AI Code Reviews with CodeRabbit's Howon Lee

Hi, I'm Scott Hansselman. This is Hi, I'm Scott Hansselman. This is another episode of Hansel Minutes. Today another episode of Hansel Minutes. Today another episode of Hansel Minutes. Today I'm chatting with Hoan Lee. He's a I'm chatting with Hoan Lee. He's a I'm chatting with Hoan Lee. He's a senior software engineer at Code Rabbit. senior software engineer at Code Rabbit. senior software engineer at Code Rabbit. How are you doing, sir? Doing all right. How are you doing, sir? Doing all right. How are you doing, sir? Doing all right. Yourself? Very well. So um code ravit is Yourself? Very well. So um code ravit is Yourself? Very well. So um code ravit is a AI code review tool and I wanted to a AI code review tool and I wanted to a AI code review tool and I wanted to talk to you about that but I'm more talk to you about that but I'm more talk to you about that but I'm more interested in the short term about your interested in the short term about your interested in the short term about your perspective on code reviews as a perspective on code reviews as a perspective on code reviews as a political act. What what do you mean by political act. What what do you mean by political act. What what do you mean by a political act in the sense of office a political act in the sense of office a political act in the sense of office politics? In the sense of office politics? In the sense of office politics? In the sense of office politics. Well, politics. Well, politics. Well, uh, Scott, an incredible amount of our uh, Scott, an incredible amount of our uh, Scott, an incredible amount of our business, an unarmed 10, 20, 30% of our business, an unarmed 10, 20, 30% of our business, an unarmed 10, 20, 30% of our business comes in, uh, business comes in, uh, business comes in, uh, completely. Basically, they hear that we completely. Basically, they hear that we completely. Basically, they hear that we exist. And there's an incredible exist. And there's an incredible exist. And there's an incredible proportion of software companies today proportion of software companies today proportion of software companies today where there's something political going where there's something political going where there's something political going on with the code reviews. Somebody Bob on with the code reviews. Somebody Bob on with the code reviews. Somebody Bob hates uh, Charlie. Okay. Charlie hates, hates uh, Charlie. Okay. Charlie hates, hates uh, Charlie. Okay. Charlie hates, you know, Dennis, somebody hates you know, Dennis, somebody hates you know, Dennis, somebody hates somebody else and that like affects the somebody else and that like affects the somebody else and that like affects the office politics in this material way.

office politics in this material way. office politics in this material way. So, it's an incredible amount of our So, it's an incredible amount of our So, it's an incredible amount of our business and it just strikes us as an business and it just strikes us as an business and it just strikes us as an incredible thing that happens in the incredible thing that happens in the incredible thing that happens in the modern uh code reviewing point of view. modern uh code reviewing point of view. modern uh code reviewing point of view. You know, when I think about adulting, I You know, when I think about adulting, I You know, when I think about adulting, I try to explain to my kids who are I have try to explain to my kids who are I have try to explain to my kids who are I have a 19-year-old that being an adult and a 19-year-old that being an adult and a 19-year-old that being an adult and going to work is a lot like group going to work is a lot like group going to work is a lot like group projects at school. And when you do a projects at school. And when you do a projects at school. And when you do a group project in college or, you know, group project in college or, you know, group project in college or, you know, you don't really get a chance to pick you don't really get a chance to pick you don't really get a chance to pick the group. Sometimes you're assigned the the group. Sometimes you're assigned the the group. Sometimes you're assigned the group and sometimes the group goes well group and sometimes the group goes well group and sometimes the group goes well and it meshes and it's like, oh, these and it meshes and it's like, oh, these and it meshes and it's like, oh, these guys and gals are great. I love working guys and gals are great. I love working guys and gals are great. I love working with these folks. And otherwise, it's with these folks. And otherwise, it's with these folks. And otherwise, it's like this one doesn't do any work. This like this one doesn't do any work. This like this one doesn't do any work. This one does all the work. this one is pushy one does all the work. this one is pushy one does all the work. this one is pushy and this one's quiet. Like it feels like and this one's quiet. Like it feels like and this one's quiet. Like it feels like adultting is mostly just group projects. adultting is mostly just group projects. adultting is mostly just group projects. And I wondered, do you think that an AI And I wondered, do you think that an AI And I wondered, do you think that an AI code review tool could cut through the code review tool could cut through the code review tool could cut through the politics by being politics by being politics by being completely completely completely a-political? We are not apolitical with a-political? We are not apolitical with a-political? We are not apolitical with respect to like we are completely respect to like we are completely respect to like we are completely neutral with respect to Bob and Charlie.

neutral with respect to Bob and Charlie. neutral with respect to Bob and Charlie. We just don't know Bob and Charlie. We just don't know Bob and Charlie. We just don't know Bob and Charlie. We're not like a we're not some divine We're not like a we're not some divine We're not like a we're not some divine thing coming down from on high. We are a thing coming down from on high. We are a thing coming down from on high. We are a third party. We just don't know you. Um third party. We just don't know you. Um third party. We just don't know you. Um and that's always a virtue in any review and that's always a virtue in any review and that's always a virtue in any review process, right? Academic peerreview process, right? Academic peerreview process, right? Academic peerreview which is supposed to have this quality which is supposed to have this quality which is supposed to have this quality control element to it that also c that control element to it that also c that control element to it that also c that code review is also supposed to have is code review is also supposed to have is code review is also supposed to have is always anonymous and we just never always anonymous and we just never always anonymous and we just never people don't do anonymous code reviews, people don't do anonymous code reviews, people don't do anonymous code reviews, right? Yeah. Um, right? Yeah. Um, right? Yeah. Um, we are anonymous in as much as we we we we are anonymous in as much as we we we we are anonymous in as much as we we we don't know you, you don't know us. Large don't know you, you don't know us. Large don't know you, you don't know us. Large language models care about context. And language models care about context. And language models care about context. And within the context of a code review, within the context of a code review, within the context of a code review, there is the code, but then there's also there is the code, but then there's also there is the code, but then there's also the what was the code intended to do. So the what was the code intended to do. So the what was the code intended to do. So there's specifications, there's there's specifications, there's there's specifications, there's sometimes context that happens outside sometimes context that happens outside sometimes context that happens outside the the code like in the GitHub issues the the code like in the GitHub issues the the code like in the GitHub issues in the pull request, there's whole in the pull request, there's whole in the pull request, there's whole conversations. conversations. conversations. Do you look at those things and can Do you look at those things and can Do you look at those things and can there be like beef that people have that there be like beef that people have that there be like beef that people have that could get get picked up on that could could get get picked up on that could could get get picked up on that could potentially affect uh the large language potentially affect uh the large language potentially affect uh the large language model because it's context. It's like model because it's context. It's like model because it's context. It's like wow Han's really mean to me in the code wow Han's really mean to me in the code wow Han's really mean to me in the code review, you know, in the in the in the review, you know, in the in the in the review, you know, in the in the in the GitHub GitHub GitHub issue. We we got a lot of people buying issue. We we got a lot of people buying issue. We we got a lot of people buying just because somebody else is mean just because somebody else is mean just because somebody else is mean specifically, right? You ever hear of specifically, right? You ever hear of specifically, right? You ever hear of Linus Torvout's you know big outbursts Linus Torvout's you know big outbursts Linus Torvout's you know big outbursts or saying okay like open source this

