Unaligned #2: Always Listening and Watching AI

Are you always forgetting things? There is way too much to remember nowadays. I can't even keep track myself. That's the problem that Rewind CEO Dan Siroker is solving. Backed by Andreessen Horowitz, Rewind is filling the cracks of the human memory. It securely and privately records everything that goes through your computer, everything you've seen and done, and leaves those recordings for easy access later. This means that you can instantly find out what you talked about on that Zoom meeting last Wednesday or get a list of takeaways. And to do items from has the potential to transform the way we record information. There's a future out there where we no longer have to take notes and all the information from our lives is at our fingertips. Coming up, Dan will tell you how that will happen and will give details on a new kind of AI powered pendant that has already been purchased by thousands. It's leading us to a new kind of digital assistant. That's very interesting. So I'm Robert Scovel, and today we're going to have maybe a little controversy on the show because Dan Siroker of Rewind AI has built a system that's always listening to me and always watching what I'm doing. So welcome, Dan. You know, you're building one of these new products, these new categories of AIs that's always listening, always helping, always doing notes. Right. Tell us a little bit about what, what you're trying to do with Rewind and why we shouldn't be scared of it. Because it's scary as hell. It's like, oh, gee, this thing's listening to me all the time and watching me.

Yeah, well, thanks for having me. So I'm happy to talk about it. So Rewind came out of the experience I had going deaf in my 20s and then trying a hearing aid. And that experience felt magical because it felt like gaining a superpower. You know, that's what it feels like to lose a sense and gain it back again. And from that, I think I learned a secret, which is that we live our lives limited by our biology, and we can't imagine what it would feel like to have what feels like superhuman capabilities without actually feeling it. So, you know, when we started, the company really was rooted around this vision of can we do the same thing a hearing aid does for hearing, but for memory instead? People typically have really poor memories. 90% of memories are forgotten after a week. Typically, your memory peaks when you're 20 years old and then every year thereafter gets worse. So that was kind of the thesis for the company can we build something that ideally is, you know, as magical and ubiquitous as a hearing aid is for hearing or glasses or for vision, but for memory instead. And so you're right to bring up the privacy questions. And you know, what's interesting is you look at the history of any new technology, it almost always follows the same pattern of sort of awe, shock, cynicism and, and, and ultimately on its path to broad based normalization, it goes through this kind of evolution. And you know, a good example of this is the telephone. It may surprise you and some of your listeners that the very first and most common objections of the telephone or when it first came onto the scene, was privacy. It's kind of hard to imagine today because the telephone is so normalized that like, what do you mean, privacy? But the reason why the telephone was such a lightning rod around privacy was back when it was first introduced, you had to go to the general store to make a phone call. Even when it came to your home, switchboard operators could eavesdrop on your conversation. And so that is the moment that we are in right now with personalized AI. The benefits and the power are there, but the technology and the capabilities to overcome the eavesdropping. The proverbial eavesdropping switchboard operator aren't there yet. And so it's not that it's not the case that people should view our product or this kind of personalized AI as a choice between convenience and privacy. It's actually quite the opposite. Is that the solution? The market leader in personalized AI will be that company that was able to overcome those privacy obstacles. And we've got lots of ideas. I'm happy to talk about what that specifically looks like. But I'm not here to say that privacy is dead. I think it's quite the opposite. I think the reason personalized AI hasn't been real yet is because no company has really built a privacy first product to solve.

Yeah, let's talk about what does Rewind do? If I'm on my Mac or Windows box, what is it doing and what are you going to take it and do with it in the future? But let's just start with the basics of what it does, because we're talking about how it's listening, it's watching, but what is it actually doing on my screen?

Yeah, so Rewind is a personalized AI powered by everything you've seen, said or heard. What that means is it's, you know, in many ways it shows up in the different devices you use. We started on the Mac, when you install it, it starts capturing your screen, does OCR on what you see, so it turns that into text. It does ASR, automated speech recognition on what you say and what you hear and stores that. And that gets stored all locally on your machine. We as a company don't have access to it. You know, if we were subpoenaed as a company, we couldn't give anything that's meaningfully useful because it's all stored locally and encrypted on your Mac. That was only possible, by the way, because Apple Silicon became good enough to make that imperceptible. So all of that happens in the background imperceptibly on your Mac. Very similarly, that's true on iPhone. If you download our iPhone app, we capture everything you see in Safari. You can do transcription, we capture your screenshots, we make that all searchable, all locally on your phone. So that's how we've started. We started with the phone, we started with Mac the phone. We're working on Windows. In fact, we're launching the end of the year on Intel's new processor, which is going to be very exciting. And more recently, we got a lot of excitement around the pendant, which is really extending that idea of capturing everything you see, staying here to the real world. We do a lot on our devices, but there's so much more to life than what happens on your Mac or iPhone or Windows. And that's what the pendant offers again, entirely privacy first. All stored locally. It's in no way for us as a company or even your employer to snoop audits data just for you, to give you a personalized AI. And that's really the value is now that you have this data, you can ask it to help you. You can ask it to write you an email to somebody you might know. And it uses the context of your past to write that draft for you or summarize a meeting or, you know, hundreds of other things that AI and large language models are so perfectly well suited for, but now tailored to your data.

