The following is an AI-generated transcript from Podsqueeze*
Rod (00:00:00) - Jerome, thank you so much for joining me and my audience today. I'm really pleased to meet you and learn more about you and your product.
Jerome (00:00:09) - Thanks for having me.
Rod (00:00:11) - Uh, before we jump into Sizzle, I want you to tell my audience a little bit about yourself and how you got involved and what your background is.
Jerome (00:00:21) - Sure. I mean, the first thing I want to do as a disclaimer is I am not an ad specialist by any measure. I would say my biggest specialty there is I have four kids, and I spent a lot of time with them and tutoring them. I think they wish they didn't have to come to me, but I'm there, I would say the number one resource in terms of tutoring. So that's all my excitement of having gone to school myself. And I'm a kid, so I don't I don't portray myself as a specialist. My background is in AI. I've been in that field for 25 years. This is my third AI startup. I've done like two others, one in B2B space text mining in 2000.
Jerome (00:01:03) - Before it was the fashion now and then I sold it to IBM and I was IBM for a few years, also in AI, putting it in the cloud. Then I switched to AI for drug discovery for a few years and other startups. And then my last role was at meta as a head of AI. There for the entire company. And then I left a year ago to do another startup, because this is the kind of thing I'd like to do. I'd like to build things. And I really thought that it was time to do something and learning and education with AI, and I thought there was an amazing opportunity there. And that's what Sizzle is about.
Rod (00:01:38) - Right? Well, that's my bailiwick. You know, I've been in higher ed my whole career, and I've always been about using technology to improve education. You know, it's always been the holy grail to have a, you know, a custom, you know, individual tutor. And my heavens, we're we're there, almost there or we're there.
Rod (00:01:59) - Why don't you tell my audience about what Sizzle does, how it works.
Jerome (00:02:05) - Yeah. I mean, so there's it today and there's the vision. So what it is today is really like, I would say a bit of a tutor in a pocket. You give it a problem and it helps you solve it. It tries to get you to give you the steps and then tries to get you to answer these steps one by one and walk you through the problem. So I kind of like a work example that tries to get you to answer the steps each time. That's the first step we're trying to do, but really our mission. We want to make learning amazing for everyone, and we want to create an experience that feels a lot more personalized, a lot more one on one. Where we'll go next is try to get people to come back to the app to learn things, you know, to commit to learning, to get feedback on how well they are doing, to give them a curriculum on how to learn a certain skill and, you know, incentivize them to go through them one by one.
Jerome (00:02:54) - So we start with homework help. And our goal is really to get people to learn really anything in the most engaging and effective way.
Rod (00:03:03) - What? Who's your main audience? Is it? Middle school, high school, college or all?
Jerome (00:03:10) - Today's really? I would say the majority of our users are high school. This is actually who we were targeting. You know, looking at high school problems and Stem problems. That's what we started. I would say it's more than 50%. We do have a good portion also in middle school and college. I would say it makes the next two quarters our audience good.
Rod (00:03:30) - When did you actually get started? To be honest. Hadn't heard about it until very recently. So when did you start? Yeah, we launched.
Jerome (00:03:37) - Actually pretty recently. We launched the app officially mid August.
Rod (00:03:42) - Oh, okay. So that's very, very new. Um, and you know, I'm going to have this in my show notes. But according to your website it says Sizzle eye is the future of learning a free AI tutor for everyone.
Rod (00:03:54) - So that sounds just wonderful. How do you make it free? How do you make any money?
Jerome (00:04:00) - So right now our goal is to, you know, our ongoing right now is to go for a wider audience and understand user behavior and collect interesting data. So we're trying to actually create better language models, get interesting data around what problems we are trying to solve. So our incentive at the moment is to build a large audience. And the goal is to create AI that can help people learn about these problems. But if we can help them learn this problem, and we'll be able to help people learn any kind of problems, because we're not hardcoding anything in the system. And we think at some point there will be a good audience that we can monetize, like people trying to upskill themselves, people trying to onboard in the company. So I think if we can do this use case, there'll be a lot of other use cases that are a lot more monetizable than kids in school. So this is when we're taking kind of a really long term approach of trying to provide something for free that in the past people have had to pay for because it will allow us to create a really more interesting and broader product that we feel will be able to monetize later on.
