The $10M+ Bet on a Beanie That Reads Your Brain | Sabi & the Future of BCI | E2282

You're staking your entire reputations on a cap that allows me to dictate my words. It does not work. There's no way to put a beanie on and then have it do transcription. Come on.

It is real. You would not believe the amount of engineering and effort that has gone into it to make it work, but it does work.

This doesn't work. I'm calling cap.

How can satellites understand what is the terrain on the Earth even through going through atmosphere? Exactly what we're trying to do is building a GPS for the brain.

You got metaphors. You got a killer idea. Congratulations, Rohit. But this is This Week in Startups. Do you We do have a demo.

This Week in Startups is brought to you by Render. Find out why 5 million developers are already using the all-in-one cloud platform Render. Go to render.com/twist and apply for the Render Startup Program to get $500,000 to $100,000 in free credits depending on your stage and backers. Northwest Registered Agent. Get more when you start your business with Northwest. In 10 clicks in 10 minutes, you can form your company and walk away with a real business identity. Learn more at northwestregisteredagent.com/twist. And LinkedIn Jobs— hire right the first time. Post your first job and get $100 off towards your job post at linkedin.com/twist.

All right everybody, welcome back to TWIST. That's right, it's Monday, April 27th. We are here in Austin, Texas, the great state of Texas, and we've got a full docket here. Things are growing so, um, vibrantly, whether it's This Week in Startups, our new show, This Week in AI, Founder University, now on 3 continents, in Japan, uh, Saudi, in the Middle East, and in the US. We have so much community opportunity in a startup. And this is my tactical, practical tip of the day, Elon, because I, we need to get these tactical practicals in here for people.

Yeah.

Tactical and practical. One person, one job. One person is the CEO of X. So there was a friend of mine, Matt Coffin, and his training philosophy when he did lower my bills, in Santa Monica. He sold it for hundreds of millions of dollars. He did great. I don't know if he ever took a job again after that. I don't think he did. No, never full-time, but he—

I know he opened a store in like, uh, upstate New York.

That was his incredibly talented wife who opened up the store. Um, she's an incredible designer. Putting it all aside, my guy Matt Coughlin said, I make everybody the CEO of their domain. It's always stuck with me. And getting people to think like a CEO, think like a founder, think like a founder is what you've gotta try to inspire everybody to do. So I try to inspire Alon as my editorials are into. Think like the founder of all editorial. What systems do you want to put in place? What goals do you want to put in place? It's not easy to do because not everybody's a lunatic founder, but you can have acts of entrepreneurship in any job, any position. If you want to impress your boss, take whatever little narrow slice of the company and just act like the CEO of that. Just act like it's your domain, your responsibility, and you're just gonna crush it. Now community, it always falls to the wayside. It's always number 2 or 3 on somebody's list, right? That's my failing as a founder to not say, you know what, if this is important for me, I should put somebody in charge of it. So I said, you know what, the sales department's pretty important here. These things are growing. I got Ricky in to be the director of sales.

Boom.

I said, you know, editorial's growing a lot here. I got too many pokers in the fire, speaking gigs, podcast, all in This Week in Startups. I brought Lan in. Lan, you're responsible. Anything that's editorial, Lan has to approve. And, uh, people still try to do end runs around Lan.

They do.

Nope. If it's editorial, Lan must approve.

Sometimes they try.

Founder University, Bianca, uh, Lucas is in charge. Launch Accelerator, Bianca's in charge. The Syndicate, Maddie's in charge. One after the other, you'll see me just deputize somebody. Back office, Kabir's in charge. They are the CEO of that. They are responsible. If you want something to grow and be sustainable, you can't make it number 2 or 3 on somebody's list. It'll never be prioritized. So I am announcing that community is gonna be our big thing. Why? Uh, it would be more interesting for me to be able to meet more fans on the regular. And I did all these different activities, Founder Fridays, meetups, live shows, but nobody owns it. If nobody owns it, it's never gonna excel. If nobody owns it, it's never going to excel. So ownership means it'll excel, and that's what we're going to do. Here's an example. Natila— is it Nikila?

Oh, Nikita. Nikita Beer.

Nikita. Yeah, La Femme Nikita. I gotta remember. Labor Nikita. Labor uses the boar as his, uh, thing. He's awesome. He's on a heater. We should do like Labor Nikita Corner like every couple of shows because I love how he manages that community. He came out with XChat and he hates Twitter groups. He wants it to be the chat metaphor as opposed to this weird hybrid that they made. So here is, uh, Noti Gang. We created a Noti Gang and I just started inviting people. Uh, I publicized the Twitter handle, I publicized the group chat link, and people flooded in. And then I did it for This Week in AI and AI Founders. So here is his inbox. If he goes to the Noti Gang on the left, yep, here it is. These are just Noti Gang members. And, uh, if you click on the management of it, like if you click on the actual title of it, this is really interesting. I think this is going to be a big deal. You'll see the, uh, open group invite link and a number there. Fourth click down, there it is, group invite link. Nope, not that, right above it, group invite link. There you go. And if you click that, uh, what you'll see is you can share that group link and you can approve people. Now when you see approve, you see all the blue check marks, that's great. When I see it, you know what I see? I see also an orange one for my subscribers. Subscribers. So I added all my subscribers and blue checkmarks first, and you can just see there's tons of people waiting to get in. So every couple of days I add 10 people. I don't know how many we're up to, but it's becoming quite a nice little chat stream. This is what we're trying to do, is just get consistent with the audience. We also created, um, one for, uh, This Week in AI, AI Founders, and one for, uh, TAG. So we're just creating these groups and then I did a nice, uh, I was driving to a ranch and I just did a Spaces and I, you could send the Spaces to the group. So then the group fills up with the people in the community and this is the community flywheel that every startup has, whether it's just 10 people who use your product or service, 100 people who read your blog, 1,000 people who read your Substack, whatever it is. If you can activate the top 1% and get 1,000 people who really, really vibe with you, it can be a game changer for your startup. So just, I want everybody in the audience who's a founder to start thinking about these two principles. One, if you don't put a project as number one, it's never gonna break out.