Yeah, this is really. There's a lot there. Let's dive into the pendant because that, that clearly you have a few competitors there from Humane, which is company started by two people who left Apple and are building a little pin, right. And you're building like a pendant that hangs around your neck. What's in that pendant? It has microphone, a little speaker and a computer inside. What's going on inside there?

Yeah, it's actually not that sophisticated. So, you know, a wearable microphone that syncs with Your phone. So almost all of the heavy lifting happens on your phone. Um, you know, I think probably humane, I think has a different approach. I think they're putting a lot of the, the, the heavy lifting in the device itself. We don't, they have little, they have.

A little display that spits laser light out in front of you and puts things on the table in front of you or on your hand or stuff like that. Right? Yeah.

And, and I'll just share, you know, I, I have total respect for that team. I'm sure they're working really hard. I, when I first saw that, it reminded me a lot of my early career at Google. Google has this one challenge in which it views the sophistication of the technology as more important as the problem it's trying to solve. And if you look at the history of Google products, from Google Wave to Google Buzz to Google Glass, it all sort of rooted in this focus on the sophistication of the technology, not does it actually solve a problem. So I'm very excited to see what problem does this humane AI pin solve. I haven't really heard that yet. I can tell you for us, what we're focused on is solving the problem of giving you perfect memory. That's the core problem that we started the company with. That's what we think AI in a privacy first approach can solve uniquely. Well, we don't have a camera, we don't have any of the bells and whistles, we don't have a laser. It's as simple as a way to help you capture more of the things that if you had perfect memory, you'd be able to recall. But because we don't, we use technology to augment and give us that.

So let's say I'm wearing one of these and I have it all set up and I meet you at a conference and I start talking to you for a bit about AI or something like that. What is it doing and what, what, what, what would it capture out of that conversation and what would it put on my phone? And how usable is that? What, what are your plans there?

Yes, so the most basic thing it will capture is what you say. So as the user of the product, you know, we want to make sure that you can capture what you say and do that in a way that you understand what happens to the data, where it gets stored, how it gets used. That I think is a really straightforward thing. It's, it captures what you say, transcribes it stores it all locally. We don't have access to it your employer doesn't have access to it. Like I said, the bigger, broader question happens what happens when you talk to somebody else? This is where consent is super important. There are certain settings in which the presumption of consent doesn't exist. So you know, public settings is one of them. Conference speaker, you know, doesn't expect, unless they say at the beginning, this is off the record, you know, both legally and sort of ethically and morally and socially. I think people would understand if you're giving a big talk at a conference, you know, people are taking notes, making people recording it. And so that's sort of the other extreme. I think that's, you know, obviously something that people would be comfortable with. The part where you're having a one on one conversation where maybe it's more private. That's where we have lots of ideas on how to, I would say, you know, walk the tightrope without stepping on landmines around the privacy of the people around you. You know, I think the naive approach that a lot of companies take is a, is a red button or a blaring sign. You know, I think that lacks creativity and sort of lacks sort of a real understanding of the problem to be solved. Because at the end of the day the problem to be solved is making sure people aren't recorded without their consent. We have several ideas on how to do that, I'm happy to elaborate, but that's the problem that we're focused on solving and I think that's what really differentiates us in this AI wearable category.

Tell me a little bit about the architecture of what's going on underneath the curtain. Do you have a large language model listening? Do you have a group of them? Tell me how this thing is working to figure out what I'm saying, how important it is, where to put it, all that stuff.

Yeah, so the basic way it works is sort of captures sound and that sound that's turned into a feed to a local instance on your phone of Whisper. Whisper is an open source model that OpenAI produced and it's fantastic at taking sound and turning it into words. That happens all locally so it doesn't go off to some like cloud transcription service. Happens all locally, transcribes what was said. And in addition to that, you know, microphones have gotten so good and even having multiple microphones help you here that you're able to sort of identify based off of several heuristics who is actually speaking. And this is the sort of the first area in which we are helping ensure people aren't recording without their consent. You know, one of the ideas we have is to be able to actually use voice fingerprinting and speaker diarization, which is basically the idea of, you know, separating out the data of what one person says with what somebody else says and use that as an indicator for what can be captured and what can't. So, for example, it's very possible that you can, you know, capture what you say and store it all locally and using, you know, voice fingerprinting, know it was you, and you give them consent and only capture the voice of others who've affirmatively, verbally opted in. That, I think, is a much better approach than sort of a red blinking light that maybe you miss. And so that's one of the other ideas that we think will really help normalize this, you know, this superpower.

