LiveKit CEO Russ d'Sa

Auren Hoffman (00:01.432) Hello, fellow data nerds. My guest today is Russ Assah. Russ is the CEO of LiveKit, an open source platform powering real-time audio for applications like OpenAI's Chat GPT voice mode and Character AI. He was previously an engineer at Twitter and 23andMe. Russ, welcome to World of DaaS.

Russ d'Sa (00:18.61) Thanks so much, Auren. It's great to be here. I got my character hat on here. It's murdered out in black. A little hard to see, but it is there. Yeah, privilege of working with them. Yeah. Get that swag.

Auren Hoffman (00:22.727) nice. Okay.

Yeah, it looks good actually. I think I want one of those. I'm jealous. Now why is voice in for us so hard? what is and there's so many companies working on it or in and around it, but why is it just so

Russ d'Sa (00:44.134) I think at a high level, it's hard because it's just a departure from the way that the typical web application works. It's just very different. So let me put it concretely. When we started to work with OpenAI, it was partially a surprise. So I built a demo at the start of 2023 that you could use WebRTC, LiveKit's kind of infrastructure that we built.

It's built on top of that protocol. I built a demo where instead of texting with ChatGPT, you could talk to ChatGPT. And it was kind of strung together, and it was a little glue and tape, but it worked pretty well, and the latency was pretty low. I mean, for the standards of that time, Like TTS and STT still took had, they were still pretty high latency. So maybe like overall turn latency. sorry. Yeah. So like speech to text.

Auren Hoffman (01:37.889) What's STD mean?

Russ d'Sa (01:42.098) is STT and text to speech is TTS. things have changed quite a bit since the start of 2023. But the vast majority of voice interfaces since then that have been built where you can interact with an AI model, the original voice mode that OpenAI released, they use what we call a

Auren Hoffman (01:42.112) Okay, species text. Got it. Okay.

Okay.

Russ d'Sa (02:08.882) cascade or a component-based model. What that means is that the user says something. It streams to the back end. The audio streams to the back end where it is converted to text. So it goes through a STT or a

Auren Hoffman (02:22.232) And how does it work? Is it converting like while a person is talking? So someone talks for 30 seconds. It's like the first five seconds converting while they're going or how does that work?

Russ d'Sa (02:26.961) Yeah.

Russ d'Sa (02:30.748) That's correct. Even like the first 200 milliseconds, 300 milliseconds is being converted on the fly. so you're getting these like.

Auren Hoffman (02:38.004) wow. And it's using some of the same dynamics where like it can predict words to have a sense of like what word it might be based on that, based on what you said in the past. Yeah.

Russ d'Sa (02:47.992) Yeah, so these are neural nets. They're just smaller neural nets than like an LLM is. And they're kind of very tuned for the specific task. So for speech to text, it's tuned for converting audio into some kind of text-based transcription that accurately captures what the person is saying. And so it's doing that on the fly.

That's why sometimes if you talk to kind of older systems like Google, I remember Google Now has existed for a while on Android phones. when you would talk to it in a voice mode, you would see the translation or the transcription of what you're saying kind of changing in real time. And then eventually kind of gets finalized where the model has like a confidence level based on all of the things that you have said in the last second or two.

Auren Hoffman (03:38.765) Yeah.

Russ d'Sa (03:44.69) what it most likely, the most likely sentence that you uttered. But at every stop, like 300 milliseconds in, 800 milliseconds in, it's kind of changing sometimes or updating words within that sentence because it's not quite sure with a certain level of accuracy as to exactly what you said. And it's looking at the surrounding words all the time. In any case, so that audio is streaming in. It's getting converted into text.

that text is flowing into an LLM, the LLM is spinning out tokens in a response. And as those tokens are streaming out, it's going to another neural network that is converting those tokens from text back into speech and then streaming that speech out to the user. And so this kind of stringing together of the user speaking into a transcriber, then into an LLM, then into a text to speech, and then back out to the user's device where it's played out.

That's kind of called the cascaded or component And I think at a high level, the reason why it's challenging even just to do this and perform well for one user is there's a lot of variables at play. Like one is, where is a user located? Where are your models located in the world? Do you have to cross a network or a long path over a network to get the user's audio to the agent on the back end that is going to process that audio? And then you have to go through

one neural network to convert it into text, and then another neural network to process that text via the LLM, and then another neural network to convert it to speech. And you've got to do all of that with as low latency as possible. And usually, you're using different services. So you might be using ServiceX or Cloud ServiceX for the speech to text part, and then another LLM for the...

Auren Hoffman (05:35.5) Okay, I can see how it like get harder and harder and everything.

Russ d'Sa (05:38.05) And then you have many concurrent conversations you're having. it's all real time, too. So it's not like a web application where you make a request, you wait for a while, and then you get a response back. You have to have a constant connection that is going, constantly be streaming the audio, figuring out when the user is done speaking. There's a lot of conversational dynamics at play here that make it quite a departure from the traditional web application paradigm.

Auren Hoffman (05:50.36) Yeah.

Auren Hoffman (06:07.406) Now I found that when I'm talking and then I kind of a little bit ramble or um, um, um, or, something like, like it, messes it up somehow. And it's like, Oh, sorry. I didn't understand. And like, I have to like, re-repeat everything, which is unlike a real conversation. I, if I don't really know what I'm talking about in a real conversation with the human, they'll like wait for me to finish. They'll be polite. They'll kind of like, or they might jump in and kind of help me a little bit.

Russ d'Sa (06:13.542) Mm-hmm.

Yep, yep, yep.

Auren Hoffman (06:36.068) I'm going to guide me or if I'm searching for a word, they might like jump in with something that's like a normal, but the LLM like doesn't really know what to do unless I, if I haven't really thought out what I'm going to say, it might mess up. Is that, is that a common thing that, that, that, that I get afflicted with?

Russ d'Sa (06:51.866) No, yeah, it's very common. So you're not alone in that. I think a lot of people experience that. And it's what I think is one of the hardest problems in voice AI and what's truly holding back. It's not the only thing holding it back, but I think it's one of the key things that are holding back kind of massive or mainstream adoption of voice-based agents for interacting with services and applications like we will in the future. And that is this concept known as turn detection.

What you described there, that falls under the umbrella of turn detection. Turn detection is knowing when a person is done speaking or done uttering their thoughts, but also knowing when is an appropriate time for you to interrupt, if at all. It might be that the right approach for that particular kind of conversation you're having with someone and the thought that they're sharing is to just wait until they're finished speaking.

Auren Hoffman (07:50.69) Yeah, or like, like I'm doing right now, ask a clarifying question or something, right? Yeah.

Russ d'Sa (07:53.49) And yes, and you just interrupt there with an or, and it felt like the appropriate time to interrupt. And I saw you interrupt, and I kind of held back and sat back and waited for you to say what you were saying. And humans kind of do, it's extremely, it's extremely hard to do. And humans do it automatically. And I think in the future, the foundational models will as well, right?