Okay?

You've gotta make somebody the CEO of blank in your startup. Number two, if you could activate your top 1% of your audience, it is going to be a superpower in terms of when you have a question for your audience, you wanna release a new product or service, you need engagement on a tweet, uh, you think you've made a mistake and you want somebody to give you some advice, just try to activate the top 1% of your followers. As opposed to growing the big number. You want to grow the people who are deeply into your brand.

This is— it's the streamer trick is what you're talking about. Like, this is all the big streamers. They keep telling you like, hey, if you want to hang out with me all the time, join this Telegram group. And then that's where they do their like heavy promotion and that's where they're the most active. And because those are the people that are going to click on everything Clavicular shares and check out, buy the merch, you know, like, so that's, that's the funnel that the influencers are running.

Yeah, and some of it's dark. We were seeing with these guys like Clavicular who had like the incel nerd guy who was giving him thousands of dollars, and he had him on a live stream and he's like, oh my God, this person, like, why am I taking $5,000 from this person? It's terrible.

Yeah, it was clearly like a, a person who, you know, like, uh, not, not, not getting out a lot, sort of a, you know, like, not, not, not the person Clavicular wanted to be promoting as like his hardcore fan.

Also, he might feel like he's taking advantage of somebody. Sure, let's just leave it at that. So don't take advantage of your fans, do the opposite, give them access.

Oh, right. And a lot of creators are doing perfectly fine, above board. I didn't mean to use clavicular to sort of crap on the entire creator economy.

There, there is something though about the whales in the system and just being careful that you are not abusing the fact that you might have a super fan or something. I have people all the time who are like, I'll give you $5,000 to go to lunch with me or whatever. Yeah. And I did it once or twice for charity. I did it for Eric Reese's 30 at some point or whatever. I did like 2 lunches and then I just, I just— these people can afford it, but now there's a reciprocity thing. They feel like they gave me the $5,000, it was for charity. So I don't even do that stuff.

Yeah, I think you have to be aware of the parasocial element of everything that we do when you're online. Like, people watch you all the time and they do start to feel like they know you. And I'm not saying it's bad, like, I don't think that's inherently a bad thing, but you do have to be careful. There's a weird—

it's a nuanced thing, right? Let's get to the word nuance. It's nuanced. And yes, the parasocial elements that we now live in from social media, you know, to streamers, it's all different.

Yeah.

Some of it's good, some of it's bad, some of it's neutral. Just, you gotta be cognizant of it, especially if you are, um, yeah. Yeah. Somebody who's an influencer. Uh, we have some news stories, we got some guests. Where are we gonna go first?

I think we should bring on our guests cuz it's a, it's a cool, it's a cool product we spotted online. And the moment Jacob and I saw this going through our feed, we were like, we gotta get these guys in. We're bringing in Rahul Chhabra and Atmadip Banerjee. They're the CEO and CTO of SABI. This is the beanie, Jason, that's got 100,000 sensors in it and it reads your thoughts and translates them into text without needing to get invasive brain surgery.

Come on.

It's a non-invasive BCI cap, essentially. Brain—

BCI, brain-computer interface.

Correct.

Adding a new member to your team is a crucial decision, and you don't want to rush into a hiring situation that you will regret. But if you're a busy founder, you also don't wanna spend a ton of time in the trenches looking for that perfect candidate. Instead, you need a partner and you need a trusted partner. And LinkedIn Hiring Pro is that partner. How do I know this? Because I use LinkedIn Hiring Pro. You want a real-world testimony? All right, here you go. We recently hired a new customer success manager. They handle all the advertising accounts here on This Week in Startups. And they ensure that we're keeping all of our wonderful partners happy. LinkedIn helped us connect with an amazing candidate right here in Austin, and he had exactly the combination of experience we were looking for. It's magic, it's alchemy, it's everything for me to find a great team member. So hire right the first time. Get started by posting your job for free at linkedin.com/twist. Terms and conditions apply. That's linkedin.com/twist. There's no way this works. You guys are punking me. Is Raul— is this guest a fake startup? It does not work. There's no way to put a beanie on and then have it do transcription. Come on.

It is real. You would not believe the amount of engineering and product effort that has gone into it to make it work, but it does work.

Not trying to be a hipster. This is a functional beanie you're wearing, correct?

This is not a functional beanie, but you would have something end of the year.

Oh, okay. That's actually a hipster beanie. Okay. So now we're—

That is a hipster beanie.

This is a fashion beanie, but there is also a technical beanie.

All right, so maybe show us what you built, how it works, and when we can all buy and for how much.

Well, it turns out that for a long time people believe that the only way to do BCI or brain-computing interfaces is to drill a hole inside the skull, put a chip inside the brain. But it turns out that like very recently we had like a paper or an academic book that came out and it was actually authored by Atma Deep in 2023 where they used fMRI signals and use deep learning models to basically be able to decode what a person is looking at using just deep learning models and non-invasive signals. And I think the world before that and world after that are like just fundamentally different, where now you have this entire understanding that you have deep learning, you have great sensors, and now you can understand what a person is looking at, what a person is thinking. You can translate thoughts to text, thoughts to images, and we just figured out how to make it and pack it inside a beanie. And a cap.

Okay, fMRI, explain what that is one more time so we understand what you're talking about and how it works.

fMRI is a massive clinical—

wait, say it again. Is it fMRI? fM, Michael, rI?

fMRI is a— stands for functional magnetic resonance imaging.

It's like the MRI scans that you get, but it's, uh, slightly different from the standard thing that you would normally get.

That is a structural MRI. This is functional, so it measures mostly the blood flow changes across different sections of your brain.

If I get an MRI, that's magnetic resonance imaging, am I correct? Yeah, yeah, yeah. This is fMRI, and the difference is this is doing blood flow. It's checking the blood flow around my brain, correct?