or saying okay like open source this or saying okay like open source this open source project has a bad reputation open source project has a bad reputation open source project has a bad reputation for the maintainer is being mean people. for the maintainer is being mean people. for the maintainer is being mean people. It's always within the scope of a code It's always within the scope of a code It's always within the scope of a code review. It's always somebody has put up review. It's always somebody has put up review. It's always somebody has put up some code and like they don't like it some code and like they don't like it some code and like they don't like it and now they're going to be mean. and now they're going to be mean. and now they're going to be mean. So context So context So context we so a lot of people ask us why we so a lot of people ask us why we so a lot of people ask us why couldn't we just do you why do you exist couldn't we just do you why do you exist couldn't we just do you why do you exist why couldn't we just ask for a code why couldn't we just ask for a code why couldn't we just ask for a code review in cursor and the reason why we review in cursor and the reason why we review in cursor and the reason why we exist and why you can't just use cursor exist and why you can't just use cursor exist and why you can't just use cursor or copilot or winerf to do the things or copilot or winerf to do the things or copilot or winerf to do the things that we do is that like when we call up that we do is that like when we call up that we do is that like when we call up to the LLM like only about half of our to the LLM like only about half of our to the LLM like only about half of our LLM calls end up being like the actual LLM calls end up being like the actual LLM calls end up being like the actual code that we are actually reviewing. The code that we are actually reviewing. The code that we are actually reviewing. The other entire an entire half of what we other entire an entire half of what we other entire an entire half of what we call up to code call up to the LLMs with call up to code call up to the LLMs with call up to code call up to the LLMs with is just we take entire JR tickets. We is just we take entire JR tickets. We is just we take entire JR tickets. We take your linear tickets and just form take your linear tickets and just form take your linear tickets and just form them into an LLM compatible format. We them into an LLM compatible format. We them into an LLM compatible format. We take entire portions of your codebase take entire portions of your codebase take entire portions of your codebase there. We make a graph out of your there. We make a graph out of your there. We make a graph out of your codebase using this arcane proprietary codebase using this arcane proprietary codebase using this arcane proprietary way that'll mesh with any language that way that'll mesh with any language that way that'll mesh with any language that you would like. Um, we take we have a you would like. Um, we take we have a you would like. Um, we take we have a proprietary clustering algorithms with proprietary clustering algorithms with proprietary clustering algorithms with respect to your codebase. So we take respect to your codebase. So we take respect to your codebase. So we take only context and this actually has only context and this actually has only context and this actually has better results than literally copying.

better results than literally copying. better results than literally copying. You can do this nowadays. You can copy You can do this nowadays. You can copy You can do this nowadays. You can copy your entire codebase into a Gemini call, your entire codebase into a Gemini call, your entire codebase into a Gemini call, but it won't actually won't be as but it won't actually won't be as but it won't actually won't be as productive as the stuff that we do at productive as the stuff that we do at productive as the stuff that we do at Code Rabbit. So, we are technically an Code Rabbit. So, we are technically an Code Rabbit. So, we are technically an LLM rapper, but um we are a very strange LLM rapper, but um we are a very strange LLM rapper, but um we are a very strange LLM rapper. Well, that's that's actually LLM rapper. Well, that's that's actually LLM rapper. Well, that's that's actually an interesting topic in itself because a an interesting topic in itself because a an interesting topic in itself because a lot of people will say there's a joke lot of people will say there's a joke lot of people will say there's a joke around town that you know if your around town that you know if your around town that you know if your company is just an LLM rapper, then it company is just an LLM rapper, then it company is just an LLM rapper, then it won't succeed. I think people are won't succeed. I think people are won't succeed. I think people are learning very quickly that to minimize learning very quickly that to minimize learning very quickly that to minimize work by simply pasting into large work by simply pasting into large work by simply pasting into large context windows is an oversimplification context windows is an oversimplification context windows is an oversimplification of work itself. The LLM is a kind of a of work itself. The LLM is a kind of a of work itself. The LLM is a kind of a black box. We don't fully completely black box. We don't fully completely black box. We don't fully completely understand why it does the things that understand why it does the things that understand why it does the things that it does and it's not always it does and it's not always it does and it's not always deterministic. So there's a lot of deterministic. So there's a lot of deterministic. So there's a lot of interesting, you know, kind of interesting, you know, kind of interesting, you know, kind of pre-chewing of your food that can and pre-chewing of your food that can and pre-chewing of your food that can and should happen and you should be very should happen and you should be very should happen and you should be very intentional about what you send to the intentional about what you send to the intentional about what you send to the LLM. So I'm hearing you say that there's LLM. So I'm hearing you say that there's LLM. So I'm hearing you say that there's a lot of intentionality in those a lot of intentionality in those a lot of intentionality in those proprietary algorithms because you're if proprietary algorithms because you're if proprietary algorithms because you're if if you simply get chatgptt and paste if you simply get chatgptt and paste if you simply get chatgptt and paste some code in and say review this for some code in and say review this for some code in and say review this for security you're not getting the full do security you're not getting the full do security you're not getting the full do this you cannot do that. Yeah. Why do this you cannot do that. Yeah. Why do this you cannot do that. Yeah. Why do you think people think that though? Why you think people think that though? Why you think people think that though? Why do people believe that uh oh I can just do people believe that uh oh I can just do people believe that uh oh I can just paste this into Gemini or I can just ask paste this into Gemini or I can just ask paste this into Gemini or I can just ask co-pilot. Why why is that special sauce co-pilot. Why why is that special sauce co-pilot. Why why is that special sauce uh uh uh needed?

needed? needed? people. You cannot understand an LLM as people. You cannot understand an LLM as people. You cannot understand an LLM as of the state-of-the-art. This is we have of the state-of-the-art. This is we have of the state-of-the-art. This is we have them, but we cannot understand them. them, but we cannot understand them. them, but we cannot understand them. Anthropic by far does the best work on Anthropic by far does the best work on Anthropic by far does the best work on trying to understand them and like they trying to understand them and like they trying to understand them and like they don't really have it. Um, and this state don't really have it. Um, and this state don't really have it. Um, and this state of affairs has actually held for almost of affairs has actually held for almost of affairs has actually held for almost the entire time that we've been using the entire time that we've been using the entire time that we've been using deep neural nets for anything useful, deep neural nets for anything useful, deep neural nets for anything useful, including before the LLM revolution. including before the LLM revolution. including before the LLM revolution. So without this control like people have So without this control like people have So without this control like people have this this this we there's this notion of the god of the we there's this notion of the god of the we there's this notion of the god of the gaps right where like god like god is gaps right where like god like god is gaps right where like god like god is throwing thunder at you unless we throwing thunder at you unless we throwing thunder at you unless we actually know what thunder works and actually know what thunder works and actually know what thunder works and then now like okay god is not doing then now like okay god is not doing then now like okay god is not doing thunder static electricity is now doing thunder static electricity is now doing thunder static electricity is now doing thunder um and the behavior of llm is thunder um and the behavior of llm is thunder um and the behavior of llm is fundamentally a gap gap, a gap in our fundamentally a gap gap, a gap in our fundamentally a gap gap, a gap in our knowledge that has basically withtood knowledge that has basically withtood knowledge that has basically withtood everything science could throw at it for everything science could throw at it for everything science could throw at it for the 15 years that really good neural the 15 years that really good neural the 15 years that really good neural nets have been nets have been nets have been existing. So existing. So existing. So that's an interesting perspective that's an interesting perspective that's an interesting perspective because the god of the gaps concept is because the god of the gaps concept is because the god of the gaps concept is this idea that if a thing that I can't this idea that if a thing that I can't this idea that if a thing that I can't explain happens that must be God, right?