Auren Hoffman (08:05.922) Yeah, I could see how this is so hard. my gosh. Like it's I'm now appreciating how hard this is. Yeah.

Russ d'Sa (08:20.114) anything that you, any problem really out there where humans do it well, but it's hard for you to describe exactly how you're doing it is a great candidate for machine learning and neural networks. But it requires a change in the architecture of how LLMs are set up today. So to give you an example, there's a couple of companies out there. So Qtie is a small company in France, really strong research team there.

who built this model called Moshi. It's a small 7 billion parameter model. But one unique kind of part architecturally of this model is that they have this dual channel setup where audio is streaming in. then, you know, I'm not an expert at this either, so I'm not going to explain the full thing. the model basically has architecturally a wiring or a connection, a connective tissue or

a joint attention mechanism, if you want to frame it in terms of machine learning parlance, between the input channel and the output channel. And what is the net effect of that? The net effect is audio is streaming in, and the model can also optionally stream audio out. And so it can decide, based on that input that is streaming in, whether it should interrupt or whether it should wait, kind of in a similar way that a human does. They're kind of.

monitoring the inputs constantly while also thinking about what they might say at the same time, and then choosing whether, dynamically, whether to insert themselves into that portion of the conversation or to wait for another moment. And if you want to see a similar approach in practice, you can try the full duplex mode in MetaAI's new app. And that is this dual channel.

thing as well. And when you go and you try it, you'll see that the actual AI or agent in Meta's app can make these dynamic decisions about whether to interrupt or not. And the latency is super low because it's not even converting to text as an intermediary for that particular application. It's processing the audio directly. And so

Auren Hoffman (10:39.198) that's so cool.

Russ d'Sa (10:41.136) That's the way it's going to go in the future. And it's going to be kind of widespread. Done this way is my guess. But we haven't gotten to the point yet where we have enough kind of audio-based conversational training data to build a large model that can do this.

Auren Hoffman (11:00.514) Why not? mean, you have movies, have podcasts, you have all these different things that are out there.

Russ d'Sa (11:05.877) You do. You do. think what's interesting, and this is

Auren Hoffman (11:10.264) We might all be talking like an Aaron Sorkin movie or something or whatever.

Russ d'Sa (11:13.394) Well, and so that's, I think, to do this really well. And again, this is the non expert, non, you know, researcher perspective that I'm bringing here. But I think to do it well, you have to have a wide diversity of human to human interactions, both in group settings and one on one types of interactions. Because just to use an example here of what you mentioned, if you take like a podcast, like even that we're doing right now,

There is a different kind of social contract. Exactly.

Auren Hoffman (11:47.938) Yeah, it's like an interview, which is weird, right? Like this is, this is not a normal interaction. Like if, if we were just going to dinner, I wouldn't just be like interviewing you. We would be having more of a conversation or something.

Russ d'Sa (11:59.443) Yeah. And if you look at the pattern, like if you were to take down, the, if you were to look at the actual turn taking pattern in a conversation like we're having right now, it's different than two friends sitting down for coffee and just on a Friday talking about what they're going to do over the weekend. There's just a different back and forth and a different cadence to how they're talking. And, and so you really need like a diverse representation in your training data to be able to do this.

at a human level kind of accuracy or precision.

Auren Hoffman (12:30.412) Now I use, I use these voice agents quite a bit. And what I find is that it's pretty good at like one-on-one. if me in, and which is way better than let's say if I was calling for customer service, I don't want to like use the phone tree. If someone just answers immediately, hi, I'm Mary, the AI. Can I help you? Like that's, that's great. And I've had a few scenarios where that's just been an incredible service where it's a little weird is when there's more than one human on the call.

Russ d'Sa (12:33.863) Yeah.

Russ d'Sa (12:39.719) Yes.

Russ d'Sa (12:48.05) 100%. Yep.

Russ d'Sa (12:59.398) Mm-hmm. Mm-hmm.

Auren Hoffman (13:00.526) Then it really doesn't know what's going on. It's kind of like, and sometimes even like my son and I will be like talking to let's say chat GPT. He's interested in football. And it's like the three of us are having a conversation, two humans, one AI, and it doesn't know when to jump in. And it's like, it's, a, and it doesn't even know its place. Like, should it be in listen mode? And then every once in a while, like, is is it him and I am in conversation every once in a while? You can like correct us with a new stat or throw something in or.

Like it's still, it's super choppy.

Russ d'Sa (13:34.096) Yeah, and so that adds another element of complexity to this problem, which is speaker diarization. So that's another issue here. Diorization. And that is the problem or the challenge, or sorry, like maybe the objective of figuring out who is actually talking in a group setting, right? Is it your son? Is it you? Is it your friend who's also there? And understanding.

Auren Hoffman (13:42.286) Speak or what? Diorization, okay.

Auren Hoffman (13:56.302) Mmm.

Auren Hoffman (14:01.516) Right. And they may have just walked in the room and left. Yeah.

Russ d'Sa (14:04.108) Yeah, and then also across an entire conversational context, Like what's been said so far in the last 10 minutes, who has said what and labeling the correct speakers and carrying through context, right? Like your son might have said something and the AI might want to respond to that particular point that was said earlier, not something that you just said. And so we haven't yet seen this now because of course one-on-one is a much easier problem to solve.

Auren Hoffman (14:23.736) Mm-hmm.

Auren Hoffman (14:29.997) Yes.

Russ d'Sa (14:31.504) But there is going to be this future where, yeah, you have AIs that are attending Zoom meetings for business kind of calls or things like that. Or another kind of clear example of this group dynamic that gets even more complicated is when you want to put AI NPCs in video games. And in video games, you might have multiple human players and one AI. You might have multiple NPCs and one human. You might have multiple NPCs and multiple humans. And it's like,

Auren Hoffman (14:54.211) Yep.

Auren Hoffman (14:57.751) Yeah.

Russ d'Sa (15:00.636) how do you kind of work out all of those conversational quirks and dynamics such that it feels natural to interact with these characters in the video game in the same way that you might interact with the human? So it is a really hard set of problems. And we're kind of just starting to get the turn detection piece correct for one-on-one conversations. But then, yeah, we have to solve speaker diarization. And then we have to do turn detection in group settings. And so it is kind

quite challenging. There are some companies out there that have the data sets to potentially solve this more easily than others. I've been talking to friends about this, Zoom and Discord have just a ton of multi-user conversational data between many humans at once that could be

Auren Hoffman (15:52.152) Yep.

Russ d'Sa (15:57.77) used for training these models. But of course, there's security and privacy kinds of considerations there at play. But yeah, it's a hard problem to solve in terms of the overall kind of conversational dynamics that humans have.