MRI was the initial work we did in 2023. Uh, that is a massive machine in a hospital, but nobody wants to go in a clinic to get the brain read for like 5 minutes. What they want is basically a cap that they can wear in-house. So we shifted from fMRI, which was a massive machine, to biopotential sensors that could essentially fit in the cap. So now what we're doing is it measures electrical activity inside your brain even from outside, and has many, many sensors and a Brain Foundation model that work together to be able to decode what you're thinking. So there's a massive change in the technology, uh, and we had to do like tremendous amount of engineering to get here, but now we're able to get it into a cap.

Okay, so you're going to be able to wear a cap and then an AI model is going to study your, um, the signals Yeah, those signals, then the model will be able to translate into something that's productive. And what you claim, what you're claiming here, Raul, is that it's going to be able to say words in a sentence. Yep, word by word. Now, as the user, do I have to think the quick brown fox, and I just slowly think each word and imagine a quick brown fox jumps over the lazy dog, or can I just be rambling and talking like I would on a podcast and you'll be—

For now you need to do word by word slowly, but I think we aspire to get to the point where you just think, and even before you put it into words, it's out on the screen.

This doesn't work. I'm calling cap. I'm calling cap on the beam. Well, there are—

this is interesting, Jason, because when I was doing research for this, a lot of people in this field are basically saying what you are, that the scalp, the skull, it's muffling these electrical signals too much and they can't get through. So I guess that's one of my questions.

Rahul, can you prove it to us right now?

How did you get around it?

Do you have this cap or not?

I'll give you an example. How can satellites understand what is the terrain on the Earth? Even through going through atmosphere. Exactly what we're trying to do is building a GPS for the brain. When you have 70,000 or 100,000 sensors and your skull has some variability, you can figure out where exactly the neural activity happened when you thought of apple versus when you thought of fox.

Do you have— okay, you got metaphors, you got a killer idea. Congratulations, Rahul. But this is This Week in Startups. Do you have a demo? Do you have a demo?

Not for this call, but we do have demos. We'll show you soon.

Okay. That's what I want to see from you guys. Well, now, where are you guys based? Where are you guys based and what's your background? I'm going to get a little— I'm going to vet this a little bit here. I'm doing a little on-air, just a little tactical, practical. I want diligence. Rahul, where are you based? Where'd you go to school? What did you do before this?

We're in Palo Alto. We're based here. I went to school in Bhopal in India, majored in math and electronics, interned at Stanford. Before that, I built autonomous drones, and then worked at Stanford on early versions of LSTM that came before Transformers. And then eventually got to a point where my college thesis was essentially based on EMG signals analyzing fatigue levels for athletes. Athletes get injured if they keep on training too much, but you can figure out at the right time when their body says that you should stop training. And so you prevent injury, you make sure that they train the best, right? And in those moments, we realize that biosignals, which are time series data, they work. And scaling laws with deep learning models work. And without getting inside the body, you can still figure out what's going on outside. And Atmaneet, who's with me, uh, he also went to college with me. He was a deep learning lead at a satellite company.

So you guys are legit, you're staking your entire reputations on a cap that allows me to dictate my words. Okay. This is a big risk here because if you don't deliver Professor X level fidelity, I'm going to call you guys out in 6 months. I better see some serious fidelity here. But Lan, do you realize how extraordinary this would be? This is basically telekinesis. Like you put your cap on. I have my cap on. You have an ear interface like we had those visual pods on, on Friday. What was that called?

ViewPods? ViewBuds.

ViewBuds, right? ViewBuds. We had ViewBuds on. You put ViewBuds on, I start thinking, we connect over view, and then we have a text-to-speech in my voice with ElevenLabs. Now you're hearing in your ears and I'm going like this.

Yeah, we would have to do this because this is the Professor X move.

But yeah, and it would literally be like me being Professor X and you being Magneto. Right. Let him go. 'You don't need to do this.' 'Charles, you don't understand.' Uh, yeah, you don't need to. What's Magneto's real name?

Uh, Erik.

Erik. Erik and Charles. 'Release him, Erik. You don't need to do this, Erik. He's not the enemy. There's a better way, Erik.' 'Charles, they'll never let us live with them.' 'No, Erik, there's good in humans. You can release—' Are you and I playing Magneto and Professor X? For Christmas, for the Halloween show would be hilarious.

Here's what I was going to say. For now, it's only 30 words a minute. So like, that's a real limitation. We'll have to get it for us. Well, what we're imagining, it would have to go at like, yeah, the speed of thought, which is very fast, faster than I can type. And I can type like 100.

Guys, did you raise money for this lunacy yet? We have. Yes.

Quite a lot of time.

Vinod, Vinod, your friend Vinod Khosla is— Vinod Khosla backed this lunacy? Yeah. Here's his quote.

That's material information.

Here's his quote: If you're going to have a billion people use BCI for access to their computers every day, it can't be invasive. So he's thinking about if we're all going to be think typing on our computers and our phones, you can't— it can't be surgery. It's going to have to be a solution like this where you just put a beanie on.

All right, Raul, next time when you get to the metaphor, say, Jacob, respectfully, Vinod Khosla put 8 figures into the company. Do you respect Vinod and his track record? I'd be like, Yeah, Vinod's pretty smart. I mean, I do think he took a little detour into, um, like alternative fuels. What was the alternative fuel we were all going to use? Ethanol or something? Sure. Yeah. Okay. So you have identified a real problem and you put together a solid solution and a business model that you believe in. So you're all set to launch your new company, right? Uh, not so fast. If you want investors and potential customers to take your new business seriously, you need to consider forming a Delaware C-Corp. And that's where Northwest Registered Agent comes in. They're going to give your new company a real identity. That means an address for your public filings, a domain, a custom website, a business email, and of course, a phone number. And that's going to take just 10 minutes and 10 clicks. They don't charge hidden fees. Customer service is available around the clock. They're not overwhelming your inbox with spam, and they make it easy to cancel at any time. So get all the advantages of a Delaware C-Corp independent regardless of where in the US you're operating from. Visit northwestregisteredagent.com/twist for more details, and the links are in the show notes. I think he was part of the, like, little detour over there with John Doerr that we're going to have ethanol and corn everywhere, and that was how we're going to drive our cars.

Salusik, Salusik ethanol and biofuels. That, that was the goal. Was that Vinod? Yeah.