Now you, you said you're using Whisper. This is the API from OpenAI, right. For doing voice to text. Right.

Well, when you say API, that kind of implies that it's in the cloud, it's all local. So that, you know, Whisper is available through, through an API from them, but you can also just download it and run it all locally. And that's what we do on the phone.

And it's.

And today, by the way, you can test it out today we have an iPhone app in the app store, the Rewind iPhone app. It runs it all entirely locally. You can turn your WI fi off. It works great to prove that, you know, it's not actually sending any data to the cloud.

Yeah, this is, this Whisper system is really insane for people who haven't used it. I've been using it on the ChatGPT app. Right. And you can be at a rock concert like you could be at Coachella in the middle of the Sahara tent and it will still hear you about two feet away from your mouth. Even though there's 120, 20 decibel music playing, it still understands what you're saying to the microphone, to the, to the phone. Right. It's, it's really good. I mean, I, Kai Fu Lee showed me NLP natural language processing system at Microsoft 15, you know, 18 years ago, and it wasn't that good. Right.

So, yeah, it's, it's interesting. You know, I actually studied computer science at Stanford, graduated 2005, and my honors thesis was on natural language processing. You know, this was 18 years ago. What's funny is like what was then called AI is now just programming. It's not even really thought of as AI, just as it comes better and more normalized so, you know, not to say that I had any vision into what the world looks like today, but it's certainly interesting to see how these things that were once, you know, thought of state art can get so much better through what large language models are.

One thing I've been doing with the ChatGPT app is if on Friday night, and tonight's Friday night, I'm out with my friends and we start having an interesting conversation on something. I say, hey, hold on, I'm going to record this for ChatGPT. And I just opened the app up, I clicked a little record button, which is in the prompt window on the iPhone app, right? And it starts recording and it listens. I just let it listen for a couple minutes. It doesn't need a whole lot, just a minute or two of that conversation and then it starts getting involved in the conversation. It's like, oh, looks like you guys are having an interesting conversation about, you know, xyz, right? And here's some more things to think about. Wow, is that, does that blow you away? I mean, this new capability of AI to listen to you, to listen to you have a conversation and yeah, it'll be interesting to see how you turn it on for, for this kind of use case. How do you get consent from the other people, you know, talking so that, you know, okay, we can record this and keep notes on it? Yeah.

Well, to answer your question, it absolutely blows me away. Every day I see something our product can do that I never. And I've seen the code. I understand. Like, I still, sometimes I'm baffled at the magic that large language models offer. And in particular, you know, we got more lucky than good. We have been at this since, you know, we started the company March 2020. We've built this product that captures what you see saying here on your Mac. Launched that almost a year ago, before even ChatGPT came along. And we just got lucky in that we've been capturing this amazing set of data in a really privacy sensitive way that pairs so beautifully with, with GPT4. It's like we've got this amazing, you know, bottle of wine and you get this steak dinner with it. You know, I'm both a vegetarian and I don't drink, but I assume people like that and that those things together are just amazing combination. You know, the reasoning of a large language model applied to your own personal data is what gives you that superpower of having a personalized AI for people.

Who aren't in the AI bubble and don't know what all these terms Are large language model is this new Mach that almost has a trillion little knobs on it. Right. It's an insane machine. Tell me a little bit about what a large language model is and why it's so important to the future of your company.

Yeah, I mean, a large language model at the most basic level is pretty simple. It basically is a system that tries to predict what the next token should be given a sequence of tokens. And the way it does this is basically by consuming or I guess you could say reading a large, large set of data of those patterns in the past. And it learns based off of the data it's consumed. Usually, you know, people say, you know, just cowers the web, some cynics will say, just uses Reddit data and you know, things content that's been produced in the world of real world reasoning, real world logic, real world text, and then becomes very, very good at predicting based off of just a few tokens, what the next set of tokens should be. And you can think of tokens as like a word or partial word. And so the way that that technology can be applied, you know, if you go to ChatGPT, how it works is you give it a question and then it predicts what the answer should be, as if it were, you know, trying to match what a human would answer. And just, just that concept alone is magical, that it can do things like, you know, as it got better and better trained on more and more data, at some level of data, it could all of a sudden do poetry and then it could all of a sudden do jokes, you know, and it can do that in a multi.

Jokes are a little, no, little weak.

I have to say. The jokes are pretty good. I mean, maybe it's because I have a, I'm a dad of three kids, so my, my dad joke bar is low. But I have seen some pretty amazing jokes and ability to understand jokes. That's the other part that I just. It blows my mind. You can now with multimodal, you can give it an image and ask it to explain why it's funny. And it understands it, it gets it. And that's something that I think is truly. I think when I saw this for the first time, two things happened. One, I realized how magical this is, but it also kind of, kind of knocked down a peg or two for me, this sort of like human knowledge superiority concept, like the idea that we're all that special because of what we can do with our brains. When you can see a machine do it and with such a basic capabilities kind of made us, made me realize maybe we're not as special as we think we are. So I don't know what your reaction is to that, but that certainly was one of mine.