Auren Hoffman (16:14.446) And where are we going in the short term, like a year from now? Like what should we expect?

Russ d'Sa (16:21.874) I would say that a year from now, you can expect that turn detection for one-on-one is solved. think we're definitely going to get to that point within the next 12 to 18 months. I lean more towards 12 than 18. We're starting to see a lot of progress already. If you look at Meta's new model, it's really impressive in how conversational it is, how it can understand when to interrupt.

I think another piece that comes maybe right after doing turn detection for a piece that comes after doing turn detection for one-on-one conversations without any visual input is adding visual input to the mix. So you'll notice that for turn detection, when you're talking on the phone with someone, humans still kind of mess up on, sometimes they talk over each other and late.

Auren Hoffman (17:16.022) Yeah, you go. You go. You go. yeah. Yeah.

Russ d'Sa (17:18.258) Exactly. So latency is an aspect of that, right? And that's going to be embedded just in having conversations over computer networks versus having conversations in real life where the latency is almost instantaneous. So that is always going to be somewhat of a persistent issue that we can, know, life, it can help kind of tune that down on the network propagation side or transmission side. But when you add like a video input,

Auren Hoffman (17:21.901) Yeah.

Russ d'Sa (17:46.642) to the mix like when

Auren Hoffman (17:47.702) Or you can see like right now, people aren't watching the video. If they are, they can see me like leaning in a bed. I'm like, so you can see I'm probably going to talk then or something, right? Yeah.

Russ d'Sa (17:53.842) Exactly. There's like nonverbal cues that humans automatically pick up on that help inform whether it's time for you to talk or whether it's an appropriate time for you to interject, et cetera. And so absent that kind of visual or nonverbal cues like

Auren Hoffman (18:08.802) And also like you might have a sense that the other person's a little bored of what you're saying. And so you might just stop talking or, you know, say, yeah, it's like these are all these things that are out there.

Russ d'Sa (18:16.306) 100 % and then I think another element here that I didn't talk too much about, but I think will kind of get encoded into these models with enough training data is that human conversations are self-healing. So if you start to talk over me, but then you see I continue to like speak and I almost like reject your interjection, you'll kind of back down, right? Or like another thing is, yes.

Auren Hoffman (18:42.796) Yeah, right. Especially in a group setting. Like I want to jump in with something cumulus, but then like it's hard. You can't always find the right time to jump in when you're talking with four or five people or something.

Russ d'Sa (18:53.368) Exactly. And then I might be saying something and I might realize, oops, like I wasn't supposed to say something here. And then you'll kind of taper off, right? And then like wait, and then finally share your point or continue. so it is, and these changes, by the way, are happening like, we didn't talk really about latency, but like these changes are happening like within 200 milliseconds. And so that's really the threshold that, you know, I think, yes, some turn taking can be longer, but

Auren Hoffman (18:58.336) Mm-hmm. Mm.

Russ d'Sa (19:19.94) it can get as low as 200 milliseconds. Like a human can react to these things almost immediately. And so we got to get like kind of the overall end to end down to that level to make this feel truly at the fidelity of a human to human conversation.

Auren Hoffman (19:36.736) Interesting. so, okay, so we're going to have the turn taking is going to be solved essentially in the next 12 months.

Russ d'Sa (19:43.414) I it will be solved. Yeah, maybe it's good to put a little bit more heft around what I mean by solve. I think it will be solved to the degree that I don't think that the latency is going to get down to 200 milliseconds in the next 12 months, the full end to end. I think it could.

Auren Hoffman (20:03.276) What is it now?

Russ d'Sa (20:06.322) I'd say the vast majority

Auren Hoffman (20:06.498) Like when I call like a chat GPT or something like that or whatever.

Russ d'Sa (20:09.65) Yeah, I would say it's like, well, so for GPT-4.0, it can get as low as 300 to 320 milliseconds for the end-to-end turn latency. However, I would say it probably settles more around like the 600 or 700 is what I would say maybe.

Auren Hoffman (20:29.23) Okay. And that's why it feels a little bit unnatural.

Russ d'Sa (20:34.428) That's why it feels unnatural. think there's another component to it that is kind of hidden underneath, which is how long does the inference take? And I think there's a transmission or propagation latency, which is going over our network in the ChatGPT case. And that's pretty consistent. I think the part where you see the variability comes from the speed of inference. And so that can vary widely. If you're using a

a llama hosted on Cerebris, for example, you can get that latency really low. I mean, you can get it down to like 150 milliseconds. The trade off there is like, how big is the model? Like how smart is that model? And what's the quality of the response? So there is a...

Auren Hoffman (21:18.35) And one of the problems with ChatGPT is just like, it just, it just is often overwhelmed. So even if you pay for like the top eight, you know, like I do, it often will just like have to, you have to reload it or it just doesn't work or something like that. Like it's super buggy on that side of things. The infrared still needs a lot of work. Yeah.

Russ d'Sa (21:37.554) And yeah, so that's like the variance between the P50 and the P95 is just so wide that I think that's what kind of contributes to that average moving up for its response speed. So that's one part. So I think that like I think turn detection in the sense that like today how it's often annoying where you talk to chat GPT.

Auren Hoffman (21:44.909) Yes. Yeah.

Russ d'Sa (22:04.07) And then if you just wait even a moment, let's say you use ChatGPT to practice being an interviewer with you for a podcast and you have the questions ahead of time, and you're going through and trying to practice how you're going to respond to things. I'm not doing that here, by the way. But you practice how you're going to respond to things, and then you pause for a moment to collect your thoughts or to formulate how you're going to word something. And then all of a sudden, ChatGPT starts to talk to you again. I think that those like,

Auren Hoffman (22:20.588) Hehehe

Russ d'Sa (22:33.441) low-hanging fruit of turn detection is going to be solved, I think it will start to feel quite realistic for those one-on-one cases where you're talking to a customer support agent or something like that, doing interview practice. I think it will get solved for those. I don't think we're going to have all of the speaker diarization and conversational dynamics between humans and 200 millisecond kind of back and forth or repartee between two humans. I don't think that's going to be there.

in 12 months, but I think within 18 to 24, it definitely could be.

Auren Hoffman (23:06.508) Now beyond voice, like I assume you're using a lot of like AI tools for a lot of, whether it's personal things or for business, like what are some of the things that you're using it for that maybe like the average person isn't?

Russ d'Sa (23:19.057) the average person. Well, the average person, that can, that that's, yeah.

Auren Hoffman (23:23.5) Or maybe the average listener of this that might be technical, but you know, something you're doing, is it for fun or for, for personal use or for, businesses?