Rod (00:05:05) - Uh, that sounds like a great idea. As long as you must have the backing to do it. Which is great. How did you do it? Yeah.
Jerome (00:05:14) - We're trying to. Yeah, exactly. We're trying to have a really broad vision and get the venture capital behind us that allows us to do this. Our goal is to be an app that a billion users will use one day. So.
Rod (00:05:28) - Well that's great. Um, how is your experience at meta, how did that inform the way you're going about producing this app, or the things that this idea came out of the blue with you? I hear you have children. Um, you know, I'm curious why you just didn't stay and do this in meta, for example, it.
Jerome (00:05:51) - Was actually very informed by my experience at MIT. So I can tell you a little bit of backstory. Um, the thing when I was a man, I think, you know, when I was there, a lot of interesting thing I did and thinking my team did like things like PyTorch using AI for content moderation, you know, pretty, pretty impactful thing using AI for ads, which works amazingly well.
Jerome (00:06:11) - And but I really wanted to do like, how can you tackle this, you know, this amazing powering AI to do something for people's lives that they feel is extremely valuable to them? Sometimes we use the AI to get people to come back to an app, you know, to stay hooked up to it. But if you ask people, you know, a month later, you know, why is that really valuable to you? It's not clear, you know? And so what I want to do is an app that when people use it like Sizzle, they feel a month later I'd be like, oh, I really learned something. I really spent some time on the app, but it brought me something that's gonna be useful for my life, and that makes me a better person. So that was really my driver for me. And, you know, how can I use all this power of AI to do something that people feel really, really good about? Like I spent every day on Sizzle 10 minutes or 30 minutes and I feel good about it.
Jerome (00:07:01) - That's my goal. Now, what's interesting is meta. We learn to get people to use these apps a lot more than people have done in education. You know, they really have recipes that work amazingly well. You know, we understand what makes products retentive. And so this is what I want to apply to the learning and education world, where retention is usually a big problem, whereas in social media apps have been rebranded better.
Rod (00:07:28) - Right, right. So that makes me wonder, are you promoting it actively on Instagram and Facebook then, or or is it just growing organically at this point?
Jerome (00:07:40) - So we are doing both right now. I think most of the growth is organic, but we're also experimenting on how to promote it on sites like TikTok and Instagram at the moment as well. But we are, you know, we're just we have delivered just a small portion of our vision. You know, the goal at some point is to get a product that people come back to, you know, we want to have something like the deli Sizzle and they come every day.
Jerome (00:07:59) - They do something and they feel like, oh, I'm making progress towards my learning goals, you know?
Rod (00:08:04) - Okay, um, give my audience sort of the range of, of subjects. It's mostly math and science. Um, how would you describe that in terms of your main audience? Your main?
Jerome (00:08:19) - Yeah. Right now we're really trying to position it more stem, you know, and, you know, the advantage of a product like we have today compared to what was in the past, like the math is that it can handle word problems, but also you can handle non math problems. You know, things like chemistry, physics, biology. Some people use it also for broader problems. You know it does things you can even ask it to for a recipe will give you step by step to this and design it for that. Yet, you know, but the technology behind it, which is, large language models tend to work equally well. The form factor right now is around step by step, step by step.
Jerome (00:08:54) - Stand to be more meaningful for Stem problems at the moment, you know.
Rod (00:08:59) - Right? Yeah, I spent some time on it and I like how that works in terms of asking a question. And then it asks you to respond. And if you don't want to just respond off the top of your head, it can prompt you for some choices, like a multiple choice test. But there's no grading per se. It's just helping people solve a problem.
Jerome (00:09:24) - Yeah. And this is actually the feature you point out as one of the things we're trying to do is how do you get people. I mean, we do a lot of them to see the answer of each step. But we try to really like, nudge them to do more, to test themselves or to answer it by themselves. So this is the game we want to be in, which is how we get the user to do more and more on their own. And that's what we'll keep pushing features in the app to get people to do a lot more and be more proficient in solving these kinds of problems on their own.