You know, it's one thing to write a transcript, it's another thing to make that transcript smart and start going out to the, oh, oh, you're talking about buying a car. Let's go out to Kelly Blue Book and start getting information or start doing some other things, make that transcript smarter. What are your plans on making the transcript the notes that it takes on me, particularly with this pendant? How are you going to make my life better by doing this system?

Yeah, it's a great question. And it starts with recognizing that just collecting what you say in here is the foundation, it's sort of the data. But to take that data and turn it into information and ultimately into knowledge or even wisdom above that, it needs, it needs context. And that's why we think going beyond just what's in your devices and actually stitching together that data across all of your devices in the real world and the, in the, in the digital world really enables you to do, you know, things like understand people. You know, like you interact with people in lots of different ways. You interview them, you DM with them, you email them, you slack them. If you had a first world or sort of first order concept of that person and all of your interactions with that person, that's incredibly powerful. You know, an example of today you can run using Rewind is, you know, I, you know, an example I shared is, you know, I ask Rewind, write me an email to Sam Altman asking to catch up. And it can use all of the details of our past. The fact that we were introduced by Paul Graham In 2010, the fact that he was an investor in my last company, optimizing an investor in my current company, Rewind, all of those details are things that if you had a truly great representation of that person across every touchpoint you have, you can give you a truly great experience. And now it writes me a draft of something in my inbox. That's something where we assume like, oh, my job is to write emails. Well, you don't need to. I mean you can ultimately, if you can prompt an AI that has the data about your past to do it for you. That's a superpower that gives you so much more productivity. And that's just one of many examples that AIs can use if it had the access in a privacy sensitive way to the data of your past.

Can I talk to rewind and tell rewind, hey, can you put this on my, you know, notes or can you remind me of this, of this the next time I'm at a Home Depot or something like that? I need a screwdriver. So can you know, how crazy can this get? Because it, with the Chat GPT app I can talk to ChatGPT. I could tell that, hey, can you listen to this for, for the last two minutes and then write me a blog post about it? And it will, it's like, oh, that's crazy that you can talk to the engine, you know, while it's.

Yeah, there, there, there's some basic things it can do today. You know, things like drafting emails. One of the most magical features people love is at the end of every one of your meetings, like a Zoom meeting, you know, a little pop up comes up from rewind. It says, you know, summarize and draft email. So you click it in one click and you get an email in your box to all of the participants in that meeting, by the way, not just. Yeah, and it's literally set and forget it. You look brilliant for having written perfect notes. And so those are the kinds of workflow things that we think are really powerful. But it all starts with sort of a comprehensive understanding of the things you've seen, said or heard and then being able to actually use that understanding to do these kind of workflow tasks for you.

Yeah, it's pretty crazy. What's possible. I'm sure you're watching all the vision stuff, the multimodal AIs that are coming from Chat, from OpenAI and others. I've been playing with the Chat GPT camera app where you aim it at. You know, I aimed it at a Ferrari on the road yesterday and I said, oh, that's a Tesla Rosa and Top speed is 180 miles an hour or whatever it said. Are you thinking about how to use the camera as part of this system that's always listening and watching and doing things?

Not really. I mean, I think eventually maybe. I think the word watching is scary. And that's where people get into surveillance. That's where people. Where I think Google Glass got a lot of things wrong around privacy. Right now we're focused on the things you say, say and hear in the real world. And similarly on Zoom when it comes to what you see, that I think is a sort of different level of privacy. And I go back to the analogy I made around the telephone. You know, there is a set of technology that we need to build to get people to understand and trust more than anything that their privacy and the data that they're, that, you know, we could be collecting is safe. So we have straight away. That's why our pendant isn't the fancy, you know, camera. I'm sure you know, some. If you had unlimited resources, you put all the bells and whistles, we would much rather a very cheap, you know, today you can preorder it for $59. You know, it may, you know, maybe cheaper, maybe more expensive, but that's, you know, we much rather a product like that than something that's hundreds of dollars that has, you know, the Vision Pro has like eight cameras or something obscene or gene. And it's a beautiful piece of hardware. I've tried it, you know, it's a beautiful piece of hardware, but it's definitely, it's not something that we think will ultimately for us, our focus is give humans perfect memory. That's a hard enough problem, enough. We would rather solve it with the best and the minimal amount of hardware if necessary and ideally with software, frankly, and then let the software do the magic, not overcomplicate things by having a huge hardware contraption.