Russ d'Sa (23:34.63) Yeah, I'd say for my voice example here is really around education. So the way that I learn about new topics now is using ChatGPT Advanced Voice Mode. And I set it up while I'm on a commute or on a drive somewhere. I pop in an AirPod and an Earbud, I guess, I was going to say. And I just have a conversation. I can ask this AI model literally anything about a topic that I want to learn about. And it

Auren Hoffman (24:02.7) And do you think about ahead of time, like, okay, I'm going for a drive. I, like, kind of think about this thing ahead of time. What I want to ask. And I have a sense of the types of questions I want to ask.

Russ d'Sa (24:13.01) Sometimes, but other times it's...

Auren Hoffman (24:15.566) Because I find like when I do it, that's how it works. When if I'm, if I'm a little bit meandering, it's, it's harder.

Russ d'Sa (24:22.658) So if you're meandering, it's definitely harder because of the turn detection. think if the turn detection gets better and it gets better at understanding how to respond to you, we actually, LiveKit, we trained a turn detection model of our own internally, and it works quite a bit better than the existing. But it's not a final solution because it doesn't take into account how you're saying things, just the content of what you're saying in addition to the audio signal.

Auren Hoffman (24:25.804) Yeah.

Russ d'Sa (24:51.332) And so it's not perfect, but it's definitely a step improvement over what is existing in the ChatGPT app. Once that gets better, I think that you'll be able to have more of a natural meandering conversation where the topic you discuss with ChatGPT can be a bit more spontaneous. But I think for me, it's a mix where sometimes I know what I want to talk about going in. Like, how does Lightning work?

And sometimes I want to talk about a topic. What I'll do is I'll actually ask it, hey, what kind of frontier topic in science or technology can we talk about? And give me some suggestions. And then it'll be like, well, are you up to speed on the latest in quantum computing? And then I'll be like, actually, I'm not. I've never understood qubits and how they work. And we'll go deep on qubits, right? Or on new and.

Auren Hoffman (25:28.775) that's cool.

Auren Hoffman (25:37.592) Right.

Auren Hoffman (25:41.486) That's cool. Okay. I need to like, so almost like, Hey, here's a topic. That's right. I don't, it's really about the questions. Like, like in any good interview, coming up with the right questions.

Russ d'Sa (25:50.812) Yeah. so hard. So hard. We've been doing a few meetups in SF recently. And it's the first time where I've been the person on the other side who's really doing the interviewing. And I have so much respect for you and other folks that are kind of in the interviewer seat, because I realize asking the right questions, being like the DJ in a sense of this session, is a lot harder than just answering questions. It's crazy.

Auren Hoffman (26:03.662) Uh-huh.

Auren Hoffman (26:17.006) Well, I haven't had like a really, a good, really long conversation with like a chat, TBT or something like that, but it is super helpful when I've got a specific thing I don't understand. I don't understand this thing about options trading or something. And then it could start explaining it to me. And then I can also level set it pretty quickly. I understand that piece of it, but not this piece, or you're getting a little bit too high level. Can you get a little bit down?

You know, get it, get it more as a 12th grade level for me or something. And then it can level set pretty quickly and get me to the answer. But where I need to get much better in is, is, is getting it to help me ask better questions somehow. Like, Hey, what questions would you ask? Or, and maybe it could come up with some really good ones or something.

Russ d'Sa (26:55.803) Totally.

Russ d'Sa (27:06.354) I see. Yeah, it could be definitely used for research and things like that. And I'll sometimes use it too in that way, not through voice. I'll use it through text for that, like if I'm going to ask some questions. Like this last meetup that we just did, we had Justin Newberry, who's the head of real time at OpenAI and also an advisor to LiveKit, but also the creator of the WebRTC protocol, which LiveKit has built on top of.

I wanted to understand how can I go deep on some of these questions based on his career and what he's done and places he's worked and companies he started and kind of tying it into what he's working on now at OpenAI. And I enabled deep research and I said, hey, put together a talk track for me. It's going to be about a half an hour talk and what kinds of questions should I dive into or should I consider? it was pretty good about that. But I do agree with you that

Auren Hoffman (27:42.411) Uh-huh.

Russ d'Sa (28:04.118) One thing that's kind of tricky is having a few back and forths and refinement, like a refinement process on a conversation like that. I think it's a little bit more one shot today. I think it, especially over voice, it needs to be a little bit more collaborative. And you know what's kind of getting in that direction is like these coding agents. So like cursor.

Auren Hoffman (28:13.953) Yeah.

Auren Hoffman (28:23.278) Especially over voice, right? Yeah.

Russ d'Sa (28:32.934) Well, my example was going to be that I don't think is outside of the realm of folks that are listening to the show. But I also, common AI use case for me is I use Cursor to build stuff. And sometimes Vibe Code, sometimes Copilot on code. But I use it. the Copilot, sorry, the Cursor agent kind of experience is a little bit better at kind of.

having this multi-step going deeper and deeper on a code base and kind of starting at a high level and then drilling into certain parts. It's a bit better at that, but it also still struggles in that you have to start a completely new chat at a certain point because it kind of rabbit holes on one thing. And it's hard to move it out of some kind of local maxima or minima, depending on which direction you want to think about it from. But yeah, I think it is still a weakness.

Auren Hoffman (29:15.757) Yeah.

Auren Hoffman (29:25.762) When I watch people who are really good at using like a cursor or windsurf in like an IDE standpoint, or like, you know, maybe using CSS better with like a bolt.new or lovable or something. When I watch the people really good at that, they are iterating with like hundreds, sometimes thousands of prompts where they're asking it to change. Hey, can you please move this? Or can I, can you move this? Or can you change this to a darker blue or.

Russ d'Sa (29:32.55) Yep.

Auren Hoffman (29:53.282) This didn't work out or now I don't understand. And it's, it's almost like this very weird conversation over text that they're having with this thing. And it's very strange, like very small number of people are going to be good at this prompt back and forth.

Russ d'Sa (30:00.53) Totally.

Russ d'Sa (30:09.04) Yep. Yep. I don't think that's the long, like, to, I think that that's a bug, not a feature, I guess, is what I'm saying. If you remember, like, the early Dali kind of, like, product, you know, that came out, or prototypes, no, not prototypes, but like, you know, beta versions of Dali, and you had to, like, go and...

Auren Hoffman (30:17.9) Yeah, I agree. Yeah.

Russ d'Sa (30:31.236) If you were really good at prompt engineering, like understanding, I got to put like Nikon something in here and like some random photo, Ansel Adams this, it's like, you know, these random words and then all of a sudden it spits out something beautiful. That's maybe kind of how I conceptualize or relate like the current coding agent, like incantations for prompts that you have to come up with.

Auren Hoffman (30:36.684) Yeah, yeah, totally. Right. Exactly.

Yeah.

Auren Hoffman (30:57.23) also just like, it's just, why is it all like the whole tech space thing is I should be able to like circle your face and it's a, I make this nose a little bit differently or change this thing or change, you know, and then, it shouldn't like redo the whole thing. And ideally, like if I want the eye, just the eye color to pop, it should just make the eye color pop or something. Yeah.