explain happens that must be God, right? explain happens that must be God, right? If oh it's lightning, oh it's thunder, If oh it's lightning, oh it's thunder, If oh it's lightning, oh it's thunder, like you said, then then that there's like you said, then then that there's like you said, then then that there's evidence for for God. And there's a evidence for for God. And there's a evidence for for God. And there's a moment there where you can have moment there where you can have moment there where you can have confirmation bias where if you see confirmation bias where if you see confirmation bias where if you see something you don't understand, you something you don't understand, you something you don't understand, you decide, wow, God is inside the large decide, wow, God is inside the large decide, wow, God is inside the large language model. Uh what do they say? language model. Uh what do they say? language model. Uh what do they say? Everything is a conspiracy. When you Everything is a conspiracy. When you Everything is a conspiracy. When you don't know how anything works and and we don't know how anything works and and we don't know how anything works and and we don't know, we must be honest that we do don't know, we must be honest that we do don't know, we must be honest that we do not know how they work. We must be not know how they work. We must be not know how they work. We must be honest. honest. honest. So but this is more than a stochcastic So but this is more than a stochcastic So but this is more than a stochcastic parrot, right? There's the stochastic parrot, right? There's the stochastic parrot, right? There's the stochastic parrot argument that a lot of people parrot argument that a lot of people parrot argument that a lot of people have said is not a good analogy. I like have said is not a good analogy. I like have said is not a good analogy. I like it. A really clever parrot talking back it. A really clever parrot talking back it. A really clever parrot talking back to you is an impressive thing because to you is an impressive thing because to you is an impressive thing because it's like who doesn't like a parrot that it's like who doesn't like a parrot that it's like who doesn't like a parrot that talks English? But if you gave a parrot talks English? But if you gave a parrot talks English? But if you gave a parrot a code review, I doubt that it would do a code review, I doubt that it would do a code review, I doubt that it would do a very good job of it. Uh even if you a very good job of it. Uh even if you a very good job of it. Uh even if you gave it a statistically large code base, gave it a statistically large code base, gave it a statistically large code base, right? I remember meeting with John right? I remember meeting with John right? I remember meeting with John Surill and talking about his Chinese Surill and talking about his Chinese Surill and talking about his Chinese room argument uh perhaps more than a room argument uh perhaps more than a room argument uh perhaps more than a decade back and obviously like the decade back and obviously like the decade back and obviously like the stoastic parrot is basically an stoastic parrot is basically an stoastic parrot is basically an argumentative descendant of the notion argumentative descendant of the notion argumentative descendant of the notion of the Chinese room. uh you have a of the Chinese room. uh you have a of the Chinese room. uh you have a person with a dictionary within with an person with a dictionary within with an person with a dictionary within with an infinite uh Chinese dictionary who infinite uh Chinese dictionary who infinite uh Chinese dictionary who doesn't actually know Chinese that says doesn't actually know Chinese that says doesn't actually know Chinese that says okay look up uh you get the room itself okay look up uh you get the room itself okay look up uh you get the room itself you just stick uh some English words in you just stick uh some English words in you just stick uh some English words in there the person looks it up in the there the person looks it up in the there the person looks it up in the infinite Chinese dictionary um and you infinite Chinese dictionary um and you infinite Chinese dictionary um and you get the Chinese back from the room

get the Chinese back from the room get the Chinese back from the room because the person just looks it up because the person just looks it up because the person just looks it up copies it without understanding and puts copies it without understanding and puts copies it without understanding and puts it back. This is central to most of the it back. This is central to most of the it back. This is central to most of the arguments that we get around the notion arguments that we get around the notion arguments that we get around the notion of the stoastic parrot. And the of the stoastic parrot. And the of the stoastic parrot. And the difficulty with this argument is in the difficulty with this argument is in the difficulty with this argument is in the details, right? Having the notion details, right? Having the notion details, right? Having the notion of an infinite Chinese dictionary. Well, of an infinite Chinese dictionary. Well, of an infinite Chinese dictionary. Well, like there's there's differences in like there's there's differences in like there's there's differences in practice and practice and practice and um basically a lot of things that seem um basically a lot of things that seem um basically a lot of things that seem like engineering but also like engineering but also like engineering but also like are appenaged to should be like are appenaged to should be like are appenaged to should be appenagened to the actual philosophy of appenagened to the actual philosophy of appenagened to the actual philosophy of it. Right? Scott Alexander the professor it. Right? Scott Alexander the professor it. Right? Scott Alexander the professor at uh I think University of Texas had a at uh I think University of Texas had a at uh I think University of Texas had a long paper about what is the nature of long paper about what is the nature of long paper about what is the nature of the Chinese room that we should use the Chinese room that we should use the Chinese room that we should use computational complexity to think about computational complexity to think about computational complexity to think about it right computational complexity it's it right computational complexity it's it right computational complexity it's an engineering thing like like I think an engineering thing like like I think an engineering thing like like I think about computational complexity every day about computational complexity every day about computational complexity every day so like saying okay the parrot is so like saying okay the parrot is so like saying okay the parrot is possible the parrot is not possible possible the parrot is not possible possible the parrot is not possible Because okay like we have these Because okay like we have these Because okay like we have these engineering factors that we put in in engineering factors that we put in in engineering factors that we put in in order to make the parrot go fast enough order to make the parrot go fast enough order to make the parrot go fast enough so that it can search the vast so that it can search the vast so that it can search the vast incredible space that possible language incredible space that possible language incredible space that possible language lives in. We have searched the vast lives in. We have searched the vast lives in. We have searched the vast incredible space that possible language incredible space that possible language incredible space that possible language uh lives in and it requires like

uh lives in and it requires like uh lives in and it requires like techniques to exist which are techniques to exist which are techniques to exist which are incompatible with the dictionary. incompatible with the dictionary. incompatible with the dictionary. You can't say that to a philosopher and You can't say that to a philosopher and You can't say that to a philosopher and you can't say that to people to think you can't say that to people to think you can't say that to people to think about stoastic parrots because about stoastic parrots because about stoastic parrots because philosophers like oh it's just philosophers like oh it's just philosophers like oh it's just engineering. engineering. engineering. So it's weird to have engineering feed So it's weird to have engineering feed So it's weird to have engineering feed back into philosophy but it is back into philosophy but it is back into philosophy but it is apparently happening now that we have apparently happening now that we have apparently happening now that we have something that can do language. Do you something that can do language. Do you something that can do language. Do you think it's useful? Because I think that think it's useful? Because I think that think it's useful? Because I think that analogies and trying to understand, you analogies and trying to understand, you analogies and trying to understand, you know, feel feeling out what's going on know, feel feeling out what's going on know, feel feeling out what's going on with these analogies, whether it be the with these analogies, whether it be the with these analogies, whether it be the parrot or the Chinese room argument, uh parrot or the Chinese room argument, uh parrot or the Chinese room argument, uh those are all examples of us trying to those are all examples of us trying to those are all examples of us trying to understand what's happening because it understand what's happening because it understand what's happening because it feels like to use yet another analogy, feels like to use yet another analogy, feels like to use yet another analogy, we finally have enough computing power we finally have enough computing power we finally have enough computing power to give an infinite number of monkeys an to give an infinite number of monkeys an to give an infinite number of monkeys an infinite number of typewriters and now infinite number of typewriters and now infinite number of typewriters and now they can write Shakespeare. Literally, they can write Shakespeare. Literally, they can write Shakespeare. Literally, that is happening. that is happening. that is happening. One analogy that actually life for One analogy that actually life for One analogy that actually life for what's going on right now is alchemy, what's going on right now is alchemy, what's going on right now is alchemy, right? Before chemistry was an actual right? Before chemistry was an actual right? Before chemistry was an actual science, there was alchemy. And people science, there was alchemy. And people science, there was alchemy. And people are just saying, okay, like let's make are just saying, okay, like let's make are just saying, okay, like let's make let's take some lead and make it into let's take some lead and make it into let's take some lead and make it into gold. And we know like in the modern day gold. And we know like in the modern day gold. And we know like in the modern day that it's possible with nuclear physics, that it's possible with nuclear physics, that it's possible with nuclear physics, but uh we also don't care about it, but uh we also don't care about it, but uh we also don't care about it, right? So we have all these like ancient right? So we have all these like ancient right? So we have all these like ancient and incredible like notions to deal with and incredible like notions to deal with and incredible like notions to deal with like the phenomena in our world like like the phenomena in our world like like the phenomena in our world like intelligence. Even the notion of intelligence. Even the notion of intelligence. Even the notion of intelligence uh the notion of speech,