Rod (00:09:53) - You know, this brings up. Uh, the question that I think higher ed, a lot of education in general is, is trying to determine how they're going to handle AI tools. I do a little bit of teaching. In fact, I teach a course in podcasting. Just, you know, I was offered this little adjunct position. I thought it might be fun. And I said, how does your school handle it? Uh, they said, well, we can't, we can't use it when we turn in a, you know, a written exam. We're not supposed to use it. And I don't know how they, you know, control that. And I said, well, for my course, there's no written exams, there's no multiple choice. It's all the audio that you're producing. So my philosophy is, you know, people say, well, you know, they're going to be put out of a job by I say, no, you're going to be put out of a job by a person who knows how to use AI better than you do.
Rod (00:10:50) - So right now, learn how to use AI as best you can, and, you know, and people are saying, well, this is the end of the written exam. You just have to ask, you know, verbally, you have to do an oral exam for many, many topics. What's your philosophy around using AI and cheating and that sort of thing?
Jerome (00:11:12) - So first I want to say I really like what you said, which is, you know, I've been in that space for a long time. You're not going to be replaced by AI, but you could be replaced by someone using AI better than you do. I think this is really two things, which means that schools just can't decide to ignore it. Right? This is a tool. It's like it will be in ten years. Like asking not to use this tool and be like, oh, you should go to school without a computer. That's not going to happen, right? This is the tool, to be honest, right? Today, like when I recruit people, this is one of the first things I ask them.
Jerome (00:11:43) - When you code, are you using GPT? Are you using copilot? It's an amazing tool. They're going to make you 50% more productive. Why wouldn't you use them? So I think it's really important to embrace this. Now. It doesn't mean that companies like ours don't have a responsibility. I think we really do have a responsibility. And this is why for us, as a top objective that we are taking is to get people to learn, not to cheat, not to get the answers, but really try to incentivize them to really do the work or really get, you know, the skills. But it also means that we are conscious that this tool exists, that it's seasonal or others. They will be tools for people to see answers. And ultimately that's not what matters if people get the answer right or wrong, right. Can they show their work? Can they do it? They're right and we can see if they are actually right. Then we can report back on that. So I think that's going to be the future.
Jerome (00:12:32) - I'm optimistic that the future will be good. It means the school embracing it and the people providing it, optimizing for the right thing. Right. Not just providing some shortcuts and ultimately don't really solve anything.
Rod (00:12:44) - Right. So you sound like you're an optimist like me. I, I write a newsletter on I'm the militant optimist, and I, I write about the good things and how technology is going to help us in the future. So I think we're on the same page there. There's been so much talk about the existential threat of AI, because what's your view on that?
Jerome (00:13:07) - Again, I think it's the same. I mean, I don't buy it in terms of replacing people. I mean, the whole history of automation has shown that it creates so much value. Then you can get even more economic activity out the end. I think it will be the same. I do think it will be a tool that's unavoidable. Again, it will feel in 10 to 20 years the same as not using a computer, which is a non-starter in most jobs out there, right? But the third thing is, I do think that with great power comes great responsibility.
Jerome (00:13:39) - And you need to be, you know, there needs to be some regulation around it. There needs to be some understanding of the deep aspect. I'm not at all admirer, which is I think the problems, the numbers talk about it are just not relevant today. It's not about AI taking over. It's still pretty dumb overall. You know, people don't quite understand how it's amazing what it does, but it's not human at all. You know, like it doesn't do things the same way. It's like it's very knowledgeable about everything, right? You can incorporate a lot of knowledge, but it's based on a lot of memory rather than deep knowledge, understanding or deep reasoning behind it. It's still very useful. So not going to replace people and we still have a lot of control over it, but badly implemented, right. It can duplicate bias, for example, which can be a huge problem. It can disrupt learning, like you said, right? It could be like a cheating tool. So how do we actually go around this? Right.
Jerome (00:14:30) - How do we design it in a way that incentivizes for the right thing? It's a lot more important to understand the objective of that AI. How is it coded versus just thinking, oh, it's going to take over the world or saying, oh, there's no problem with it. You know, it's right. The reality is in the middle, you know.
Rod (00:14:47) - Um, getting back to your app. Um, it's available on the, on the web. And there's an app. There's also an app you can download to, I guess, iOS and Android. I'm assuming I haven't tried the app yet. And what can you tell us about competition? I know Khan Academy has been around for a long time, and they have a new AI powered tutor called Khan Conmigo. Um, how can you compare your product to Conmigo for example?
Jerome (00:15:19) - Yeah. So first I'm in the project, as you mentioned. I mean, we started with an app both on Android and iOS, and we just launched the web version.