Russ d'Sa (31:07.472) Yep. Totally. Yeah.

Russ d'Sa (31:17.81) 100%. And that's where we're going to go. mean, it is going to converge on Jarvis from Iron Man. But we're just not there yet. This is the early innings for that stuff, too.

Auren Hoffman (31:28.334) Where else do you think these like, assistants are going to evolve over the next few years?

Russ d'Sa (31:34.874) I think that there's going to be two kind of paths that you're going to see emerge. Already are emerging to a degree, but again, it's still early days for either one. I think on one side, you're going to have the co-pilot. And the co-pilot is going to be there with you to help with creative work. So I'm envisioning something like I'm working on a

an image or a painting or something like that or a digital painting, I guess. Maybe we'll have humanoids that also can do physical painting but with brushes. you're working on some piece of art or it might be a song, it might be a movie, et cetera. And you are really like the director of this and the copilot is helping you with some of the mundane tasks or the mechanical tasks that are

kind of embedded or part of that creative work that you're doing. And so it's like you're partnering crime through that kind of experience or creative process. And we're going to see one kind of whole path of applications that are infused with AI that is your co-pilot, not so dissimilar from the early versions you see now with a lovable or a cursor or a windsurf.

And then on the other side, and this one's a bit earlier than these copilot examples, is the autopilot examples. And I think the autopilot version of this is where it's less oriented around creative work and more oriented around kind of like mundane or rote or automatable tasks that you really don't even want to do anyway. I think human beings love to be creative. I think that that's like just...

Auren Hoffman (33:25.143) Yeah.

Russ d'Sa (33:26.61) embedded within us as a species, I think there's other parts of work and labor that we'd all rather not do, to be honest.

Auren Hoffman (33:36.994) Yeah, totally. It's like even like copying, go to this thing, copy the numbers from here, put it in a spreadsheet, send this report to this person or something like that. Like it's not always an API for all those things. Yeah.

Russ d'Sa (33:42.48) Yeah.

Russ d'Sa (33:46.898) Yeah, does anybody actually want to do those kinds of things? And so I think in the autopilot use case, there's going to be a bunch of applications or agents, I think is like the new word that we're using to replace the word application. Or the agent is the application. These agents are going to go, and you can effectively tell them to do stuff for you. You can tell them what task you want them to go and tackle. And they'll go off, and they'll do it. And they might ping you back.

and ask you for clarifications, or they might end up being blocked on something and they need your advice or guidance on how to proceed, or they might need you to type in a password or unlock something for them or give them privileges over a resource that they don't have access to. And that's going to be the other kind of way that you interact with this technology.

Auren Hoffman (34:37.454) It's really hard to do like figure that out though. Like it's like even like a very small example of my, my doctor, like, like I have my blood test and then he like takes the, and he doesn't do this, but someone's office, like literally takes each thing from the blood test and puts it in some sort of like essentially like a Google sheet to track it over time and track these metrics over time. So I'm going to have to go in and literally

Russ d'Sa (34:42.074) It is.

Russ d'Sa (34:46.993) Yep.

Russ d'Sa (34:57.724) wow. Yeah, yep.

Auren Hoffman (35:01.966) Go do it and have to know. And sometimes the blood test changes what it is. You have to know what's what and what is, and then it's got to be HIPAA compliant. So you can't like send it to open AI. would have to be a local model that would live on your local things and stuff. it's just like, it's, it sounds like it should be easy, but he's tried like eight different services. None of them work yet. And so, and he's worried about getting it wrong. He doesn't want to scare the, you know, his client if it's not right.

Russ d'Sa (35:16.624) Yep. Yep.

Russ d'Sa (35:32.114) 100%, there's a lot of challenges on the security compliance side and red tape that have to be either adjusted for or dealt with in some way. I don't want to say circumvented because that's not really a great word or doesn't connote the right kind of approach to these kinds of things. But they have to be taken into account. Some of the

the constraints within certain problem domains, like finance is one example, health care is another great example. But I think once that is navigated, right, like I think over time, once we kind of figure out how to integrate AI into these workflows, like to use the example with your doctor taking these blood tests and looking at them over time, what it unlocks, I think, is going to be very profound in terms of value. So very kind of,

directly tied to the blood test example you used. I had a blood test or a series of blood tests for my kidneys because I was working out a lot, eating way too much protein, because that's the internet told me to do. And I had my blood tested and they said, yeah, you have like too much protein in your blood and your kidneys aren't filtering out and we think your kidneys are like a 70 year old's kidneys.

Auren Hoffman (36:56.561) my gosh.

Russ d'Sa (36:57.522) A bunch of tests proved that I was fine and there was nothing wrong. Thankfully, thankfully, knock on wood. But what's interesting is that if you looked back at my blood tests from when I was 25, 15 or so years ago, I was already showing a naturally higher sign of protein in the blood before I was doing any of this crazy diet or working out.

Auren Hoffman (37:01.023) Yeah

Auren Hoffman (37:19.863) Got it. So if like somehow they just realized that they could have circumvented all that.

Russ d'Sa (37:24.594) Exactly, or they would have been able to contextualize what I was showing at 25 with what I'm showing now at 40. But what they weren't doing, what the doctor wasn't doing was because the doctor is a human and they just can't scale to the number of patients they have, the only time it got on their radar that this number was creeping up or that it was already high or too high was when it went from being like green into the yellow or whatever color they associate zone.

Auren Hoffman (37:28.065) Yeah.

Auren Hoffman (37:49.728) Yeah, yeah, totally. Exactly. Yeah.

Russ d'Sa (37:51.76) And they're like, wait, this is too high, but it turns out that that entire range is supposed to span ages like, you know, zero through 70 or 80. And they're not contextualizing it to someone who's 30 or someone who's 41 or whatever. Right. And so it's, I think when you, when you can add AI or computers to this mix and they have the context and they're compliant and all of that, all of a sudden you can unlock more value for patients because you can scale software better than you can scale.

Auren Hoffman (38:03.917) Right.

Russ d'Sa (38:21.65) It's not like to say a doctor isn't valuable. It's just to say that they have limited time, right? Whereas a computer doesn't eat, computer doesn't sleep, runs 24 hours and computes way faster than a human brain does. yeah.

Auren Hoffman (38:34.126) Now on the founder front, one of the things you've been pretty vocal about is like having founders not raise until they have product market fit. Um, unpack that a little bit for us.

Russ d'Sa (38:42.759) Yeah.

So I think that there are founders that have raised when they don't have product market fit, and they've succeeded. So I don't think that this is a hard and fast rule that you always must follow or anything like that. I think from my own experience, I tend to think of raising money as a responsibility. My

Auren Hoffman (38:52.323) Yep.