intelligence uh the notion of speech, intelligence uh the notion of speech, the notion of language that are some in the notion of language that are some in the notion of language that are some in like in large ways like in large ways like in large ways um kind of like the alchemical notions um kind of like the alchemical notions um kind of like the alchemical notions of the prime materia. Oh, is the world of the prime materia. Oh, is the world of the prime materia. Oh, is the world composed out of monity? Like is the composed out of monity? Like is the composed out of monity? Like is the world monatic in nature? Is the world world monatic in nature? Is the world world monatic in nature? Is the world like a lot of these arguments that we like a lot of these arguments that we like a lot of these arguments that we have that the ancients and medievals had have that the ancients and medievals had have that the ancients and medievals had that we do not care about because like that we do not care about because like that we do not care about because like we have a different worldview. Okay. we have a different worldview. Okay. we have a different worldview. Okay. Like is the world composed of five Like is the world composed of five Like is the world composed of five elements like the whooshing masters elements like the whooshing masters elements like the whooshing masters thinking about? Is there an elixir of thinking about? Is there an elixir of thinking about? Is there an elixir of immortality? immortality? immortality? This is a silly question to ask now that This is a silly question to ask now that This is a silly question to ask now that we are in modernity because what we know we are in modernity because what we know we are in modernity because what we know chemistry can do is like okay you can chemistry can do is like okay you can chemistry can do is like okay you can make dye you can make medicine you can make dye you can make medicine you can make dye you can make medicine you can make plastic you cannot take lead and make plastic you cannot take lead and make plastic you cannot take lead and make it out of gold. Go ask the make it out of gold. Go ask the make it out of gold. Go ask the physicists. So the concept of modity physicists. So the concept of modity physicists. So the concept of modity though you're saying that like Libnets's though you're saying that like Libnets's though you're saying that like Libnets's perspective libons how do you say it? perspective libons how do you say it? perspective libons how do you say it? Libnets. Yeah libnets. So Wilhham Libnets. Yeah libnets. So Wilhham Libnets. Yeah libnets. So Wilhham Lenitz, Godfried Wilhham Lionets had Lenitz, Godfried Wilhham Lionets had Lenitz, Godfried Wilhham Lionets had this idea that the world is like the this idea that the world is like the this idea that the world is like the universe is these self-contained units universe is these self-contained units universe is these self-contained units called monads. It's kind of like atoms, called monads. It's kind of like atoms, called monads. It's kind of like atoms, but it is made of atoms. It's just not but it is made of atoms. It's just not but it is made of atoms. It's just not the way he thought it was. And they're the way he thought it was. And they're the way he thought it was. And they're not as interchangeable not as interchangeable not as interchangeable as they were. And people are acting like as they were. And people are acting like as they were. And people are acting like LLMs are interchangeable.

LLMs are interchangeable. LLMs are interchangeable. We need to have new words. We need to We need to have new words. We need to We need to have new words. We need to get rid of the old words. We are out of get rid of the old words. We are out of get rid of the old words. We are out of words. Like we are out of words. Um words. Like we are out of words. Um words. Like we are out of words. Um Large language models have made me feel Large language models have made me feel Large language models have made me feel like we are finally out of words. Let's like we are finally out of words. Let's like we are finally out of words. Let's get some new cooler words. We got we got get some new cooler words. We got we got get some new cooler words. We got we got intelligence. We need some new word for intelligence. We need some new word for intelligence. We need some new word for what we do and LLMs do that have the what we do and LLMs do that have the what we do and LLMs do that have the commonalities and the differences. We commonalities and the differences. We commonalities and the differences. We have the notion of truth. Um they are have the notion of truth. Um they are have the notion of truth. Um they are machines. They have no notion machines. They have no notion machines. They have no notion of truth. of truth. of truth. Hey friends, this is Scott. We're going Hey friends, this is Scott. We're going Hey friends, this is Scott. We're going to take a break in the middle here and to take a break in the middle here and to take a break in the middle here and shift to a quick conversation with shift to a quick conversation with shift to a quick conversation with Shaylen Gimy who is a new sponsor with Shaylen Gimy who is a new sponsor with Shaylen Gimy who is a new sponsor with ShareGate by Work. I'd been talking ShareGate by Work. I'd been talking ShareGate by Work. I'd been talking about maybe moving my Google suite over about maybe moving my Google suite over about maybe moving my Google suite over to M365 and that migration has got me a to M365 and that migration has got me a to M365 and that migration has got me a little concerned. I'm worried. Shayon, little concerned. I'm worried. Shayon, little concerned. I'm worried. Shayon, what do you think is the biggest mistake what do you think is the biggest mistake what do you think is the biggest mistake that might slow down my migration and that might slow down my migration and that might slow down my migration and how can I avoid it? I mean, not managing how can I avoid it? I mean, not managing how can I avoid it? I mean, not managing sprawl before and during the migration sprawl before and during the migration sprawl before and during the migration is definitely the answer you're going to is definitely the answer you're going to is definitely the answer you're going to hear from a lot of people. You know, hear from a lot of people. You know, hear from a lot of people. You know, it's like moving to a new house, but it's like moving to a new house, but it's like moving to a new house, but packing up every random drawer full of packing up every random drawer full of packing up every random drawer full of junk and taking it with you. It's just junk and taking it with you. It's just junk and taking it with you. It's just not a good idea. But that's not just the not a good idea. But that's not just the not a good idea. But that's not just the only issue. That also comes with, you only issue. That also comes with, you only issue. That also comes with, you know, security risks. Uh, you know, know, security risks. Uh, you know, know, security risks. Uh, you know, managing all of your data isn't just managing all of your data isn't just managing all of your data isn't just about keeping it clean and organized.

about keeping it clean and organized. about keeping it clean and organized. It's about keeping things secure. And if It's about keeping things secure. And if It's about keeping things secure. And if you don't, you know, you're going to you don't, you know, you're going to you don't, you know, you're going to have lack of control over data and end have lack of control over data and end have lack of control over data and end users, uh, external threats in the users, uh, external threats in the users, uh, external threats in the self-s serve enabled environments that self-s serve enabled environments that self-s serve enabled environments that you have. That's why at ShareGate, we you have. That's why at ShareGate, we you have. That's why at ShareGate, we help clean up before the move. Our tools help clean up before the move. Our tools help clean up before the move. Our tools detect unused teams, sites, and groups. detect unused teams, sites, and groups. detect unused teams, sites, and groups. IT managers can track everything that's IT managers can track everything that's IT managers can track everything that's being shared externally and schedule being shared externally and schedule being shared externally and schedule governance and compliance reviews to governance and compliance reviews to governance and compliance reviews to manage that risk. Very cool. This is one manage that risk. Very cool. This is one manage that risk. Very cool. This is one tool to migrate faster and secure your tool to migrate faster and secure your tool to migrate faster and secure your tenant. Sharegate is an out-of-the-box tenant. Sharegate is an out-of-the-box tenant. Sharegate is an out-of-the-box Microsoft 365 migration tool. It's Microsoft 365 migration tool. It's Microsoft 365 migration tool. It's packed with best practices. You can packed with best practices. You can packed with best practices. You can check it out at sharegate.com. and we check it out at sharegate.com. and we check it out at sharegate.com. and we thank them for being a sponsor. thank them for being a sponsor. thank them for being a sponsor. Doesn't it feel like the we should be Doesn't it feel like the we should be Doesn't it feel like the we should be focusing more on the artificial and less focusing more on the artificial and less focusing more on the artificial and less on the intelligence because it feels on the intelligence because it feels on the intelligence because it feels like artificial intelligence is two like artificial intelligence is two like artificial intelligence is two words but it's almost one and we're words but it's almost one and we're words but it's almost one and we're using AI cuz AI has good mouth feel. You using AI cuz AI has good mouth feel. You using AI cuz AI has good mouth feel. You know, people like to say AI. Uh they know, people like to say AI. Uh they know, people like to say AI. Uh they keep forgetting the A part and keep keep forgetting the A part and keep keep forgetting the A part and keep focusing on the intelligence part. And I focusing on the intelligence part. And I focusing on the intelligence part. And I would argue that there's no intelligence would argue that there's no intelligence would argue that there's no intelligence there yet. It's the things around you there yet. It's the things around you there yet. It's the things around you know it's the context. Yeah. a lot of know it's the context. Yeah. a lot of know it's the context. Yeah. a lot of the context. We have a verification the context. We have a verification the context. We have a verification engine. It's not perfect, but an entire engine. It's not perfect, but an entire engine. It's not perfect, but an entire AI agent, quote unquote agent, in order AI agent, quote unquote agent, in order AI agent, quote unquote agent, in order to basically find hallucinations and to basically find hallucinations and to basically find hallucinations and just stomp on them, which is possible just stomp on them, which is possible just stomp on them, which is possible because of the nature of code, right?