Jerome (00:15:26) - So you can use it on the desktop, just the web browser just yesterday actually. So that was our.
Rod (00:15:33) - Really? I was one of your first users today.
Jerome (00:15:35) - Yes. It was really recent. I mean I do think right now competition is good. I think people are trying to figure out what is the way that learning will leverage this language model like power. There's like which model? Um, my view, I think it's interesting to see what everybody is doing. I think it's super interesting to see them embracing it and kind of mingling. And I think it's very interesting and pretty well done implementation for you. Another is that I think the interface will be a lot less chat-like and a lot more fit for purpose. I think one of the challenges is just a pure chat interface, like trying to use Khanamigo is like you do all this work and ask you a lot of questions. It's kind of cool. You know how it works. But then after, okay, you've been successful and it just looks the same.
Jerome (00:16:26) - It says, hey, you did it. Okay, cool. Let's move on to something else. You know, as a student, if it's like, okay, dynamic progress today, you know, and you need incentives, right? You need the feeling of completion. You need a feeling. So having an interface is just like text that comes back each time. And there's no like, you know, nothing that shows you you're making progress, you know. So I do feel like people will integrate this technology and a lot more fit for purpose UI, you know, like last for us, right? We showed you okay, here are the steps. You did it right. Here are some options. And you can click. You can answer. You can. There are more like you know multimodal interaction to it than a simple chat. Chat is one way of doing it. But these are other ways to do that. So that's all bad. And the second would be like you need to understand the user and you need to personalize what's going on.
Jerome (00:17:13) - Right. So as you interact multiple times it's chat can be not starting from scratch. Right. You need to start really getting a deep understanding. We haven't really implemented that in the app, but this is really the future for us. Deep understanding of users and personalizing to what they are and who they are, you know.
Rod (00:17:28) - Right. So that's another step where it's like ChatGPT or I use Perplexity AI. It keeps your threads as long as you have an account, you can go back and continue with the thread.
Jerome (00:17:41) - It doesn't really learn across the thread. Right. So like what you did with it, one thread didn't really help the next friend to understand who you are, what you're trying to do. Obviously in a pure chat, I mean, China is very general, right? But in a learning environment where we can understand what you're trying to learn and keep that going. Right like that. You connect with us today or in a week, we can see how much progress you made.
Jerome (00:18:03) - We can see what you're trying to learn and keep the context going. Right?
Rod (00:18:07) - Yeah, of course, it reminds me of, you know, we've I've always been a proponent of learning management systems. Maybe the students don't like it, but some of the teachers don't like it either. But I helped to bring Blackboard into my institution. I worked for back in the late 90s, and they had either built it their own, their own technologies, or they often incorporated other technologies. I can certainly see having a plug in to a learning management system using LTI, where they would go into Sizzle and have some hooks so they could analyze, you know, the outcomes. Is that something that you're involved with yet or planned to do to sort of team up or make it make an interface, an API for learning management systems that might help you right now. Yeah, yeah, it just might help you with the analytics piece of it in terms of tracking progress and so on.
Jerome (00:19:11) - Mm. Yeah. So at the moment we are just direct students and direct learners. That's our go to market. We do feel that there is a big opportunity there because as this learner user product we can see where they are struggling, where they're having a problem, where they're working from. And so imagine like if your classroom was all using Sizzle, right. We could report back to the teacher, hey, for this class of problems, this is where your students are actually struggling. And they would not actually be very complicated for us to do that. And like we do have that information and we can analyze it. And there are amazing tools right now to be able to do that. The challenge for us is like, you know, how do we get plugged in? There's complexity in integrating. So I do see that as part of the future, especially because in our view right now, we may not even want to sell to school. We'd love to even give it for free to them.
Jerome (00:20:01) - Um, but, you know, we're not keen on doing complex integration or complex long sell cycles. Yeah, but I'd love to find a way to do this. I'd love to find a way to give that insight back to the teachers. And maybe they can just recommend us. That's the. That's the trade I would love to do right now, you know?
Rod (00:20:22) - Yeah, I can certainly agree with that because first of all, students, they move around a lot. They change their majors. They, you know, think that less than half of students that start at an institution end up graduating from there. You know, so having it personalized to that student that can follow them throughout their career is certainly a great way to think about it. Absolutely. Um, looking at the you know, I follow AI, I'm not follow it in depth in terms of the technology, but I'm just curious in terms of the large language model, do you, do you use something like ChatGPT you build your own models.