Russ d'Sa (39:10.418) There's pension funds that invest in VC funds, And our LP is there. And my partner is a special needs preschool teacher and she has her money in pension. And so I think of raising money as a serious responsibility. And I think of it as a job to bring a return on capital to the folks that are taking a bet on me. And so I think what's tricky for me anyway,

Auren Hoffman (39:14.04) Yeah.

Russ d'Sa (39:39.258) And the way I think about raising capital and the responsibility associated with it is that if you don't have product market fit, then you don't really have an indication or a confidence that you can deliver a return. And so when you raise money before you have that confidence, it puts a lot of stress on the founder, or at least it puts a lot of stress on me, in my experience, where I suddenly feel like I am

kind of under the gun. if I had the capability or capacity to be able to not raise, to experiment a bit, to find something that has at least some indication of product market fit, maybe not full blown product market fit, but some indication or signal, then I have the confidence level where I can say, OK, I'm going to start to dump gasoline on this fire or on this small spark.

And I think from a stress perspective, I think like you're just in a much better mental state as a founder. And so in the past, I've done companies where I've raised money, sometimes quite a bit of money before I had really any idea what the company was going to do. I just wanted to start the company to be a founder, to say I was a founder of something. And Livegate was the first time I didn't do that. And it's worked out a lot better for me.

Auren Hoffman (40:56.588) Yeah, we've been there.

Russ d'Sa (41:03.122) both in terms of the company's success or progress and in terms of my own mental state of running the company.

Auren Hoffman (41:12.194) What other advice would you give to founders?

Russ d'Sa (41:16.102) I would say the main piece of advice that I would give to founders is

Russ d'Sa (41:25.66) Keep in mind that we don't have a lot of data points. A lot of the advice that you get as a founder is coming from a good place. It's well-intentioned. But it's two things. One, it's folks that are trying to pattern match to what we've seen before. And we just don't have that many data points about what it takes to make a successful company. And then there's also a lot of other variables at play, like the market at large and the space that you're playing in.

competitive landscape and your own team and how fast or quickly you can iterate and react to what's happening in the macro that's out of your control. There's a lot of variables. And we try to come up with some kind of mathematical equation that maps x to y, right? Like x being the inputs or many x's, x1, x2, x3, et cetera, being the inputs, and then y being the output as to what you should do next. And I just don't think that we have that much data.

to say with any precision about what you should exactly do. So I would take kind of the advice that you get that is sort of pattern matchy with a grain of salt. The other thing that I think is really important for a founder to bear in mind is that nobody understands your product, your space, what you're doing better than you do. When you go into a VC pitch and you feel super nervous because they're your

these folks are investing large amounts of money and they seem like they know a lot, they actually don't know as much as you do about what you're working on, right? So they're there to learn and get up to speed and exactly, if they do, there's a problem. I've also had founders come to me and say, well, I'm thinking about bringing on this person as an advisor to help kind of guide the product or serve as like a product advisor, head of product. I'm like, if you need to do that, I mean,

Auren Hoffman (42:58.21) Yep. In fact, they do, there's a problem.

Russ d'Sa (43:17.914) I think you're in trouble. I think you need to intimately, you have to be the PM. You are the number one visionary PM for what this product is going to be and where it's going. And you cannot farm that out or outsource it. so I think that that's generally is like, think about advice in that way. Don't put too much stock in it. And then the corollary to this is think first principles. I know it gets said a lot, but

Auren Hoffman (43:18.978) Yep. Yeah. You have to be the PM. Yeah. Yeah.

Russ d'Sa (43:47.812) like nobody knows what the right choice is better than you will. Like incorporate learnings, incorporate that information, but make a first principles decision. Reason about it, not based on what other people do. Reason about it based on the information you have and make the best decision given what's put in front of you. And then kind of react to what you learn as a result, right? And kind of be comfortable. I think it's good to build the muscle of being comfortable with

making mistakes, acknowledging that you did the best you could with the information you had, and then doing something differently now that you have more information. So I think that that's another piece of advice I'd give.

Auren Hoffman (44:31.914) On your ex bio, you describe yourself as a late bloomer. What does that mean? And what, is that?

Russ d'Sa (44:41.126) So I think.

Auren Hoffman (44:44.876) Like kind of didn't excel in kind of high school college type of thing or, or what do you mean by Lee Boomer?

Russ d'Sa (44:44.912) I think what's interesting.

Russ d'Sa (44:49.938) Well, I think it's an extra. There's an interesting kind of juxtaposition in my career, at least that I kind of when I look back at my career, the way that I view it is I've been pretty early to stuff like early to some ideas. I was in the fifth batch of YC. I was the second front end, 75th employee at Twitter. I joined 23andMe.

I interviewed there and got a job offer, but turned them down and do YC when there were five people. And then I went back after my YC company failed and, and I went and I joined them as like the 30th or 35th employee. And so I've been like early to things, right. And I have more examples, but you know, I won't, I won't belabor the point. But then when I have like, I think I've been early to things that have been done by other people. So I think like I've been good at kind of like looking and seeing what is out there and kind of like selecting into things that I think are.

are going to become big movements or things, big parts of the future. But then when it comes to starting my own company, I think I've been pretty immature. And I think that it took me five tries to really learn what responsibilities and thinking thought process is required to be a good founder, to learn some of the stuff that I just told you, like around

like first principles thinking and not putting too much stock in what other people tell you or the direction other people tell you you should go with your product. Thinking strategically about go to market, like how do you prioritize different product initiatives properly and which ones are going to be the things that in the near term you have to kind of weigh in on versus like waiting for other parts that might pan out in the long term and kind of pushing those off despite them being these shiny objects.

knowing when to grind versus when to go for the greenfield opportunities. And so I think that it took me many tries to learn that. And I don't think all of it was just from doing startups. think some of it was due to just getting older and learning more things adjacent and going through life a bit more. But that's why I say, in my ex bio, I talk about early, early, early, but then

Auren Hoffman (47:13.895) huh.

Russ d'Sa (47:14.212) As a founder, think a late bloomer for sure.

Auren Hoffman (47:17.514) You mentioned 23andMe, like why wasn't that company more successful? what, it had so many good things going for it. It had the mind share and the market share. was early. So many people I know were customers, millions and millions, maybe tens of millions of customers and everything. It has all these ways of learning from one another. It had a...

Russ d'Sa (47:26.15) Yeah.

Russ d'Sa (47:33.788) Yep.

Russ d'Sa (47:40.036) I know. Yeah.

Auren Hoffman (47:45.886) founder that has billions of dollars personal net worth and so she can self like, like all these things you think would have made it successful. Like why was it not successful?