I mean, listen, he's a big swing guy. Advanced handicapping here on Twist. I don't know if this is his next ethanol. Or if this is his next Sun Microsystems. I think he's famous for Sun Microsystems, if I remember correctly.

He said to me— Oh, he was the first investor in OpenAI.

That too. That too. I think he made a donation to the nonprofit.

Yes. He is the co-founder of Sun Microsystems.

First VC investor in OpenAI. Yes. I think these are just details was what he said to me on stage when I said, how does it go from being a nonprofit to a for-profit? He said, oh, Jason, these are just details. I was like, yes, I, I'm gonna keep that answer in my pocket. Yeah, the next time I get pulled over and they're like, are you, son, do you have a, a weapon? And were you watching the Knicks game while you were on FSD? I'd be like, officer, those are just details. Yeah. All right, listen, Raul, I wish you guys the best of luck. You need to come on this show the moment you have it able to do one word, and that one word is calicanus. You can say calicanus, not Lon or J.

Okay, if you could say— that's not a common one.

If you could say J-Cal, I'll give you two syllables. If you can say J-Cal, I'll buy the first one. All right, great job. Where can people learn more of this lunacy that you're doing here? This is— it's called— this is my type of crazy. What is it? It's called sabi.com.

You can reserve your device right now, and through the end of the year, we will have something for sabi.com.

Well, there it is. I'll give you another—

I'll give you another professional tip here. If you get a 4-letter domain name, that's pretty credible too. So if this is— if you're making all this up, Raul, you have set up the best amount of credibility. You have a 4-letter domain, you got Fidel Kossler, and you got a Stanford PhD. This is either Theranos or it's SpaceX. I am here for it, Raul. I am hoping you get to Mars. I want you back on this program the second I can put a beanie on and tell Lon, you're fired. We go to Mars. That's it.

That's going to be your first beanie. My first thought is going to be, oh man, come and trim your beard.

That's— I'm going to be, go to the barber. What's the barber you go to over here?

Brazos Barbershop, right across.

That's it. If it can make me go, Brazos Barber and take Lon from his homeless era into his tight as right era. I'm good. All right, guys. Great job. Thanks, guys. Very entertaining. Thanks. Very entertaining. I love it. Thank you for being here. These are my guys. I am rooting for these guys. If they do this, you realize this changes the world. I mean, this is a world-changing—

As soon as we saw this go by on X, we were like, we got to book these guys. That is a revolutionary technology. If you figure out how to read my thought. There are— I will say we didn't get to talk about it. There are some pretty epic privacy safety, like a hat that reads my mind. Like, does it read? Do I have to think I want to share this with Sabi, or is it just reading everything that I've got going on in there?

I mean, this is going to have to be on a local model on your phone encrypted because Lon's going to be like, here's, here's Lon Sabi. Ready? Oh no. Gabagool.

Yeah. It's like, oh man. Prosciutto. Where? Oh, gelato. Where am I going to go for gelato next?

Next week, pistachio gelato. Yeah, here's Jakehouse. Here's Jakehouse. For the Bulldog.

Yeah, I wonder how much powder is there this week. Bulldogs, brisket.

Yeah, Sig, get a new Sig, get a Sig P365.

I want to be able to strongly differentiate. Here are the thoughts I want to share with Sabi, here are the thoughts that are just for Lon. And that's— that's—

be honest, you know, they should do a collab with Com. And that would be really great for them to make this for equanimity and meditation. I always tell people my big 5, you can do an infographic here, like make this into a short. These are my big 5. If you're feeling like unbalanced and be dead serious here, number 1, your sleep. Number 2, nutrition. Number 3, uh, working out, exercise. So sleep, nutrition, Exercise, right? Those are obvious pillars of being healthy as a person. I'm going to give you two more: meditation and socialization. Meditation, you just pop out the Calm app— shout out to Calm— and just meditate. 10 minutes, 5 minutes, 50 minutes, doesn't matter. Any amount of meditation will reground you. And then 5, socialization. Invite 3 friends to dinner, open up a bottle of wine, whatever you're into. Maybe 420, I don't know, no judgments, long, you know, whatever, blunt rotation. And, and just, you know Enjoy life a little bit and socialization. Anytime you're feeling out of sorts, take Jake Howe's Big Five. Try to hit all five in one day. Hmm.

It's not as easy as you think. No, not really. It's not. If you get the nutrition is the hard one for me. Nutrition and working out. I can do the other three, but those two are hard.

Here's an easy one for nutrition. Order a goddamn salad and then just eat a chicken breast. I mean, is it going to kill you? It's not. It's not. It's not going to kill you.

I like some veggies. I don't mean to make it sound like that.

Like it's that hard.

But you know, sometimes you want a burger. Sometimes you want a burger.

I mean, I love a burger, but here's what I do. This is my very simple nutrition system. Uh, one for you, uh, two for your health. So you want the burger, that's great. The next day, just eat like an egg white omelet or something nice, and then have a salad for lunch. One for two. There you go. Two for one, one for two. If you're three for three, that's a Gavone situation. Yeah. Now you just try to go from three 3. Am I right in Pulp Fiction? And, uh, in, uh, Inglourious Basterds? Is this the 3 or is this the 3?

This is the American 3. This is the German 3. I'm going with the German, okay?

Yeah, because the Germans are fit. So just try to get, you know, 1, 2, 3 here.

All right, what's next on the show? All right, uh, so up next we're gonna have Oliver join the show. As you know, Jason, as you may recall, we are offering a $5K bounty and a guest spot on this very podcast that you're watching right now for anyone who manages to build a live AI sidebar. There are, there are a bunch of rules. The main idea, we want 4 active AI personas watching the show, making commentary, adding important tidbits. The idea would be like an AI panel of producers that we could consult throughout the livestream that would listen in to everything we're talking about and make helpful suggestions. You were comparing it to Howard Stern's crew. He's got, Robin Quivers there to read the news. He's got Fred Norris with the sound effects and background. He's got Gary Delavate, the producer that has all the context. He's got Jackie, Jackie Martling, and later Artie Lang is the cut-up who's throwing out one-liners and being funny. And so we were looking for how could you— could we use AI to simulate this whole group without actually having all of these people in the studio? And a lot of people have submitted.

