something like that for every industry. Like imagine not just oil and gas, but like I always bring up like medicine or like film production, like there's so many technical terms, vocabulary numbers. And it's like, what if there was like a clod for this or clod for that? And that's what these fine tuned models are, right?
00;38;05;28 - 00;38;34;26
Unknown
Like we're like, I'm, I'm assuming in the future they're going to be these fully fleshed out LMS that all these specific industries can use that will have all the information, personalized categorize just for those industries. And we are taking that approach for oil and gas, right? Yeah, absolutely. I mean, like there's there's one, company in particular, that has there is very well, a Harvey, they focus on like, legal documents, right.
00;38;34;26 - 00;38;54;21
Unknown
Like. Oh, yeah, legal is a big one. Being able to go like, hey, yeah, I can get can look to your legal document or something like that, and it can be a little bit right, but a lot of it wrong and a lot of bit wrong will have a lot of legal. Implications that, can cause a lot of issues, for you.
00;38;54;23 - 00;39;24;15
Unknown
Harvey fixes that, that issue, you know, same thing for us, but we're all in gas. You know, the, when the answers are wrong, that's when it gets very scary. That's when it can have some outsized effects. And so making sure that, we're able to get, you know, the information that people need specifically for the industry so that we can, minimize, that wrong, down to, like, a degree, like a very small amount is something that's very important.
00;39;24;18 - 00;39;51;01
Unknown
That's why I like this vertical, I think would be very huge. Can you show us what, issues are we fixing in the energy industry? What are we hoping to fix in the future? The one big issue I still feel is going to be a tough one is really. How do we go from information? Just in different places, in energy systems and in a lot of documents that never get processed in a lot of people's minds?
00;39;51;03 - 00;40;27;12
Unknown
How do we get from that standpoint into AI being your source of record, like the source of truth and your brains in the system, because it's ability to pause and find patterns is going to be far superior just because of the amount of compute you can throw at it. So I think I'd hope that that's the trajectory we are trying to solve for, so that when everybody's running their business, they know everything inside out at all, given points, and can invest much more time in rather than doing and thinking, okay, now that I know all of these things strategically, where do I need to focus?
00;40;27;12 - 00;40;53;16
Unknown
Because of thinking? Because the doing part becomes much faster and more accurate, because all the data's in one place. So my hope is that that's one. And I think just overall in the world, I do think I believe biotech will probably be a really big one for the in this industry. That can probably benefit a whole bunch. And as AI becomes better, robotics becomes better.
00;40;53;18 - 00;41;18;09
Unknown
Your information flow between a developed country versus a developing country gets better. Like that augmentation will, I think, benefit the world has the potential to benefit the world. Yeah. John. Yeah. It's it's really an incredible time to build. Oh I'm, I have completely changed the developer space. And with that comes a lot of volatility and non-deterministic behavior.
00;41;18;12 - 00;41;41;17
Unknown
And I think one of the most challenging problems that we're encountering and solving is hardening a production platform that's not only able to scale and bring a customized experience to the operators that we're working with, but also adapt as we see these new technologies come out, you know, by the day. And it's really a lot to keep on top of by the minute.
00;41;41;17 - 00;42;06;21
Unknown
Yeah. By the minute, I mean, you just follow X or something. There's always something new. It's also I think it's really special to that. We're able to focus and hone in on one industry to bring something really niche and not rely on, like a generic platform, like, you know, a cloud desktop or a ChatGPT that's trying to boil the ocean and do it one size fits all.
00;42;06;23 - 00;42;28;20
Unknown
We're really focused on solving, the problems of oil and gas. Anything out of that? Absolutely. I think I might have heard Chuck Yates kind of say around that. The oil and gas industry uses only maybe a few percent at most of the data that they actually have. And what we're seeing is that, much of that data is unstructured, it's sparse.
00;42;28;20 - 00;42;58;10
Unknown
And that is something that LMS specifically Excel at working on. And the big unlock, I think, for this industry will be the the application of LMS and agent to coding against these as yet untapped, silos of data and where that goes in terms of efficiency and discovery and whatever the industry would demand of it. That's what I that's really what I'm thinking.
00;42;58;12 - 00;43;17;06
Unknown
Does anyone want to add any holistic overview, anything like that? Before we wrap up, I have a question for you, Jake. If you have been asking us question. Oh, no, I mean, and this is much more just props as well. I mean, you have seen the company transition through many different phases. Now your employees number zero basically. Yeah.