Russ d'Sa (47:57.682) I think that there's a, you as you've done startups, of course, in the past, and I think there's a lot of different things that can contribute to a startup not working out, right? There's like a thousand different ways you can get killed as a founder or your company can get kind of killed. But I think fundamentally, there is one thing that in my opinion kind of held 23andMe back.

from being successful. This is something that I kind of said to them or folks at work there, including Anne. She's wonderful, by the way. But yeah, including Anne. Back in the day, even after I'd left, and that was,

Fundamentally, science moves slower than technology. so 23andMe was really early in the game doing this kind of personalized genomics and personalized medicine as a goal. And I think it's a wonderful goal. I think it's going to be a thing that happens. within the next 10 years, we're going to have a lot of the things that I think Anne and the team there envisioned in the future, even back when I was working there.

And so I think timing is really important. And what's tricky is that like 23andMe came in, know, they're the first player and, um, and they have all, you you said like a lot of people, you know, like have all done 23andMe and they've kind of captured the market share. And we were Oprah's, you know, favorite product back in, in 2008. And I remember packing boxes because yeah, like we were, we were times, uh, you know, one of times inventions of the year, I think times invention of the year in 2008 or 2009 and

Auren Hoffman (49:35.981) wow, that long guy, my gosh.

Auren Hoffman (49:43.235) Huh.

Russ d'Sa (49:44.89) I remember just like being inundated with like these purchases of the spit kits. And so we were having to go and work overtime packing boxes like the whole team. it was great energy and it was a great, yeah, great, great moment. But I think that when it comes down to it, you you get your DNA sequenced and you take a look at like what your gene, what we can say about your genetics today at that time.

Auren Hoffman (49:53.608) so fun. Yeah.

Russ d'Sa (50:10.33) it just wasn't that high impact except for a few individuals. So like BRCA1 with breast cancer, APOE with Alzheimer's, LRK2 with Parkinson's. For a very small set of those markers, genetic markers, we can say something very impactful to your life. But for the vast majority of people, we can't really say anything.

Auren Hoffman (50:15.213) Yeah.

Auren Hoffman (50:30.766) And even this is like, okay, APOE like, what are you going to do differently? It's just like, okay, you maybe get some exercise more and I don't know, exercise and that's it. there's no other thing.

Russ d'Sa (50:35.234) That's true too, yeah. Exactly. Yeah, exactly. Right, we don't really know how to combat those kind of things. you know, like some of the mitigations are kind of like Brian Johnson-esque. It's like, you gotta do this thing for like 40 years and then maybe you award off like the Alzheimer's onset and maybe you won't. it's...

Auren Hoffman (50:58.198) Right, right.

Russ d'Sa (51:02.0) It's one of those things that gets, it's really hard to commit to something like that without having like a degree of precision or confidence that it's really gonna work out if you do make that investment. And so I think that for 23andMe, the big challenge was we have the technology, we can sequence the DNA, we can do it affordably, but what we can say about what your DNA says and...

the kinds of actionable advice we can give you is still very limited. We don't know enough. Maybe we need AI to go and analyze a full genome.

Auren Hoffman (51:37.73) But like, why, like if we start marrying that with outcomes and marrying that with, people who did these things that got this or people who got this drug differently and stuff like that, it just, never got like the databases never got merged. And obviously you could do it in, you'd have to do it anonymous way where you protect privacy, but it just never happened. So like, I feel like we could get there. Like we can get all this not like it has.

Russ d'Sa (52:01.65) It didn't.

Auren Hoffman (52:05.55) We have the data now, like now we just merge it in and it's like, boom. It's like, I would opt into that in a second. Like I think everybody who did that would opt in.

Russ d'Sa (52:14.362) You know, and I think that they would, I think the things that make the, there are some complicated dynamics at play. like, you know, they had the FDA thing that came down on them. They were the first company really that had to deal with how do we work with the FDA and like bring this genomic information to light. you know, how do we avoid being a diagnostic tool or interpreted by the FDA as a diagnostic tool? And so,

Auren Hoffman (52:34.285) Yep.

Russ d'Sa (52:41.948) There was a lot of red tape that 23andMe kind of trailblazed. And so that's one thing that slows things down, this integration of systems. And so how to work with them and like there's, you know, there's some fear there too. It's like, how are they gonna interpret, you know, if you're a bit cavalier about it or a bit more aggressive, are they gonna interpret it in such a way that they shut everybody down? And then it's like, this is why we can't have nice things scenario. And then no genomic companies can do anything. And so I think that that's one problem.

Auren Hoffman (52:47.244) That's a good point. Yeah. Okay.

Auren Hoffman (53:05.238) Yeah.

Russ d'Sa (53:10.096) The second thing is that, and this is kind what I was getting at with, I think, the fundamental part that held back 23andMe, I'll take an example of, to use an example of what I think is a good situation or a good setup is Elon Musk is kind of like a master of this particular sort of phasing or staging of his companies. So with Tesla, the mission is to electrify the world and reduce our

or eliminate our reliance on fossil fuels. And the way he's going to get there is not by doing that directly. It's first, he's going to make really, really nice cars that people want to buy. then the money that you make from that is going to fuel, almost literally, like pun unintended, is going to fuel your foray into robotics and into power walls and into solar panels.

Auren Hoffman (53:54.38) Yeah, expensive cars too. starts and then gets cheaper.

Auren Hoffman (54:00.814) Yeah.

Russ d'Sa (54:09.324) And then those things are going to cost money and you're going to make more money from that and get to the next stage. And I think that that's what he's done with SpaceX too. It's not like we're going to colonize Mars and become a multi-planetary species. Well, we're not going to do that straight away. We're going to take other people's payloads into space and sell satellite internet and maybe do civilian space travel and maybe replace airlines for traveling around the Earth.

there are all these kinds of things that kind of help build the company up over time, even from a revenue perspective, such that you reduce your reliance on having to raise capital, external capital, and become more self-sufficient. And I think that, like, and you can see OpenAI is moving in this direction too, and towards their path to building AGI. And this is something that 23andMe really just didn't have fundamentally is, you I think that they thought...

And this is my take from the outside. I've been out of that company for, I worked there from 2008 to 2009, right? So like almost a couple of years. And so I've been out of that for a long time and yeah, 15 plus years. But my take from the outside is that selling kits was, I don't know if it was also a subscription. I think they moved to a subscription model, but there really wasn't this kind of like organic.

Auren Hoffman (55:15.534) 15 plus years here.

Russ d'Sa (55:31.108) strategic path of how to layer in value over time as the science progresses. one thing that I remember talking to them about was like, Hey, you guys should work on like your own, not saying this is a good idea because hardware is hard, but like your own fitness band or your own health based product that you can constantly monitor and layer it into your genetic information. And, know, kind of like the direction Apple watch is starting to take though. I would never want to compete with Apple in retrospect, but you know,

Auren Hoffman (55:47.436) Right, right, right. Yeah. Yeah.

Russ d'Sa (55:57.202) there just wasn't like a staged approach to building this business over time on your way to this ultimate mission of personalized medicine for all. And so that to me is a fundamental kind of failure.