All right, Oliver, this obviously presents a number of technical tasks. We had about a dozen people participate. Participate in this. So we've decided instead of just awarding a winner today, that we'll just put, got more work on your plate. Everybody knows you do the demos for This Week in AI. Go to thisweekinai.ai. Oliver is my protégé. He's my protégé. He's really great on air. He's making great demos. So in between the roundtable for This Week in AI, which I host, he does great demos explaining all the basics and even some intricate stuff that you can do with AI. So the application's really hard. In this case, I can identify what's easy and what's hard about this. What's easy about this is creating personas. Hey, you're a fact checker, take this transcript, look for facts. I mean, we do that all the time with the newsletter project you're working on. Fact-checking is easy, but fact-checking a live stream, usually our listeners will do that for us, you know, and they'll go to Claude or they'll go to ChatGPT or Grok, they'll ask a question, they'll cut and paste the answer back, or they'll just tell us, oh yeah, you know, here's the, these are the of, you know, 10 highest market cap company— market cap companies from the '90s, whatever. But this is happening in real time. So what we're asking people to do is to take a live stream, take a live transcript, and then figure out what's the facts in it, and then check the facts in real time. Or what are the discussions and then make a joke about the discussions. That is, is that the hard part here, Oliver?

Yeah, I mean, so on a basic level, it's a pretty simple task. And, and what our, you know, the Noti gang has built has been pretty simple on a base level. What is it doing? Is it making the speech to text and then it's sending, you know, 50 words to Gemini, for example, and it's saying, and it's including the prompt of the persona that it made. So if it's a fact checker persona, it's taking the 50 words saying, hey, is there any facts that you can quickly find that don't line up here? Um, or anything that Jason mentioned that we should correct him on. So on a simple level, that's pretty basic. I would say you just have to connect, you know, an API for the, um, analysis. So on the one that I set up, I was using Gemini, and then you can connect, um, ElevenLabs, uh, speech to text, or. Or OpenAI's Whisper speech-to-text. Or DeepGram. So there's a few different options. Or DeepGram.

Armchair uses DeepGram. Yeah.

Big cloud providers may offer you cheap compute, but you'll end up paying the difference in engineering costs and hiring extra developers. You don't wanna waste time configuring virtual networks. None of us do. Or your access policies. You want your team building your product. So it's time to look at Render. Render is the all-in-one cloud platform for developers that allows you to deploy, scale, and secure your apps and agents with zero ops. Most cloud platforms ask you to split your focus between product and infrastructure, or they force you into platform constraints. You know you're gonna aggro in 6 months, but just connect your GitHub repo to Render and you are live. Web services, cron jobs, the whole stack in one platform. It's time to find out why 5 million developers are already using Render. $5 million. Go to render.com/twist and apply for the Render Startup Program. You'll get anywhere from $500 to $100,000 in free credits depending on your stage and who your backers are. That's render.com/twist. What is Deepgram? I've never heard of that.

It's another option instead of ElevenLabs for a— or Whisper for a— yeah, speech-to-text. It makes transcripts basically. Got it.

It works well. And if you remember the Clipper I made, which takes clips from the show. Yes. What that did is it transcribed the whole show with timestamps using DeepGram and then used a tool that my OpenClaw had that would cut it up based on those timestamps. So DeepGram's great. But the issue that I, we've ran into specifically, you know, for, for all three of the co-hosts that we've tested is that it's taking— it has limited context, which means that specifically for mine, which is the pod commentators, it would every 100 words, it would send to Gemini to get the breakdown. And what that does is it just has the 100 words. It doesn't have the full context of, you know, the whole topic we're talking about.

And that leads to is— yeah. So here's what I think we might want to do just for the fairness of the contest. Go ahead, Lon, give your comment.

I was just gonna say, I feel like there's really two levers so far that that differentiate the quality of the products, at least the, the 12 to 13 I've been through so far. One is ease of use. Some of them, it's a URL, it's instantaneous, it starts working right away. Others, you gotta massage it, you gotta work with it, you gotta set stuff up yourself. Uh, that's, that's sort of one. And then two is the quality of the personas. As you'll see, a lot of them, the personas are all pretty similar. They're making kind of similar comments. And then some of them have done a really good job differentiating where it's like, this is the troll, this is the fact check, and they're giving better quality. So like, I feel like that's where a lot of the variability is coming.

All right, so listen, this is a contest. So it's, you know, and I just kind of threw it out there, it's my money. So ultimately I'll decide with the counsel, uh, of you two guys and the rest of the team. Sure. But I think in fairness, what matters most, the high order bit, is Is it doing great fact-checking? Is it doing great trolling? And is it doing great jokes? Those are three very different things, right? Those are the three personas that I think matter the most. The fact-checker. Yeah. Who's just giving facts. The trolling, who's just roasting you, etc.

Being cynical, being snarky.

Cynic. Okay, let's call it the cynic. Yeah. Or no, I think the roastmaster is even better.

Well, I mean, a lot of them have gone with— Or no, maybe the cynic is better. The troll is what a lot of them are using because that's what we've sort of defined Dave Erickson described it as.

But what would you say, Lan, then? Is it the cynic? What would have the most overall value for the show? This Week in Startups, All In, whatever. I mean, I would say— the cynic, the troll, or the roast master?

I like cynic the best because I feel like that's a clear role. Like a troll is just— it's had a negative connotation. They're just there to disrupt. A cynic is a helpful thing to have on a show like this because they're exactly like what you were doing for Sabi, like somebody to sit there and be like, 'Cause this really worked challenging the frame.

All right. So let's keep it simple. I think the joke and the troll thing is very hard to do because— so let's take those off the table and let's narrow the contest down to the fact-checker and the cynic. Okay. Because those two, I think, would have the most valuable— would be most valuable to us. Yes.

We'll let everybody in the chat room know we're reshaping it in some way. They're just simpler. Because I think we've made the mission too big with 4 personas. It's— it's— It's challenging. And, and, you know, like, well, as you'll see if we show some of these off, uh, it— some people— some, some of them, the differentiation is not really there. It's just a lot of commentary. It's not very personality-driven.