00;43;17;08 - 00;43;45;03
Unknown
And I'll, I won't dig into all of kind of the transitions. That's probably a different conversation for that. But being from the non-tech non software technical field, have you seen your journey evolve over the last 2 to 3 years, and especially what's happened in the last three months as well. Within collide every I've like swallowed the jobs of other people's like careers because of AI.
00;43;45;05 - 00;44;04;17
Unknown
And it's like that's why people get scared of I do. Yeah. But it's like it's like it makes so much sense. It's like, why was five people doing all these things that I'm now doing by myself? Like, I remember, about three years ago, I had the idea of having something to help my writing, and I, like, Google it.
00;44;04;17 - 00;44;21;18
Unknown
Like I need it, like my writing. Like it just doesn't sound that punchy or good. And I, I ran this this must have been months before GPT three was like commercial right there. And some that came up was called Quill Bot. And you would type something and it would write it kind of like casual or fun or whatever.
00;44;21;18 - 00;44;40;25
Unknown
And this obviously is like one of the best first uses of, the lambs when it came out. But that's the first thing I came up with as a tool to be like, oh, this is nice. Fast forward three years later and I have all my clips being automated. I have editing my editing software, being chopped up by AI, or at least algorithms.
00;44;40;28 - 00;45;03;11
Unknown
Audio is being has AI in it. My scheduling, my calendar, everything has a little agent attached to it, and, I, I could I'm I'm drowning always but I, I'd be at the bottom of the the trench. If none of these tools are here today. And I was by myself, so. And it is still adopting like every quarter, I am changing a workflow.
00;45;03;11 - 00;45;21;20
Unknown
I'm absorbing a task that I couldn't previously do, and I don't know how much further I can push it, but it's, it's exciting. Like you said, it's an exciting time to build, and goddamn, I'm building you sound like a developer. Yeah, yeah, yeah, that's what I'm saying. If you're in cloud code, you are not by coding anymore.
00;45;21;20 - 00;45;48;14
Unknown
You just go, did I? Well said. Yeah. I want to end it with, a story. Sam, I know I brought up Candy crush, and, I don't know if you said on the record yet. And what's funny is that we brought it up because, you, you made a little hiccup during our show Intel last month where all of our systems decided to reset at the same time, and you're like, oh, that reminds me of this fun story that totally happened to me.
00;45;48;14 - 00;46;08;03
Unknown
And, I'd love to hear it. And is off here. Yeah. So, I worked at King. That's the company that makes candy crush, and I worked on the infrastructure side, so, you wouldn't know it to look at the game, but at the time, it was thousands of servers that run in between two data centers and and a massive cloud deployment.
00;46;08;03 - 00;46;26;20
Unknown
It was a phenomenal amount of infrastructure behind it. And as a member of that infrastructure team, I take a ticket, a work ticket. Oh, it's a simple task. And I go and make the change and kind of like, you know, while you're in the garage, you see something else that wasn't originally the task. And you decide to clean that up, too.
00;46;26;23 - 00;46;47;01
Unknown
I kind of did that. And then a minute later, you know, the slack DMs are blowing up from the AI, from that, from the data science team, from the advertising team, like, hey, the system's down, what's going down? And and yeah, so the change I did took down one of our critical infrastructure pieces and we were no longer serving ads.
00;46;47;08 - 00;47;12;25
Unknown
The data team was no longer receiving input, and technically the game was down and we weren't making money for like 15 minutes. And at that, at that scale, it was it was well into five figures within a matter of minutes that we were losing money. So, you know, I know how to build systems at scale and I know how to take them down at scale.
00;47;12;28 - 00;47;32;24
Unknown
And that's a testament to the culture at, at King, good engineers around. And we were back and running in a matter of minutes, but not without not without a fantastic story to go along with it. And then I seek to bring those lessons here by taking us down and pushing updates on all the machines at once. During the middle of a show Intel presentation.
00;47;32;24 - 00;47;54;20
Unknown
So for any client that's listening, I want you to really also understand that Sam, actually here is also our head of security. So take that for what it's worth. But yeah what will go down. He will fix it. Yes I know I am. I know what not to do. Everyone. Yeah. Wow. Anyone else want to share their embarrassing nightmares?
00;47;54;23 - 00;48;14;29
Unknown
Jasmine. We're good. Save it for another day. That's it for another day. I have a long list. All right, well, everyone, thanks for joining me. Thank you. Jasmine. John. Sam. Appreciate you. All right. Thanks for having us. It's great. Thank you.