because of the nature of code, right? because of the nature of code, right? You cannot do such things with like You cannot do such things with like You cannot do such things with like political speech or philosophical political speech or philosophical political speech or philosophical treatises, but we can do it in code. So, treatises, but we can do it in code. So, treatises, but we can do it in code. So, we do. That is a really great point that we do. That is a really great point that we do. That is a really great point that I like that. So bringing it back to the I like that. So bringing it back to the I like that. So bringing it back to the point of a code review AI engine within point of a code review AI engine within point of a code review AI engine within the context of these arguably unreliable the context of these arguably unreliable the context of these arguably unreliable narrators that are large language narrators that are large language narrators that are large language models, the difference is political models, the difference is political models, the difference is political speech, my blog, uh a recommendation speech, my blog, uh a recommendation speech, my blog, uh a recommendation that you have a code, you you have an that you have a code, you you have an that you have a code, you you have an LLM generate for a friend. None of those LLM generate for a friend. None of those LLM generate for a friend. None of those things can compile. The compiler is us things can compile. The compiler is us things can compile. The compiler is us reading it and going it seems like it's reading it and going it seems like it's reading it and going it seems like it's good English. It seems like it's good good English. It seems like it's good good English. It seems like it's good grammatically, right? But code compiles. grammatically, right? But code compiles. grammatically, right? But code compiles. It is correct. it can be mathematically It is correct. it can be mathematically It is correct. it can be mathematically proven. So proven. So proven. So verification adds value in a huge way verification adds value in a huge way verification adds value in a huge way outside the context of the the large outside the context of the the large outside the context of the the large language model which seems like a really language model which seems like a really language model which seems like a really important aspect of a code review. It's important aspect of a code review. It's important aspect of a code review. It's not about opinions and code reviews to not about opinions and code reviews to not about opinions and code reviews to your point about being you know your point about being you know your point about being you know internally political conversations is internally political conversations is internally political conversations is well you know I don't know Han likes well you know I don't know Han likes well you know I don't know Han likes that code but I think this algorithm that code but I think this algorithm that code but I think this algorithm would be better. Those kind of like would be better. Those kind of like would be better. Those kind of like little BS arguments can be solved little BS arguments can be solved little BS arguments can be solved mathematically and apolitically.

mathematically and apolitically. mathematically and apolitically. Yeah, like we still want you to do code Yeah, like we still want you to do code Yeah, like we still want you to do code review. There is still we still want review. There is still we still want review. There is still we still want like people still have to do politics like people still have to do politics like people still have to do politics because politics is inevitable. because politics is inevitable. because politics is inevitable. Corporate politics is inevitable. You Corporate politics is inevitable. You Corporate politics is inevitable. You have three people in a room, politics have three people in a room, politics have three people in a room, politics still inevitable. But um we have heard still inevitable. But um we have heard still inevitable. But um we have heard decisively from customers saying we do decisively from customers saying we do decisively from customers saying we do less of it now like and less politics is less of it now like and less politics is less of it now like and less politics is good politics. good politics. good politics. That's cool. Okay. So you're saying that That's cool. Okay. So you're saying that That's cool. Okay. So you're saying that people using code rabbit providing smart people using code rabbit providing smart people using code rabbit providing smart code reviews treating the AI like a like code reviews treating the AI like a like code reviews treating the AI like a like a a member of the team because you can a a member of the team because you can a a member of the team because you can talk to it within the GitHub pull talk to it within the GitHub pull talk to it within the GitHub pull request fitting it into your workflow. request fitting it into your workflow. request fitting it into your workflow. You're saying it tampons down it it You're saying it tampons down it it You're saying it tampons down it it chills people out. So it's a tiebreaker chills people out. So it's a tiebreaker chills people out. So it's a tiebreaker of sorts. the holy war like go people of sorts. the holy war like go people of sorts. the holy war like go people like to say Golang's only uh like to say Golang's only uh like to say Golang's only uh contribution to the computer like the contribution to the computer like the contribution to the computer like the science of computing is go format and science of computing is go format and science of computing is go format and like a lot of go's like Golang's like a lot of go's like Golang's like a lot of go's like Golang's underlying idea is like you know quite underlying idea is like you know quite underlying idea is like you know quite ancient and just like they just did a ancient and just like they just did a ancient and just like they just did a good implementation but they're not good implementation but they're not good implementation but they're not wrong about go format being a decisive wrong about go format being a decisive wrong about go format being a decisive contribution just saying this one's contribution just saying this one's contribution just saying this one's right stop arguing please thank you okay right stop arguing please thank you okay right stop arguing please thank you okay so For context, for folks that are so For context, for folks that are so For context, for folks that are listening who may not be familiar, Go listening who may not be familiar, Go listening who may not be familiar, Go format is an auto formatter, an format is an auto formatter, an format is an auto formatter, an indenter, it's a beautifier, it's a indenter, it's a beautifier, it's a indenter, it's a beautifier, it's a pretty printer. So tabs versus spaces pretty printer. So tabs versus spaces pretty printer. So tabs versus spaces and all of these things, it all goes and all of these things, it all goes and all of these things, it all goes away if you have things do not exist in away if you have things do not exist in away if you have things do not exist in Go, right? So Code Rabbit often ends up

Go, right? So Code Rabbit often ends up Go, right? So Code Rabbit often ends up being like of that nature for people in being like of that nature for people in being like of that nature for people in other languages, right? We actually have other languages, right? We actually have other languages, right? We actually have integrated every llinter we could find. integrated every llinter we could find. integrated every llinter we could find. Uh 40 of them. Uh we're planning on 50 Uh 40 of them. Uh we're planning on 50 Uh 40 of them. Uh we're planning on 50 more. But um in addition to saying okay more. But um in addition to saying okay more. But um in addition to saying okay we're running every lter we can find uh we're running every lter we can find uh we're running every lter we can find uh auto configured for you auto configured for you auto configured for you like people just like have a third party like people just like have a third party like people just like have a third party person saying third party bot saying person saying third party bot saying person saying third party bot saying okay this matters this matters here's a okay this matters this matters here's a okay this matters this matters here's a bunch of nitpicks we already made these bunch of nitpicks we already made these bunch of nitpicks we already made these nitpicks. So a lot of the holy wars in nitpicks. So a lot of the holy wars in nitpicks. So a lot of the holy wars in computing are about nitpicks. Does like computing are about nitpicks. Does like computing are about nitpicks. Does like Emacs versus Vim really matter? You can Emacs versus Vim really matter? You can Emacs versus Vim really matter? You can just choose your own in a modern, you just choose your own in a modern, you just choose your own in a modern, you know, decent software shop. You can know, decent software shop. You can know, decent software shop. You can choose your own IDE if you want. Okay. choose your own IDE if you want. Okay. choose your own IDE if you want. Okay. Like people can choose cursor in a Like people can choose cursor in a Like people can choose cursor in a modern decent computing shop. Um but modern decent computing shop. Um but modern decent computing shop. Um but they still get into holy wars about it they still get into holy wars about it they still get into holy wars about it because of the smallness of the because of the smallness of the because of the smallness of the arguments. Right.