Rod (00:21:01) - How does that work?
Jerome (00:21:03) - Actually both. Right. So at the moment if you use the app, sometimes it touches GPT four, sometimes it touches there's our own model actually built you know by fine tuning some other models. So we have actually. You know, a good set of models right now, like almost like a dozen running for different tasks and even the problem solving, you know, some of that is rotted to own. So we're trying to learn to do this very well. And we say it's the state of the art right now in problem solving. But as you're looking at as we're collecting really interesting data, we're starting to deploy on things as well.
Rod (00:21:39) - Yeah, I remember trying with ChatGPT 3.5, it wasn't so good with some arithmetic coming. It would make mistakes. So I assume they've fixed that in 4.5 or whatever.
Jerome (00:21:54) - Right. So we use it right now and it's way better. 3.5 I would say my view at the moment is that you will benefit from having calculations done by real calculators.
Jerome (00:22:09) - You know, I think there are some simple tasks that a language model has a hard time doing, like arithmetics or even complaining, like floating points, for example, manipulation. And there is this notion of tool use which you can kind of do, in which it's not really well developed. And I think that's one of the future ways to make the system much more accurate, you know.
Rod (00:22:34) - Interesting. Yeah. It's hard for me. It's hard enough for most people to get their head around how these large language models work. I mean, I have a general idea in terms of adjusting all this text and so forth and predicting the next word and so forth, but integrating that with math and some high, you know, high end calculations. That interface seems. I just can't imagine how that works. I don't know if there's anything you can say to help me understand that a little bit better. How do you interface the two?
Jerome (00:23:08) - Yeah. So so today.
Jerome (00:23:08) - Right. You can do some arithmetic. I mean, if you ask simple arithmetic to GPT four, I will do it pretty well. And it happens because. Well, this is where there's things like if you start doing floating point operations, I think it will start seeing the limits pretty quickly. It's not a universal calculator. Right. But it's such a general understanding of a lot of the pieces of the world that many of these kinds of algorithms are embedded in that system. It's pretty amazing. By just reading all the information in the world, the system is able to kind of approximate and sometimes even implement some of these algorithms. But there's limits to this. Like it's just it's not a calculator and it will make mistakes, especially from things where it's a very detailed calculation. So what you can do is you can get the system to kind of know how to use a calculator. It's what's called tool use because they know the API to the calculator let's say, or to WolframAlpha for example.
Jerome (00:24:09) - Okay. And so they can learn to use that. And because an API call is a lot more like actual language manipulation, then instead of doing like oh pi divided by, you know, e the power of two, they will basically send that to the calculator and get the answer, you know.
Rod (00:24:29) - That helps me understand about WolframAlpha. And I know that sort of interface would really help move that along.
Jerome (00:24:37) - And remote from.
Jerome (00:24:37) - That is a very different, different technology, not a language model. It's much more like a set of like wrapped up rules that ultimately use a standard calculator in some steps at the bottom, you know. Right. And so it's usually right, but it's a lot less flexible.
Rod (00:24:54) - In looking briefly, I haven’t that much time to spend on the website. But for those out there that are interested in this, I was intrigued by the fact that I think one sample or one example I saw was that you, a student, could take a picture of a word problem from a book or their own handwriting, I don't know.
Rod (00:25:17) - And then your system would, would attempt to, to, to answer that. Is that correct? Is that possible?
Jerome (00:25:26) - Yeah, that's the main model. You know I mentioned that earlier. It's a bit like the model. You know, that photo method popularized a few years ago. So really the difficult thing is you come, you take a picture of the problem or you upload a picture and the system does OCR, you know, it does character recognition. It can even parse all the formulas out of it and then it solves it, or at least it gives you the steps to solve it and walk you step by step through it.
Rod (00:25:49) - Well, that is just intriguing. I would love to test it. I co-authored a book a years ago in Pharmacologic Calculations and. It Is fun to try to take a snapshot of some equations there and see what your system can do with it. But it's just fascinating.
Jerome (00:26:07) - And you'll see.
Jerome (00:26:08) - At the moment the system has the standard limitation, which it's actually pretty good at coming up with steps.