I just want to make it easier to judge the quality, Oliver. So I just think in judging the quality, ease of use of the product, so the fit and finish of the project, and then the quality of the output— those two would be the weights we would use. Is there any weight I'm missing here, Oliver, in terms of picking a winner and being fair about it? And I, and I could pick two winners and split it. You know, it's my money. I can do whatever I want here. This is a, a rolling bounty here.

No, I think those two things are great, but I also just wanna give two specifics that Lon and I found that I think would really help a lot of the submissions, which are speed. How could you increase the speed? Yeah. Uh, so if we're talking about a subject, how quickly is it gonna give us the notes. So in our Slack channel, it's me listening live and like typing as fast as I can. Like, oh, Jason, what about this? Jason, what about this? Also in the model, in the one that I tested, it would write something that was hard to read and I wasn't sure what it was talking about, if that makes sense. So yeah, you got, it's got a, in one of the ones that Lon has is it actually has the transcript and then the commentary.

One of the submissions. Yeah.

What, what? Our armchair— our armchair did a really good job as it highlights the section of the transcript that the commentary is about. So you can see immediately, oh, the troll is picking on this thing Jason said. Oh, the fact check is about this thing Lan said. Right. Okay.

So this speaks to the fit and finish of the product. Yeah. Which is to say, for this to be useful to the host of the show or the audience, because I think this serves two roles. One, I could run this privately and just see when I'm on Megyn Kelly what facts she's getting right, what facts she's getting wrong, and I could be challenging her on a fact and say like, okay, is that actually true? And then I could go look at it, which by the way, everybody's doing right now. There was a famous Larry Summers episode of All In where he claimed, you know, uh, I think he said Chamath's looking stuff up or whatever. That's kind of all fair in love and war now in terms of, you know, a podcast. People are looking up stuff in real time. So let's go into our demos here. So we always like to demo or die. Who's going to lead the demos? Uh, I'm, I'm actually—

I started sharing already. This is, uh, this is Armchair, the one of the first ones that we set up. Okay. Uh, it is by— I'm looking up his name right here— Mark Kolbrügge set this up. I believe that's how— and as you can see, this is what I really liked about it. It's highlighted this block of what we're talking about Uh, you know, you're— this is—

be more specific here so that the audience listening to SportsCast, please.

The quote here is, as you said, there was a famous Larry Summers episode of All In where he claimed that Chamath was looking stuff up. It's highlighted that. And then Gary, their producer with context— oh, we're using this right now? Yep, this is live. I'm, I'm doing this live. Uh, Gary says in April 2025, All In had a tariff debate with Summers and Chamath Saks using Grok ChatGPT. Live per Reddit buzz. They couldn't confirm whether Chamath was actually looking stuff up. And then here a little later, I mentioned Mark Kohlbrugge and Gary says, yes, he found it. Mark created Armchair. He found his, you know, he found his URL. He corrected me on the spelling. Wow. Yeah. So you can see it's actually— it's working live.

And okay, wait, let me, let me try using this now. This is incredible. Holy Cow, let me do this right now. Um, let's do it, uh, based on a political topic. I'll do a political topic right now. Uh, obviously we know there was a, an incident at the White House Correspondents' Dinner this weekend, and the buzz online was which side of the aisle has the most extremists. Is it the right with the Oath Keepers and the Proud Boys and what happened on January 6th where there was violence at the Capitol? Or does the left have the violence problem with the recent New York Times podcast where they were talking about soft murder or something and they were talking about microlooting and stealing from the Louvre? And then we've had the 3 assassination attempts, including this weekend on President Trump, all from left mentally ill. I think you got to be mentally ill to do an assassination. So I'm curious, of these instances— I mentioned a handful of them— but if you were to take political violence in the last 20 years, I'm curious which side would have the most notable actual violent behavior in the market, the Democrats or the Republicans? All right. So as you can see, I spoke slowly, by the way, Lon. Yeah. And I gave the details as best I could. And that might be how you would hear me tee it up on All In.

Yeah, not too different. So Gary confirmed the shooting. April 25th, 2026, at the White House Correspondents' Dinner at the Washington Hilton. Gunman breached checkpoint, fired shots. Trump and officials evacuated unharmed. It even named suspect Cole Thomas Allen, was arrested. And it gave, gave citations. It gave you this, the Wikipedia citation for the event. Yeah. And then the troll said, your framing assumes January 6th defines right-wing extremism, but you missed Antifa riots, the 2020s, $2 billion in insured damages, campus cancellations. The troll says your selective memory isn't data. And then the troll— Hold on.

Pause there for a second, Lan. I didn't bring up Antifa. That wasn't on the top of my head because we haven't had an Antifa situation in a little while. But during BLM riots on campuses, there was a lot of Antifa violence. Yeah, I think when Biden was in office or maybe the first term, I can't remember all the things that were ascribed specifically to Antifa, but Antifa is a known violent agitator. So that's great.

That helped me. Yeah. So, I mean, and as we keep going, the troll and Gary, they seem to be more active. I think there are other personas in here, but we're hearing a lot from the troll and Gary, the producer. But yeah, here's another one. It confirmed you were talking about that April 22nd New York Times podcast with Jia Tolentino, Hasan Piker, and Nadia Spiegelman. It mentioned microlooting, the Louvre theft, social murder, and 3 more links that you could click to check out. This is so great.

I mean, this— I'll tell you why this is great for me. I'm frequently saying during the show, hey, producers, get me this piece of information. You are? Because I can't remember everything. You know, no one person can. And maybe Ben Shapiro, because he does monologues. So I suspect, you know, he's so good at debating, he's got such a narrow focus. Ben Shapiro is so good at, you know, doing these that he's written a really great script for himself, like he's a monologuer. So monologuers get to practice their monologue, take 4 or 5 takes. I'm live. You don't get 4 or 5 takes, right? So this would be very helpful for people who are live just to remember. I don't know who the other co-hosts were on that New York Times podcast. This would have filled me in here. Correct.