How is networking at conferences going to change because of, let's say, this pendant? At Davos, I met a guy, he had a badge on his name was Peter Pyatt, right. And it said he worked for the Bill and Melinda Gates Foundation. But I didn't know that he discovered the Ebola Viru. When I met him, if I had, we'd had a much different conversation, right? So one can it, can I talk to it and say, hey, I just met Peter Pike. Can you tell me a little bit about him? Right? Put, put it on my phone or something. I'd love that kind of interaction. And then it, you know, if we talk for a couple minutes and we're like, oh, I need to. Can I get your email? Can we figure out how to talk to the pendant to, you know, remember the guy's email or his Twitter x account or LinkedIn account, stuff like that?

Yeah, I think of all those things, the thing that feels most real is talking to yourself and taking action items. You know, right now I think it's a bit of a leap to have it be interacting in the conversation with you. I'd rather it be passive, imperceptible, passively capturing what you want. And you can say, hey, it's great to meet you person's name. I'll send you an email later today. And then it knows because it heard and if good analogy is imagine it was a perfect chief of staff that followed you around. Everywhere you go, you know, the President has this, and the President whispers in your ear, this is who this dignitaries and this leader. They keep track of action items for you. That's something that I think a pendant or even any kind of personalized AI should be able to do for you effortlessly. And then everything beyond that of being actively involved in the conversation, I think that would be great. But that's further along, you know, I think starting with kind of being that passive, you know, safety net, I think is a great place to start. And over time, as people more comfortable and they see the power, they can rely more and more of their brain outsourced kind of to the device and not actively having to sort of remember action items and, and that kind of thing.

I've gone through 3900 companies on x.com all in AI and I've grouped them up into different groups. I put, put you guys into this group called Holodeck, which is a group of companies that are doing things while I talk or eventually leading to a world where I can talk to an environment, say, hey, take me to the Taj Mahal and boom, it'll take me to Taj Mahal. Right? Let's say in five or ten years when we have glasses that can do that. A lot of these companies in this group thing are trying to build personalized A.I. i. E. They're trying to aim at a world where there'll be something that looks like a human or a robot that you talk to and talks back. Are you heading down that path or do you. Because it seems like you're, you're close to that, but you're really focused on utility without that human, you know, appearance kind of thing where it's talking back to you. It's a human thing talking back to me and having a conversation with me. Me. Right, yeah.

And we are absolutely that we are a privacy first personalized AI and you can, you know, right now you type to it, you can't talk to it. Obviously if you, if on our mobile app you use your microphone dictation, you can. But you know the feature we have today in a product called Ask Rewind, you can literally ask it any question about anything you've seen, said or heard and it gives you an answer. You can ask it to summarize everything you've read about a topic. You can ask it to do work for you to, to take all of the disparate signals and data you have around things you've seen, said or heard. And imagine that's the kind of data that you have at your fingertips, but using the reasoning of a large language model. So that's absolutely where we're focusing on.

Yeah. LLMs hallucinate once in a while or confabulate a wrong answer. I was in an Austin restaurant asking ChatGPT what to eat at the restaurant and it gave me six items, two of which had never been on the menu before. Right. It just made them up. And it makes them up quite confidently. It's hard to tell what's the difference between the made up one and the ones that were real. Right.

Yeah. So we have a great answer to that. We have a great solution, which is we give you sources, and those sources are moments from your past. You know, for example, I gave earlier the real example of write me an email to Sam Altman asking to catch up when it says, hey, you know, it's great to catch up. I remember finally our introduction from Paul Graham in 2010. It has a citation, a link I can click that goes exactly to that moment in my past, that email that was introduced. So that that is a key way to eliminate hallucinations, is to actually have source material that backs it up. I also think that our approach is so much less likely to hallucinate because what we're doing isn't training a model. We're not training a model on your past. We're actually using the power of the large language model, which is great at reasoning, giving it the context of your past and then giving a prompt to use that context using the reasoning of the large language model. So when it's in the context window of the prompt itself, that's what gets used. And you can refer directly to that in the response. So it massively improves the hallucination issue.

There's like 400 companies building developer tools, which is just also insane to me. I've never seen this kind of innovation for developers. How are you guys looking at the world of AI and what are you excited about other than rewind?

Yeah, I mean, I think developer tools are incredible. I think there is an era of engineer of which I'm probably one of them that was used to doing it yourself. And now some of the folks on my team, some of the younger folks, they are so much better than I ever was. By pairing with an AI, they're able to learn entirely new languages. Right now, for example, one of our top focuses is on building a Windows app. It's entirely new programming language, tools, technology frameworks, but with Copilot trained on a ton of C code. You can build things faster than if you were an expert in it yourself. So I think that the paradigm of building things is incredibly different. It's kind of unsettling how easy it is to build things today. And that I think also means technology will get better and better and grow exponentially because that last layer of technology is used to make the next layer. That's kind of the definition of an exponential. So I'm super excited about that. I also think AI has a tremendous opportunity to augment ourselves. You know, I think a lot of people fear AI replacing people, which it certainly will. I think half of jobs will disappear just simply because of AI. But if you think about the jobs in which you can use technology to make your to do better, I think that's a big, big opportunity as well. So I'm a big believer in AI augmenting human capabilities, not necessarily replacing them as the big opportunity for the future.