Auren Hoffman (56:09.258) Interesting. Now you before we started talking, you had mentioned you had like some interesting like a coincidence or something like that. I'd love to hear it.

Russ d'Sa (56:18.386) Oh, yeah, it's not a big deal. But what's so funny on the on my journey, right? You asked me about being a late bloomer and stuff and how I've kind of seen trends and like interesting things happening out there and then chosen to try to work at those companies or, you know, provide my my skills to defer in furtherance of somebody else's mission. One of the first companies, if not the first company that I interviewed at

When I was a bachelor, when I was doing my BS in computer science at Davis was this company called Rapleaf. And yeah, that was one of my first interviews, if not my first one, right out of college. And I don't think that we met. I think that I interviewed with someone there, but rightfully so. I was kind of being a jack of all trades and a master of none.

Auren Hoffman (56:56.369) my gosh, really? Okay.

Auren Hoffman (57:15.854) Ha

Russ d'Sa (57:16.412) you know, part product, part engineer, not good at any one of those two things. And so I don't, I don't blame Rappley for not, for not giving me a job.

Auren Hoffman (57:21.678) I actually I'm checking my, okay, this is a live, by the way, I don't want to, but, um, I have a, a, a, I had a call with you or at least, at least I have a call here. I don't know if it was with you or not. Um, in February 13th, 2007, which is the super early days of it. Okay. Yeah. And it's a four or eight.

Russ d'Sa (57:34.118) No way.

Russ d'Sa (57:44.326) That was right before I YC, yeah.

Auren Hoffman (57:48.642) phone number. don't want to say the phone number because it might still be your phone number. Okay, great. Yeah. So I'm just looking at it here. that was that that's actually and then and then later on, I see we had a we had like an overview.

Russ d'Sa (57:50.84) Yeah. I think my mom has that phone number now. We swap numbers. She had an unlimited data plan. Yeah.

Auren Hoffman (58:13.622) Yeah. So that's right. So that's crazy. my God. So you're in my ear, my old school emails going back. Yeah.

Russ d'Sa (58:19.468) I am. thought it was so wild that I was coming on the show and I'm like, my gosh, this is like, when I...

Auren Hoffman (58:24.438) Yeah. yeah. So it was a recruiter, a recruiter forward us to you or something. Yeah.

Russ d'Sa (58:32.24) It was a recruiter or like, did I apply off like a Craigslist ad? I can't remember, but it was either a recruiter or a Craigslist ad that I had applied to.

Auren Hoffman (58:43.694) Yeah. It looks like a recruiter forwarded to us. And then somehow I think you and I did talk. looks like I don't have the notes. I usually have my notes, but I don't have the notes, but it looks like we did talk back in the day, 2007. Uh, that's awesome. That's a great, that's a great, this is a great coincidence. All right. Last question. We ask all of our guests, what conventional wisdom or advice do you think is generally bad advice? And you're not allowed to say follow your passion because, um, I feel like that's like,

Russ d'Sa (58:51.654) Wow.

Russ d'Sa (58:58.886) Crazy, Yeah. What's up?

Auren Hoffman (59:11.054) 90 % of what people say when I ask them this question.

Russ d'Sa (59:15.154) I think, well, maybe I sort of answered it though in a different way earlier, but the conventional advice that I think is bad advice is...

Russ d'Sa (59:35.834) I guess it depends on how you interpret advice. Don't do what people tell you to just because they tell you to. I think that it's like first principles thinking. Every single person that you talk to and ask for help, well, I think most people, if you have friends that are good friends, when you ask them for help, they're well-intentioned and they want to help you. But again, everybody's story is different, including yours. And so advice is not going to directly apply. It's like if you go into, you know,

an exam in school, like for computer science or for math. And there's going to be homework problems that are similar to the questions that are on the test, but the numbers are changed or the situation is tweaked. And you have to incorporate your learnings from those problems, but also understand how to adapt your answer or the way you go and solve that problem to.

this unique situation that's put in front of you, right? Like these tweaks that have happened to the problem. And that's kind of like a, not a metaphor, but an analogy for how you kind of take other people's advice is to like, take it for what it's worth. It's learnings from a different scenario, maybe an adjacent one, or maybe a similar one. think about, incorporate that into your own thinking. Like that's, don't take anybody's advice kind of wholesale. Derive your own answers.

Auren Hoffman (01:01:01.39) That makes a lot of sense. It's interesting because like people, one of things that most people are really good people and they're trying to help you. And so you meet these people, they're evangelical about a very specific thing, meditation or protein or some sort of diet or workout routine or something. it really is because it worked for them, just like religion.

Russ d'Sa (01:01:10.918) Yeah.

Russ d'Sa (01:01:18.13) Mm-hmm.

Auren Hoffman (01:01:27.726) Like it worked for them. So they want a date. Well, it worked for me. So therefore I'm recommending it to you. Um, or this type of interview way of interviewing people or this type of way of running a business, whatever it is, like I'm evangelical about it because it worked for me, but just cause it works for me, it doesn't mean it's going to work for you or your company. Cause we're all in different situations. As you mentioned, we have different DNAs, our company's different DNAs are. So just like that meditation might not work for you just because it worked for me.

Russ d'Sa (01:01:54.898) 100 % the world is way too like life is way too high dimensional to to be able to to really say that like somebody's exact pattern is gonna apply to you and you're gonna you know get the same results in the end I don't know how much time you spent in like on reddit with like people like that do exercise but one thing you'll notice in a lot of

threads where someone that knows what they're doing when it comes to exercise or weightlifting or working out, one thing you'll notice is that in a lot of threads with experts, true experts, is they'll say, why MMV? Or this is what worked for me. It doesn't mean it'll work for you. And then they'll list out what their routine is. And I think that that applies to really all advice.

Auren Hoffman (01:02:36.727) Yeah.

got it.

Russ d'Sa (01:02:49.126) Take it for what it is, right? It's an anecdote. And figure out how elements of that anecdote apply to you and which ones don't apply. And then make the best decision for yourself and your unique situation. That's what I think.

Auren Hoffman (01:03:04.332) That's really, I love this. All right. Thank you, Russ, for joining us. Roll the dice. I follow you at. d'Sa d'Sa, which is an amazing X handle, which you probably only can get if you're early there. that was like, that's an incredible three letter X handle. I definitely encourage our listeners to engage with you there. This has been a ton of fun and really interesting.

Russ d'Sa (01:03:15.964) That's right.

Russ d'Sa (01:03:25.754) Yeah, it's been wonderful to have the conversation. And I really appreciate you giving me the opportunity to come on here and asking the questions. And also just really nice that you looked up that old email. And we're able to connect almost 20 years later in this capacity. Yeah.

Auren Hoffman (01:03:43.96) Crazy, yeah.

All right.