This is very impressive, by the way. I think— I mean, for me personally, we were, we were originally thinking we were going to pick a winner. So far, Armchair is my pick for the winner. It's the best job of the actually adding helpful context in the personas that we defined.

I want this, by the way, as a Zoom plugin or ODBC on my desktop. Yeah. So Oliver, is that possible for me to set this up and just have it running when I'm on a Zoom call with a founder or I'm on, uh, you know, a podcast, whatever I'm on, somebody else's podcast? Can I have this just running and using the Zoom stream? How does it grab the stream? It's just on a technical basis.

I can share my screen here, and this is actually how the a project that I looked at is built. So this is a different one. Of course, this is not Armchair, it's called Pod Commentators. I also think it's called Cast Side, uh, multiple names here. But you can see I'm looking at the Zoom through my Chrome browser window and I'm streaming into Pod Commentators so I can see what I would see in my Zoom window. And also it's popping up on— and I'm also getting takes from Rex, the cynical commentator, popping up on the kind of side of the screen so I can see.

Is that through Zoom or is that a plugin? How is that? Oh, it's a browser window. So it's—

yeah, yeah, it's all through your browser. If you, you run the Zoom in a, in a tab on your browser and then you link these tools to that tab and it just listens in and runs it live. Incredible. It is incredible.

Incredible. Both of these are very strong contenders. Um, You know, I have a lot of edits here, but man, I am just shocked at what you can get done with a bounty contest. These two are very impressive. We got one more, and then we're gonna give— just so we're gonna— just so we're fair, because I, I'm literally gonna cut this $5,000 check, and I might give it, you know, $3K to one person and then two honorable mentions for $1K each. I may give $5K to one person, but I think we should do— give everybody the ability to do two iterations. 'Cause I don't wanna drag this on and be unfair, but since we didn't have any rules, I wanna be clear 'cause I'm a man of my word. If I'm gonna give $5,000 and I said I would pay $5,000 for this, I wanna pay the $5,000, but I want it to be my spec. Here's my spec. I wanna be able to run this either on a public stream or I wanna be able to run it privately when I'm on a Zoom. Okay. Very simple. Either way. Um, and then I'll narrow it down to, I just need, I don't need 4 personas. I need 1 persona. I'll take two personas, a fact checker with citations, facts, and then a cynic. I'll take a cynic. I'll take a fact checker. So I'm cutting that down by half. You don't have to try to get the jokes right. That's my job. That's on me and Lan. I just want fact, real-time fact checking and real-time cynic, you know, like giving me the other side of it. I love the idea of the transcript being shown and the addition. I think that's key because that allows me to be scrolling up and down. So if the guest was talking or if I was being interviewed, I'm being asked my next question. I can zoom up and down and say, oh yeah, you know, in that previous answer I gave, I mentioned this podcast. It was Hasan Piker and Jia and this person, one for the New Yorker. Jia works for the New Yorker and this person works for the New York Times. So I want to make sure I get my facts checked. Exactly. So that real-time fact-checking is so incredible. Okay.

The possibilities are, I think, could go a little further even here. As these models have bigger context windows, we can feed in our full docket and it can, you know, say, hey, this guest posted about this on X that we already have in the docket, and it can connect there. Okay, yeah, that could be in its memory.

Yeah, okay. I don't want to be unfair and make it a never-ending— I don't want to keep moving the goalposts, but since— and I didn't think anybody would build this, I just said it, hey, you know, if somebody builds this, I'll give them $5K. I didn't think 15 people would come out here, so now we've— I've got to do a little cleanup work for the next bounty. We're gonna announce our bounties with like a very specific process. So we're gonna do this 2 more weeks in a row. This is incredibly exciting. Um, we're gonna do 2 more weeks in a row. So like maybe next Friday and then the Friday after. This way people have like essentially 3 weeks of refinement. So today's Monday, April 27th. So we'll do May 1st. Well, that's plus 7 will be May.

This is right. May 1st is this Friday, then May 8th is the next Friday, and then May 15th is the Friday after that.

Okay, so May 8th is, uh, the second week of the— will be the second check-in, and then the final winner will be whatever you get done for the 15th. Fair enough?

I think that's fair. Do you want to— do you want to check out number 3? We've got Patrick. Please give me number 3. All right, here's— here's number 3. I'm gonna flip over to the tab right now. Uh, Sidecast, maybe it's called. You could see I've already been filling in the transcript and the agents are on their way. Now my one comment here It works amazingly. This was no setup. You just go to this URL and I'm just— it's working all on its own. The one, the one negative—

this is running right now or this is running on a previous episode?

This is running right now. You could see it's, it's, it's transcribing us live right down there. The one back— the one back draw of this one or drawback, all the personalities are pretty similar. They're all very snarky. It's sort of like having 4 trolls watching the show.

I think that's the easiest thing to correct for though. So So just again, for whoever produced this one, what's the name of this one?

BMD Pat, I believe, is what he's calling it. This was Patrick Hughes, our viewer. Fantastic.

Yeah. Can you scroll back to the political violence discussion?

I think I started it after the political violence discussion.

Oh, so you actually fired this up. Okay. I fired this up.

This is fantastic. Anyone could go to BMDpat.com/pod and check it out.

This is amazing. Uh, Fact-Checking and Cynic. We're going to narrow down to those two. We're going to put a Notion link in the enhanced show notes for today's show in the description on YouTube about the contest. So there'll be a contest page with the rules, the dates. So if you're listening to this and you want to join, you can join now, uh, and you can just beat everybody. But go ahead and join. Uh, if you don't want the prize, you can donate it to charity. You pick the charity as long as it's not something crazy and offensive. Um, and the person gets to come on the show and get their flowers. We will give the Twist Cup, the Twist Cup, to somebody on the show. There'll be some sort of Twist Cup, the Twistivation prize, something.

Twistivation. Yes, we'll come up with it.