Yeah, it's pretty crazy. I have a special needs son who has a speech impediment. It's really hard to understand him. It's hard for me to understand him and I'm his D. Yeah. So if he, if he, if you meet him in the street and he asked, ask you something or needs your help, it's hard to understand him. How, how is the computer going to get to the place where it really understands the edge cases of human beings? People who have speech troubles or, and, or don't know the native language very well. Right. Might know Spanish really well, but not English.

Yeah. And you know, I think there's such an opportunity for computers not just to take something that you could think of as a disability or something that sort of makes it harder to be in the world. For me, that's hearing. But for others, there's other ways technology can augment that capability, but then so much more. I think being able to actually have a conversation with anyone in the world in any language, and have it simultaneously translated as if you're speaking that native language, that's not decades away, I mean, that's months away before that becomes absolutely normalized. Today you can do that kind of with your phone, but with wearables, I think that becomes even more powerful. And I think technology has that promise. I mean, that's if in some ways that is what technology is, it's a lever, it's a way to do a thing. But better because you use technology to do what it's good at and you do what you're good at. And so I'm incredibly optimistic for the opportunity it has for, you know, for your son, for the world more broadly. And so I'm not one of these doomsday AI naysayers. I'm a optimist. I think AI is going to make everyone's lives better. And we're all going to look back at today and I think, be surprised. Why were we so cynical? The same way people were cynical of the telephone, you know, I think we'll look back the same ways at AI today.

Tell me a little bit, first of all, what's your business model? Tell me a little bit about the fundamentals of the company you're building. What's the business model? How was it funded as well?

Yeah, so the business model is very simple. It's a subscription. We have a free product that you can download today at Rewind AI, we have an app in the App Store that's free. You can download that Rewind AI. And our focus on that free product is allow you to access your data forever. You can download it. It's frictionless. We want to give you that first magical aha moment. Usually that's like if a tab crash or you discarded an email, or maybe it's that first time it wrote a draft of a summary of a meeting. So that's what we give you for free. And then if you want to upgrade, you can subscribe for $29 per month or $19 if you're willing to pay when you're upfront. So it's a SaaS subscription. We will never sell your data. We don't even have access to it if we could. And we also never do advertising. So we're very much from a business model perspective, aligned around privacy and doing what's right for our users. I love the purity of the fact that if we build a great product, people love it, they'll pay for it. If we don't, they'll fire us, they'll churn, and we need to make the product better. So that's our business model. We've raised $33 million from great investors like Sam Altman, First Round Capital, Andreessen Horowitz and NEA. We've got 25 million in the bank. We've got six years of Runway, so we're not going anywhere. But, yeah, that's the business.

That's crazy. Before we get to a demo, the Mac is really interesting because most people don't understand that if you buy a new Mac today with an M2 chipped in it, that 21% of that is just for neural network. Is for AI inferencing. And you're running in that space, right?

Yeah, I mean, in fact, the only way we could have built the first version for rewind was Apple Silicon. What it did with this, it's called a system on a chip design. It took the key parts of what our app would have had to do in software and the CPU and offloaded it to hardware. So things like compression, taking everything you see on your screen and storing it efficiently on your machine. If we didn't do that efficiently, you know, we compress things around 3,750 times smaller. If we didn't do that, you wouldn't be able to store things indefinitely. But we do. And so effectively, people have around 14 gigabytes a month of data they consume. That's plenty to have years of storage on your machine without even thinking about it. Similarly for automated speech recognition running Whisper locally, that's also leveraging the system on a chip for OCR that's running. So Apple, I think it was very smart to build this architecture to enable software companies like us to really stand on the shoulder of giants and make it imperceptible. That's our goal, is we want it running imperceptibly so it's not hogging your resources, it's not hogging your battery, and it's giving you the superpower without any of the downside of the computation.

Can we see your screen? Because you're an advanced user of this, I'm sure it might be interesting for people. Plus I'll be able to overlay your screen on top of the interview while you're talking because it's pretty crazy what it does.

Sure, yeah. Let me get that ready for you.

Yeah. Give us all your private data. We want Sam Albin's phone number. What was fundraising for you? Because most people don't have the experience you do, getting this kind of money from these kind of quality of people. Any advice for other entrepreneurs? I mean, there's thousands of companies that are chasing funding right now, right? Yeah.