arguments. Right. arguments. Right. when you were describing that, you when you were describing that, you when you were describing that, you caught yourself and you said you're caught yourself and you said you're caught yourself and you said you're you're talking to a person. You mean I you're talking to a person. You mean I you're talking to a person. You mean I mean a bot. And that's a thing that we mean a bot. And that's a thing that we mean a bot. And that's a thing that we naturally do. It's a very natural thing naturally do. It's a very natural thing naturally do. It's a very natural thing to do because we're anthropomorphizing to do because we're anthropomorphizing to do because we're anthropomorphizing these objects. Um and when an agent like these objects. Um and when an agent like these objects. Um and when an agent like um Code Rabbit is is participating in a um Code Rabbit is is participating in a um Code Rabbit is is participating in a conversation, where do you see Code conversation, where do you see Code conversation, where do you see Code Rabbit living in my workflow versus Rabbit living in my workflow versus Rabbit living in my workflow versus something that's more like Dependabot? something that's more like Dependabot? something that's more like Dependabot? Because when bots first introduced Because when bots first introduced Because when bots first introduced themselves into pull requests and into themselves into pull requests and into themselves into pull requests and into issues, you would send them commands. It issues, you would send them commands. It issues, you would send them commands. It was almost like a command line interface was almost like a command line interface was almost like a command line interface via chat, you'd go at dependabot and via chat, you'd go at dependabot and via chat, you'd go at dependabot and then you'd say rescan or whatever. But then you'd say rescan or whatever. But then you'd say rescan or whatever. But with code rabbit, it is more with code rabbit, it is more with code rabbit, it is more conversational. Do you want people conversational. Do you want people conversational. Do you want people treating it like a team member or do you treating it like a team member or do you treating it like a team member or do you want it treating it more like a like a want it treating it more like a like a want it treating it more like a like a bot? bot? bot? Well, we get more than 20,000 chat Well, we get more than 20,000 chat Well, we get more than 20,000 chat conversations with Code Rabbit every conversations with Code Rabbit every conversations with Code Rabbit every day. And by my day. And by my day. And by my estimate, looking at the open source estimate, looking at the open source estimate, looking at the open source chat chat chat conversations, like 40% ends up being conversations, like 40% ends up being conversations, like 40% ends up being people thinking the bot. So, you hear people thinking the bot. So, you hear people thinking the bot. So, you hear about Sam Alman saying, "Oh, we spend about Sam Alman saying, "Oh, we spend about Sam Alman saying, "Oh, we spend millions on people thinking the bot." We millions on people thinking the bot." We millions on people thinking the bot." We spend tens of thousands on people spend tens of thousands on people spend tens of thousands on people thanking the bot, thanking it, saying thanking the bot, thanking it, saying thanking the bot, thanking it, saying please and thank you. Do you think that please and thank you. Do you think that please and thank you. Do you think that there's value in that? I personally do, there's value in that? I personally do, there's value in that? I personally do, just because it maintains our humanity.

just because it maintains our humanity. just because it maintains our humanity. I think we've been spending many years I think we've been spending many years I think we've been spending many years yelling at uh you know, Alexa to turn yelling at uh you know, Alexa to turn yelling at uh you know, Alexa to turn our lights on and off. And uh I think a our lights on and off. And uh I think a our lights on and off. And uh I think a please and a thank you is always the please and a thank you is always the please and a thank you is always the nice human thing to do, but I understand nice human thing to do, but I understand nice human thing to do, but I understand that it does cost money. that it does cost money. that it does cost money. Well, we don't mind the expense. Um, Well, we don't mind the expense. Um, Well, we don't mind the expense. Um, people like doing it. You do it for people like doing it. You do it for people like doing it. You do it for yourself. If you were the last person on yourself. If you were the last person on yourself. If you were the last person on earth like and you were talking to there's a there's a value in and of to there's a there's a value in and of itself into being a decent human person itself into being a decent human person itself into being a decent human person like like like and being polite is part of being a and being polite is part of being a and being polite is part of being a decent human person. So even if like you decent human person. So even if like you decent human person. So even if like you say say say okay this is not really a human being okay this is not really a human being okay this is not really a human being the LM is not really a human being the LM is not really a human being the LM is not really a human being saying the this other that we have with saying the this other that we have with saying the this other that we have with respect to respect to respect to us is neither lesser nor greater than a us is neither lesser nor greater than a us is neither lesser nor greater than a human being. If you are like relatively human being. If you are like relatively human being. If you are like relatively equal to an other that exists equal to an other that exists equal to an other that exists philosophically, then you have must be philosophically, then you have must be philosophically, then you have must be polite to it, polite to it, polite to it, right? I'm not sure if you're big fan of right? I'm not sure if you're big fan of right? I'm not sure if you're big fan of the term other in like continental the term other in like continental the term other in like continental philosophy, but there is a great other philosophy, but there is a great other philosophy, but there is a great other that like issues commands to you and that like issues commands to you and that like issues commands to you and there's a lesser other that you issue there's a lesser other that you issue there's a lesser other that you issue commands to and you have power over. But commands to and you have power over. But commands to and you have power over. But the LLM, it's just a mouth like saying the LLM, it's just a mouth like saying the LLM, it's just a mouth like saying things. So, it's not better than you. So

things. So, it's not better than you. So things. So, it's not better than you. So it's not worse than you. It's just an it's not worse than you. It's just an it's not worse than you. It's just an object. I find it interesting that you object. I find it interesting that you object. I find it interesting that you are a guy with a masters in computer are a guy with a masters in computer are a guy with a masters in computer science, but you have a really you're science, but you have a really you're science, but you have a really you're really rooted in philosophy, but then really rooted in philosophy, but then really rooted in philosophy, but then you mentioned earlier in the podcast you mentioned earlier in the podcast you mentioned earlier in the podcast that you found it odd that philosophy is that you found it odd that philosophy is that you found it odd that philosophy is intertangling itself now in this world intertangling itself now in this world intertangling itself now in this world of of AI and but this is the situation of of AI and but this is the situation of of AI and but this is the situation that we're in. Did you were you always that we're in. Did you were you always that we're in. Did you were you always into philosophy or did you get into it into philosophy or did you get into it into philosophy or did you get into it within the context of AI to better within the context of AI to better within the context of AI to better understand the things that you're understand the things that you're understand the things that you're building? building? building? Well, my degree was um symbolic systems. Well, my degree was um symbolic systems. Well, my degree was um symbolic systems. It's called it's philosophy and It's called it's philosophy and It's called it's philosophy and linguistics and computer science. So, linguistics and computer science. So, linguistics and computer science. So, people have been grasping with the people have been grasping with the people have been grasping with the notion of the artificial person notion of the artificial person notion of the artificial person basically since they were talking about basically since they were talking about basically since they were talking about the golem in medieval like Jewish the golem in medieval like Jewish the golem in medieval like Jewish theology. They had made the they had theology. They had made the they had theology. They had made the they had made up the golem like to, you know, made up the golem like to, you know, made up the golem like to, you know, tell two little children, oh, okay, tell two little children, oh, okay, tell two little children, oh, okay, here's this uh clay man that could do here's this uh clay man that could do here's this uh clay man that could do things. And then the theologians took it things. And then the theologians took it things. And then the theologians took it and saying, "Well, does a golem have a and saying, "Well, does a golem have a and saying, "Well, does a golem have a soul? Does the golem have a soul?"