We'll do like a Twist Bounty. Twist Bounty 1 will be done. So we're going to put that in the links, and we'll link to all the existing, uh, submissions. Yes, this is fantastic. I'm going to announce another contest. Contest right now. I'm gonna announce another one. All right. This is inspiring to me. I think the audience is gonna love this as well, cuz it shows what the tools can do. Uh, and this is adjacent. I own the domain name annotated.com. Hmm. I had talked publicly that I'm looking for a developer or a team. I'm gonna narrow the scope. I want this person to actually put this in the world for all of us to use. I want an annotation tool. Hmm. This tool is Uh, a sidebar Chrome extension. It allows you, if you go to annotate it, to create an account. I don't know, log in with whatever service you want. X and Google, I think, are like probably the two big ones. And X is important for a reason. Then you put in a URL or it uses whatever page you're on. So if you're on the New York Times and the story comes out, you know, or a podcast comes out or a YouTube page, so the two Two examples I'd like to see is a news story and a YouTube page specifically. You pick the YouTube page, you say, it asks you, where do you want to start? Where do you want to end? You say 1 minute in and out at 3 minutes. It takes that, pulls the transcript, pulls the video, makes a landing page, annotated.com/lons, L-O-N-S, slash, you know, YouTube video. The, you know, uh, like a, a, a intelligent title. It then puts the clip on that landing page. We rip the clip 'cause it's fair use. Maximum clip size is, let's make it 90 seconds. That would be fair use. 90 seconds is your maximum clip size and it downgrades it to a smaller file size, right? So make it like, I don't know, 240 pixels, not even 480. So we're not stealing content. We're using fair use, and you can look up fair use. Fair use would be, I'm taking a small amount, I'm not using in the original resolution, I'm not interfering, but you have to write a commentary on it. And the commentary has to be a certain length. So either text or audio, whatever, but I just think text is fine. So you comment on it and then there's a button that says, um, you know, I'm going to do a fair use claim, or, you know, I, I believe you're stealing the content. And then we have to make a determination if it's fair use or not. You can't just use it to clip and steal. Or you take a New York Times story and you say, clip this section. You highlight a section, you say clip, and then you put your commentary on it. It links back to the original. It gives credit to the source. Maybe it takes a screenshot. In other words, we don't wanna screw the person's original content. It is, we just want to let you annotate somebody's original content. That's it. That's the contest. It has to be a live service so you can make whatever you want right now.

Is it like a social network? Like the idea is I have a feed and I'm looking at other people's— annotations. That's the, the sort of a community.

The annotations will be saved based on account. You can do an anonymous annotation.

And, uh, you know what, you know what it reminds me of, Jason, in a weird way? Do you remember Delicious?

It is like a Delicious, like a bookmark.

It's like a hot— it's an enhanced multimedia Delicious.

Correct. Yeah. And, uh, yes, you have a stream of everything you've done, right?

So that would be delicious. You would bookmark things that you wanted to read, but I could check out somebody else's bookmarks.

And on the annotated homepage. The annotated.com homepage is just download the sidebar, and on the thing it just says, um, it just gives like a, a feed of whoever posted last, and then you can vote them up and down, and you can comment under somebody else's. So when you say this section— yeah, but under your section where you're fact-checking the New York Times, or you're, you know, taking a Ben Shapiro clip and you're arguing the other side of it from his Oh, and you could do a podcast too. So audio, video, or text. Then you do the audio, uh, of a Ben Shapiro thing. I can comment under yours, so it has thread comments underneath it.

Yeah, yeah. I mean, that would be the idea over time, is people like a leaderboard, I think, would be really cool. Like, here are the 10 most annotated pieces of media.

Anyway, there's a bunch of ideas. Uh, I'll put this up for 5 grand. 5,000. Another 5 stack I just gave. That's too bad. We'll do it over We'll do it. Oh yeah, it's 5 dime skis. Uh, I own the rights to it, so if you make it, I own it. That's it. Because I'm gonna make this an actual public service, but I might hire you for $5K a month to maintain it for me. I have to come up with the business model eventually, but I might pay you to if you're good at this. But anyway, this would be like, show your work. Worst case scenario, you make $5K. Uh, and worst case scenario for me is I throw it away, uh, and I burn $5K. Best case scenario, I hire you or I partner with you to make it a commercial service. Uh, this is what I always dreamed of building because a lot of times, I don't know if you ever had this, Oliver, you're like, that's BS or that's brilliant, and I want to clip it and then comment on it in fair use. That's the key, it has to be fair use. So you can't take the whole New York Times story. Let's set the, um, let's set each clip from text. Let's keep audio and video clips are 90 seconds or less. Text clips, what would you say, Elon?

I mean, I think if you're taking 100 words— yeah, I was going to say 75 to 100 words. Beyond that, you're sort of pulling a big chunk of the article, but 100 words—

let's go with 100 words.

You said 2 sentences, you know, 3 sentences that stand out to you particularly.

If you want to do more than 100, you just clip the next 100, and then you could thread that with the previous Yeah. So let's say you wanted to go through a New York Times article and correct 5 points. It's just 100 words at a time. Yeah.

Or you could just take—

and the person still has to go to the New York Times to read the full story.

I mean, that's what I was going to say is you could take the most offending sentence, annotate that, and then just add your comments about the entire article. People would still be able to get back to the article. Correct.

And always a link to the original source. Oliver, what do you think of this contest?

You like this one? Yeah, I think linking back to the original content, people love to see the top stories and I love it. I'm excited to get my clipper working for me, reviewing those clips, having a take and dropping in my take.

Well, Oliver's going to, he's going to work Sunday and he's going to just drop this and he's going to take the $5K.

I'll just, I'm taking what I'm doing after this call is I'm taking this transcript saying make all of Jason's ideas and I'm entering the competition. There it is. Perfect. Yeah.

You just tell your agent, go make this. Yeah. Yeah. Shake Al and make him what he wants. All right, everybody. Great job. Oliver, great job. Look in the show notes for Bounty 1. Bounty 1 is the Realtime Fact Checker cynic. Bounty 2 is annotated. So now we have 2 bounties. We'll have a bounty page. This is going to be a recurring thing here on the show. Bounties for everybody. All right, everybody. It's been another amazing episode of TWIST. Bye-bye.