So I don't know if you saw this, but we did this kind of crazy thing where we actually raised our Series A in public. So about in April, I was just getting inundated with investors trying to meet and we didn't really need the money. We just raised our seed from Andreessen and I realized, wow, actually you would build so much trust with users and customers to just put our pitch deck out there. You know, I'd been giving this pitch a lot. You know, it's seven minutes I put it out there and absolutely went viral. We had like 4 million views or maybe 2 million. I don't know what it was. And at the end of the pitch, almost kind of like a joke, I said, go to Rewind AI Invest, and if you're interested, fill out this form. And it's like, what's your name? What's your firm? How much do you want to invest and at what valuation? And of course, are you an accredited investor? That's required if you do this publicly. And we got like a thousand responses and. Which is awesome because then I could plot kind of like what is the fair market value of your company, which is so hard to do as a private company. So make sure we don't get over our skis. We have 20 people want to invest at a billion dollar valuation, but that was kind of insane. We ended up raising at a $350 million valuation. And what I also loved about that is it actually it's selected for the kind of investor that doesn't have this ego associated with us because the founder coming to their Sandhill Road office and kissing their ring. We wanted the kind of investor who's like, look, we're in this for the long term. We recognize you're busy, we'll put in an offer. Our ego isn't so big that we can't just submit a form and get to know you. And by the way, I didn't just take the biggest, as I said, the biggest offer. I actually then spent the next week, I put this out there as well. I sent basically back to back meetings for five days straight, really getting to know the right investors. And so that's how I raised our Series A. We still haven't spent a dollar of it, but, you know, it's, it's, it's good to be in the right place at the right time. I'll just say that.

It sure is, and it is. The funding door is wide open at the moment. It might not be for long, but it is open right now. So congrats. Yeah. Do you have, do you have the ability to give us a demo now? Sure, yeah.

Let me, yeah, let me share my screen here and I will give you a quick demo. So this is my screen and you'll see a nice effect here because this is the, the, the, the interview, the podcast site. But if I want, you know, I could open up, rewind. I could rewind time. So I'm just going back and forth here. This is our conversation just 20 seconds ago. Everything on my screen from the past is copy and pasteable. I can just select it and copy it. So if I ever want to see anything in the past, I want to go back to. That's a very simple thing. I can search for anything I've seen my screen. But the real power comes from our Ask Rewind feature here. So I'll just ask it a question here. Let me try something. Write an email to Robert Scoble thanking him for having me on his podcast. Right. So let's try this. And so what this will do is it'll take every time your name is referred to in the past, it'll take the context of our prior conversation, and it will actually create a email that I can copy and paste and send to you. You can see here it uses this interview itself, the link to the squadcast link, and it actually uses that as a source. So I can go back to that moment in the past, the transcript of our conversation, to generate this. That's one of many examples. This is a good one because your name appears a couple times in my screen. Certainly appears a lot in Twitter. So that's just one of many examples of using the context of your past, using the sources of things you've seen, said, or heard as the basis, the context for what you generate. And after this meeting, I'll copy and paste this and I'll send it to you. So you get a nice personalized email from me. Don't think too less of it, because AI helped me make it.

That's awesome. This new world is just so helpful to get into. The problem is how do we get people into this world where they're willing to put a tool on their Mac that's always listening, always watching and building these kinds of memory aids.

Yeah, I mean, I think it starts with recognizing that the most important thing to get right is trust and privacy. And that was true of many, any technology revolution. I think people, Facebook, in many ways did the world a disservice by making it feel like there's a trade off between convenience and privacy. People think, well, and when you put pictures on the Internet, you lose your privacy. I think you can have both. And maybe I'm too optimistic, but I earnestly think we can build a product that delivers on the promise of personalized AI while not compromising your privacy. And that's the thing we're trying to solve. And you know, maybe we're wrong. Maybe people don't care about privacy anymore.

Most people do. And certainly most people who have never tried thought of trying a tool like yours. Right, because they haven't even really thought about what a large language model can do. I mean, you know, probably in one of the future shows, we have my psychiatrist who built a system to listen to her therapy sessions. You know, and I was patient number one. She talked to me for half an hour and it nailed all my mental illnesses, I mean, in highly scientific, detailed notes, like seven different kinds of notes, just from a 30 minute conversation where it was just listening, not watching, not doing. It's pretty insane what the capabilities of this technology are. Right. And when people see that, they get afraid because. Oh, geez, you mean if I talk to a microphone they can know all of my mental illnesses? That's mind blowing.

Yeah. And I think it's reflecting and understanding what the concerns are. What's the concern around what's going to happen to the data and how could it be in a place you wouldn't want? But I don't think those are unsolvable problems. A good example is today our product integrates with GPT4. Most people are fine with that, but I would say, and most people, I mean, most early doctors of our product are fine with that. But to really go to, you know, if you look at the Geoffrey Moore's, you know, crossing the chasm technology adoption curve, like to really get to the early majority or certainly the late majority, you need a product that's entirely local. You need a product that if you turn your WI fi off, everything works. And that's what local LLMs will offer. You know, we've actually, we demoed this on stage with Pat Gelsinger, the CEO of intel, just a few weeks ago, running Ask Rewind, our Rewind product, entirely local, using llama2 on one of the new Windows intel chips. And it was magical. You know, he gave earlier in his keynote, he asked, he told the story of how, you know, he wears a hearing aid as well. And one of his favorite sounds is the sound of his granddaughters calling him papa. And we actually asked, you know, locally on the machine that was sitting there on stage, what's Pat's favorite sound? And it's able to do that using the data it captures, all locally. So that those are the kinds of things. That's what I mean by making the eavesdropping switchboard operator obsolete. Those are the kinds of technologies we can build to give you the power of personalized AI without any compromises around privacy.