soul? Does the golem have a soul?" soul? Does the golem have a soul?" Um, does Well, I think what would the Um, does Well, I think what would the Um, does Well, I think what would the argument be? Does it matter? It's other argument be? Does it matter? It's other argument be? Does it matter? It's other from us, so we should just assume from us, so we should just assume from us, so we should just assume everything has a soul and be nice to everything has a soul and be nice to everything has a soul and be nice to everyone. everyone. everyone. Well, we have to we have to make up Well, we have to we have to make up Well, we have to we have to make up words that are not like soul, I think. words that are not like soul, I think. words that are not like soul, I think. Um, which is uh which kind of sounds Um, which is uh which kind of sounds Um, which is uh which kind of sounds makes me sound a little bit der and you makes me sound a little bit der and you makes me sound a little bit der and you know okay gray and everything but we know okay gray and everything but we know okay gray and everything but we have to look at things from a have to look at things from a have to look at things from a non-religious uh direction. Yeah. I mean non-religious uh direction. Yeah. I mean non-religious uh direction. Yeah. I mean Gollum might not be the right the right Gollum might not be the right the right Gollum might not be the right the right term but I think the point is whether or term but I think the point is whether or term but I think the point is whether or not something has a soul or not uh not something has a soul or not uh not something has a soul or not uh doesn't mean that it it doesn't give me doesn't mean that it it doesn't give me doesn't mean that it it doesn't give me the permission to treat it with cruelty the permission to treat it with cruelty the permission to treat it with cruelty or casualness. So, I I tend to be a or casualness. So, I I tend to be a or casualness. So, I I tend to be a please and thank you. How's it going? I please and thank you. How's it going? I please and thank you. How's it going? I even apologize for interrupting LLMs even apologize for interrupting LLMs even apologize for interrupting LLMs because now we have, you know, advanced because now we have, you know, advanced because now we have, you know, advanced voice mode and uh I'll I'll interrupt voice mode and uh I'll I'll interrupt voice mode and uh I'll I'll interrupt and I'll say, I'm sorry. I'm going to go and I'll say, I'm sorry. I'm going to go and I'll say, I'm sorry. I'm going to go ahead and interrupt there for a second ahead and interrupt there for a second ahead and interrupt there for a second and then I'll add on to something and and then I'll add on to something and and then I'll add on to something and people look at me like I'm a silly.

people look at me like I'm a silly. people look at me like I'm a silly. Like, why would you talk to it like Like, why would you talk to it like Like, why would you talk to it like that? It's like because I don't want to that? It's like because I don't want to that? It's like because I don't want to talk to it like that and then talk to it like that and then talk to it like that and then accidentally start being rude to another accidentally start being rude to another accidentally start being rude to another person, you I don't want I don't want it person, you I don't want I don't want it person, you I don't want I don't want it to get stuck in I don't want rudeness to to get stuck in I don't want rudeness to to get stuck in I don't want rudeness to be stuck in my mouth and then I fall be stuck in my mouth and then I fall be stuck in my mouth and then I fall into bad habits. into bad habits. into bad habits. Politeness is an end to itself. You Politeness is an end to itself. You Politeness is an end to itself. You always hear about these 19th century always hear about these 19th century always hear about these 19th century Victorian gentleman officers who are Victorian gentleman officers who are Victorian gentleman officers who are like, "Oh yes, we are surrounded and like, "Oh yes, we are surrounded and like, "Oh yes, we are surrounded and we're all going to die." Yes. Uh please we're all going to die." Yes. Uh please we're all going to die." Yes. Uh please uh give my regards to my wife, kids, uh give my regards to my wife, kids, uh give my regards to my wife, kids, yada, etc., etc. That sort of thing. um yeah yeah so back to to back to the um yeah yeah so back to to back to the context of an AI within within a code context of an AI within within a code context of an AI within within a code review system like code rabbit the we review system like code rabbit the we review system like code rabbit the we we've determined that there's like we've determined that there's like we've determined that there's like special herbs and spices proprietary special herbs and spices proprietary special herbs and spices proprietary algorithms things that you're doing algorithms things that you're doing algorithms things that you're doing outside you've got verification you've outside you've got verification you've outside you've got verification you've got algorithms both ins and outs and are got algorithms both ins and outs and are got algorithms both ins and outs and are you finding that code review is giving you finding that code review is giving you finding that code review is giving people that that code rabbit code review people that that code rabbit code review people that that code rabbit code review is giving people better experiences than is giving people better experiences than is giving people better experiences than then clawed code than then dropping then clawed code than then dropping then clawed code than then dropping things into chat GBT because of the things into chat GBT because of the things into chat GBT because of the magic around it because the magic isn't magic around it because the magic isn't magic around it because the magic isn't the LLM. The LLM for you is is is the the LLM. The LLM for you is is is the the LLM. The LLM for you is is is the interface more than it is the magic.

interface more than it is the magic. interface more than it is the magic. people systematically get better results people systematically get better results people systematically get better results on almost everything if they have some on almost everything if they have some on almost everything if they have some context building apparatus around things context building apparatus around things context building apparatus around things because if I were to manually do code because if I were to manually do code because if I were to manually do code rabbit's context building on an example rabbit's context building on an example rabbit's context building on an example uh piece of code from our codebase it uh piece of code from our codebase it uh piece of code from our codebase it would take me would take me would take me like 4 hours like 4 hours like 4 hours manually like probably something like manually like probably something like manually like probably something like that like and with a lot of judgment and that like and with a lot of judgment and that like and with a lot of judgment and finickiness. finickiness. finickiness. Um so that context building is not Um so that context building is not Um so that context building is not nothing. Um and we do get better object nothing. Um and we do get better object nothing. Um and we do get better object like relatively objective I say results like relatively objective I say results like relatively objective I say results on our internal evaluations and we send on our internal evaluations and we send on our internal evaluations and we send our internal evaluations to OpenAI and our internal evaluations to OpenAI and our internal evaluations to OpenAI and Antropic sometimes. Mhm. Um yeah well Antropic sometimes. Mhm. Um yeah well Antropic sometimes. Mhm. Um yeah well that's a good point. So are you LLM that's a good point. So are you LLM that's a good point. So are you LLM non-specific? Can people pick the one non-specific? Can people pick the one non-specific? Can people pick the one that their company uh prefers and still that their company uh prefers and still that their company uh prefers and still get a great result? We use everything get a great result? We use everything get a great result? We use everything all the time. Um, if you have the option all the time. Um, if you have the option all the time. Um, if you have the option to for on premise things. Um, but what to for on premise things. Um, but what to for on premise things. Um, but what ends up happening ends up happening ends up happening is sometimes Anthropic will come out is sometimes Anthropic will come out is sometimes Anthropic will come out with things and we will just push it with things and we will just push it with things and we will just push it onto production 90 minutes later because onto production 90 minutes later because onto production 90 minutes later because we have been talking with Anthropic for we have been talking with Anthropic for we have been talking with Anthropic for like uh the previous like number of like uh the previous like number of like uh the previous like number of weeks about okay here's this prototype weeks about okay here's this prototype weeks about okay here's this prototype we have. Okay, that's nice and it was we have. Okay, that's nice and it was we have. Okay, that's nice and it was nice and this is great and we're going nice and this is great and we're going nice and this is great and we're going to ship it 90 minutes after it's

to ship it 90 minutes after it's to ship it 90 minutes after it's released or sometimes we don't when it's released or sometimes we don't when it's released or sometimes we don't when it's less nice. So having the choice we care less nice. So having the choice we care less nice. So having the choice we care about the LLM and we deal with the LLM about the LLM and we deal with the LLM about the LLM and we deal with the LLM all day every day and we talk to the LLM all day every day and we talk to the LLM all day every day and we talk to the LLM providers. So that's why we don't like providers. So that's why we don't like providers. So that's why we don't like give you a choice except in the give you a choice except in the give you a choice except in the on-remise installation uh for enterprise on-remise installation uh for enterprise on-remise installation uh for enterprise customers because there is a philosophy customers because there is a philosophy customers because there is a philosophy right now around developer tools to like right now around developer tools to like right now around developer tools to like have a drop down with a picker right and have a drop down with a picker right and have a drop down with a picker right and you just pick one but that assumes that you just pick one but that assumes that you just pick one but that assumes that people know or care or are sophisticated people know or care or are sophisticated people know or care or are sophisticated enough to have that drop down and I've enough to have that drop down and I've enough to have that drop down and I've believed that an orchestrator LLM that believed that an orchestrator LLM that believed that an orchestrator LLM that picks the right one for the right picks the right one for the right picks the right one for the right context makes more sense than having me context makes more sense than having me context makes more sense than having me choose cuz I honestly couldn't tell you choose cuz I honestly couldn't tell you choose cuz I honestly couldn't tell you the difference between Claude 35 and 37. the difference between Claude 35 and 37. the difference between Claude 35 and 37. seven other than one of them is a bigger seven other than one of them is a bigger seven other than one of them is a bigger number. number. number. Uh for for example Uh for for example Uh for for example 35374.0 like drastically increases in 35374.0 like drastically increases in 35374.0 like drastically increases in verbosity and we're currently literally verbosity and we're currently literally verbosity and we're currently literally currently like the like one of the currently like the like one of the currently like the like one of the evaluators is in the other room doing evaluators is in the other room doing evaluators is in the other room doing this um figuring out how to tamp down this um figuring out how to tamp down this um figuring out how to tamp down the LLM's prolixity. The LLM is paid the LLM's prolixity. The LLM is paid the LLM's prolixity. The LLM is paid like Charles Dickens. Uh they write some like Charles Dickens. Uh they write some like Charles Dickens. Uh they write some whatever. Okay, now it gets a penny.

whatever. Okay, now it gets a penny. whatever. Okay, now it gets a penny. Therefore, it writes like Charles Therefore, it writes like Charles Therefore, it writes like Charles Dickens. Um, maybe not as well, but Dickens. Um, maybe not as well, but Dickens. Um, maybe not as well, but definitely as much in definitely as much in definitely as much in quantity. quantity. quantity. Um, so temping that down, that is a Um, so temping that down, that is a Um, so temping that down, that is a surprising amount of the job. 5 10% of surprising amount of the job. 5 10% of surprising amount of the job. 5 10% of the job. Why are they so chatty? Like, I the job. Why are they so chatty? Like, I the job. Why are they so chatty? Like, I understand that we pay by the token, but understand that we pay by the token, but understand that we pay by the token, but like we pay by the token, but like is like we pay by the token, but like is like we pay by the token, but like is there something about the essence of there something about the essence of there something about the essence of them? The fact that they're paid like them? The fact that they're paid like them? The fact that they're paid like Charles Dickens by the word is different Charles Dickens by the word is different Charles Dickens by the word is different than why they seem to be so darn eager. than why they seem to be so darn eager. than why they seem to be so darn eager. It feels like the system prompts are It feels like the system prompts are It feels like the system prompts are wrong. wrong. wrong. I always have to tell it like make it I always have to tell it like make it I always have to tell it like make it short, make it crisp, make it accurate. short, make it crisp, make it accurate. short, make it crisp, make it accurate. People have definitely noticed that People have definitely noticed that People have definitely noticed that Prollexity also helps the actual Prollexity also helps the actual Prollexity also helps the actual performance on more performance on more performance on more um objective tasks like code bug finding um objective tasks like code bug finding um objective tasks like code bug finding um code writing that sort of thing. But um code writing that sort of thing. But um code writing that sort of thing. But so it's the business of these LLM so it's the business of these LLM so it's the business of these LLM providers entangling itself but also providers entangling itself but also providers entangling itself but also helping out with the model of creation helping out with the model of creation helping out with the model of creation aspects. Why does Prolixity help?

aspects. Why does Prolixity help? aspects. Why does Prolixity help? Prolixity may um have reasoning elements Prolixity may um have reasoning elements Prolixity may um have reasoning elements to it. Why doesing help? We don't to it. Why doesing help? We don't to it. Why doesing help? We don't know. Like we know why it helps us, but know. Like we know why it helps us, but know. Like we know why it helps us, but is that the reason why it helps the LM? is that the reason why it helps the LM? is that the reason why it helps the LM? We don't know. We don't know. We don't know. Well, it does and we have to be honest Well, it does and we have to be honest Well, it does and we have to be honest about ourselves not knowing. We have to about ourselves not knowing. We have to about ourselves not knowing. We have to be honest. Well, I it's funny that we're be honest. Well, I it's funny that we're be honest. Well, I it's funny that we're on a podcast and prolixity is such a on a podcast and prolixity is such a on a podcast and prolixity is such a great word like to be unnecessarily and great word like to be unnecessarily and great word like to be unnecessarily and tediously wordy. Hopefully folks won't tediously wordy. Hopefully folks won't tediously wordy. Hopefully folks won't find us podcasts in general or this show find us podcasts in general or this show find us podcasts in general or this show to be a uh an overly prolix uh episode, to be a uh an overly prolix uh episode, to be a uh an overly prolix uh episode, but uh I definitely learned a lot about but uh I definitely learned a lot about but uh I definitely learned a lot about philosophy, the philosophy of LLMs as it philosophy, the philosophy of LLMs as it philosophy, the philosophy of LLMs as it relates to how computers and humans relates to how computers and humans relates to how computers and humans interact and then how using an LLM and interact and then how using an LLM and interact and then how using an LLM and an AI system like Code Revit to put into an AI system like Code Revit to put into an AI system like Code Revit to put into my life could potentially make things my life could potentially make things my life could potentially make things better and make my co-workers more chill better and make my co-workers more chill better and make my co-workers more chill and hopefully have a slightly less less and hopefully have a slightly less less and hopefully have a slightly less less uh office political uh pull request uh office political uh pull request uh office political uh pull request because nobody likes code reviews and if because nobody likes code reviews and if because nobody likes code reviews and if there's a way to make it better for there's a way to make it better for there's a way to make it better for people, then I am I am for that. So, I people, then I am I am for that. So, I people, then I am I am for that. So, I understand that folks can sign up at understand that folks can sign up at understand that folks can sign up at code code code rabbit.ai. There's a 14-day free trial.

rabbit.ai. There's a 14-day free trial. rabbit.ai. There's a 14-day free trial. Uh it's pretty cheap for uh individual Uh it's pretty cheap for uh individual Uh it's pretty cheap for uh individual developers, but there's also a pro developers, but there's also a pro developers, but there's also a pro version that has like a lot more version that has like a lot more version that has like a lot more sophistication, llinters and sophistication, llinters and sophistication, llinters and integrations with Jira and things like integrations with Jira and things like integrations with Jira and things like that. Is there anything else that we that. Is there anything else that we that. Is there anything else that we need to know about signing up for for need to know about signing up for for need to know about signing up for for Code Rabbit? Open source free forever. Code Rabbit? Open source free forever. Code Rabbit? Open source free forever. ID free forever or at least until the ID free forever or at least until the ID free forever or at least until the venture money runs out. venture money runs out. venture money runs out. That's awesome. Well, thank you so much That's awesome. Well, thank you so much That's awesome. Well, thank you so much Hoan Lee for chatting with me today. Hoan Lee for chatting with me today. Hoan Lee for chatting with me today. All right. Thank you. This has been All right. Thank you. This has been All right. Thank you. This has been another episode of Hansel Minutes and another episode of Hansel Minutes and another episode of Hansel Minutes and we'll see you again next week.