Yeah. And to me, it's not a privacy thing running locally like the Peter Pyatt example I gave you. I wasn't connected, I didn't have Google, I didn't have WI Fi. I was in a conference room where none of that was working. Right. And so I want, if I am going to really rely on such a system to memorize everything and help me out as I live my life. I want it to work when it's not connected. And are times even in modern society when you're not connected, you know, you go to Yellowstone national park, there's no connection unless you brought your own Starlink, you know, and most people don't. So you're out in the middle of nowhere. Well, I want my pendant to still be working. Right?

Yeah. And we don't do the local stuff because of that. I mean, that's, that's a nice to have. We do it because we want to prove that your data is safe. We want to prove that it's not going to some nefarious place that we, it's not going to our servers. And so that's why, you know, that's why I called sort of the WI FI test that was, you know, I think what, what people picked up on is, wow, this thing doesn't have to. That's where I think people get fear is like what happens. And today when you make a phone call, you pick up your phone and you call somebody. Are you afraid that some switchboard operator is listening to it? Probably not. Maybe you're afraid of the NSA and maybe that's, that's fair. But you know, the switchboard operators, they're out of the business. They don't exist. And I think that's what we're trying to overcome with our product is how do we build it so that people don't worry about the switchboard operator. And, and that's where we are today with personalized AI, I think we can get past that and it's just a question of building the right technology and futures to do.

Very cool. Well, thank you so much for giving me a rundown. I think you're real exciting. Do you have two or three other companies that you're watching in the AI space? Doesn't necessarily need to be a competitor. But what has you excited right now?

Yeah, there's a couple companies I'm excited about. I won't name them because I don't want to give too much attention, but there's the idea of taking a pull request or basically taking a ticket in my backlog, like of our linear. We have linear as a project management system and taking that ticket and turning it into a pull request that I can review and sort of taking all that low hanging fruit of things that engineers on your team would have otherwise had to build, I think is super exciting. There's some, some interesting companies in that category.

There's a couple on my developer tools list on x.com.

Yeah, yeah. So I think that's a huge, huge opportunity. You know, I think communication can get better through AI as well. You know, there's very often, you know, they'll have these, these automatic dubbers that will kind of take your video and, and not just dub it, but also make your lips move. You know, I do think that there are some fears that that might, you know, cause some deep fakes. But I think that's, that's, you know, that's interesting. It creates, you know, I think especially in Silicon Valley, people have this very myopic view of the world and they've lose sight of the fact that there's like this whole planet out there, mostly non Americans, certainly mostly non tech people who probably don't know what you do or care what you do, maybe don't speak your language. And I think breaking down those barriers with AI is also incredibly exciting.

Totally, totally agree. It's been a pleasure talking to you on one of the first shows that I'm doing. You're representative of a whole bunch of new kinds of AI companies that are coming along trying to help people by using this new technology. So it's really exciting. You know, one last thing. I'm starting to meet entrepreneurs who are running orchestras of models. Not necessarily all LLMs, but a group of LLMs that might be arguing with each other, correcting each other's answers. Right. Rating each other's answers, that kind of thing. Tell me about it. You know, five, ten years from now, what, what does software look like to you?

Yeah, I think five or ten years from now software is going to look very much like it does today, but the abstraction layer will be much higher. So the things that otherwise, you know, it's maybe the equivalent of going from assembly to an object oriented programming language. You can now go up to literally writing in sentence form or drawing on a napkin what you think the technology or software should do. And it abstracts away all of the complexity of, of making that real in the same way that memory managed object oriented programming language abstracts away all of the underlying assembly that it generates. So I think that looks like that's the future of software creation. The apps themselves will be amazing and I think they will be described best as superpowers. That is the true opportunity for technology and software is to give us the capabilities that we didn't even realize we were missing until we felt them. And that's what we want to do for personalized AI.

Thank you so much for being on the show. You get this. You get your tool at Rewind AI and you can buy the pendant. What do you call the pendant again? It's the Rewind Rewind pendant. I got one on order.

You can pre order it. Yeah, you can pre order that at rewind AI pendant for 15.

Yeah, I got one on order. I can't wait to see that. That's a whole new genre of consumer electronics coming, so that's going to be interesting to watch for you do. Thank you so much for being on the show.

My pleasure. Thanks for having me.