Jerome (00:26:13) - And we sort of step from our understanding, like around 90% relevant each time and pretty helpful. But it will get calculations wrong at times. You know, there's a good like 20% of problems where it does get some of the calculations here and they are wrong, but the steps are usually accurate. So you have to be careful. And that's something we're working on solving and making some progress on.
Rod (00:26:36) - Yeah.
Rod (00:26:37) - How do you validate this going forward. Do you have a team?
Jerome (00:26:43) - Some. We have benchmarks actually. So we do have benchmarks. So we have a set of problems both like a standard benchmark that exists out there and also like a set of problems. We have collected ourselves and and we do evaluate what we're doing against this benchmark and see what accuracy we get, you know?
Rod (00:27:00) - Yeah that's great. I bet there are listeners out there that can't wait to try their hand at this. And I'll just tell them it's Sizzle AI. But the URL, I really, really you're lucky to get a three letter URL szl.ai.
Rod (00:27:25) - So if you're looking, how do we get to.
Jerome (00:27:27) - This one of the shortest out there right. Yeah.
Rod (00:27:30) - That's great I really appreciate that. Oh the other thing I want to ask you about before we go, it looks like your logo looks like a Japanese character. Is that correct?
Jerome (00:27:40) - Yeah. I mean, it's the kanji character for fire.
Rod (00:27:45) - For fire. Okay.
Jerome (00:27:50) - We did, you know, bringing in a designer. This is actually how I came up with that logo and most likely we'll get replaced at some point.
Jerome (00:27:56) - So we're actually in the process of discussing that as a team, you know.
Rod (00:28:01) - Interesting. Yeah. Somebody saw that the Japanese version is all about recipes and how to cook maybe.
Jerome (00:28:10) - The name Is interesting. You know, the reason we picked this name is because, you know, this is about making learning a lot more fun and engaging and ultimately a lot more amazing. And this is about taking time from people who spend on social media today, right? It's like that, so our goal is really ambitious.
Jerome (00:28:27) - We want people to learn every day in a way that feels good to them and engaging. And so we're not really trying to compete against, you know, like kind of, you know, Coursera. We're trying to compete against TikTok and Instagram and, and Twitter for people's time, you know.
Rod (00:28:40) - Well, you’ll have to offer them badges, you know, as they complete different things, you know, to collect badges.
Jerome (00:28:48) - We want to get there at some point.
Rod (00:28:48) - Yes. Yeah, I'm sure you will.
Rod (00:28:50) - Let's run. This has been fascinating. It sounds like you have a wonderful product. I certainly wish you the best. I'm sure you certainly have the background for it, that's for sure. Is there anything you want to leave my audience with? Any. Anything even more exciting that's around the corner that you can.
Jerome (00:29:06) - No, no, I mean, just go try it out, you know, like in the AI. Or you can type Sizzle in in the App Store.
Jerome (00:29:13) - We do have a lot of things coming up, you know like so I think stay posted on one of their I mean the next feature will push in will be things where you can actually get feedback on your homework or how you completed it, which actually both students and teachers could use. Right. So you give it a solution and the system will tell you how well you have done. And then after that, we're going to really try to get you to commit to learning and come back to the app every day and give you a plan, you know, how do you master that skill or how do you solve this exercise on your own? You know, so a lot of exciting things in the future. But the app is functional right now and go try it and give us feedback.
Rod (00:29:49) - Exciting world we're living in. It changes so quickly.
Jerome (00:29:53) - Yes, yes. I mean it's the things, the opportunity things have to change. And I think education is ripe for it. You know, like the model of education hasn't changed in hundreds of years.
Jerome (00:30:06) - And I think we can do better. But that's not that I know how, but I think you can use you can learn how to do better. That's my philosophy, you know.
Rod (00:30:14) - Oh yeah.
Rod (00:30:15) - Oh well I'll certainly follow you and look for good things. So thanks so much again. All right.
Jerome (00:30:22) - Thank you so much for having me.
Rod (00:30:23) - Bye bye.
END OF TRANSCRIPT
*Podsqueeze was used to generate the transcript of my podcast. My links to Podsqueeze include referral links. If you decide to try it out using those links, I receive a small commission at no extra cost to you. This helps support the podcast, allowing me to continue providing valuable content.
Use this code: 10OFF3MONTH at Podsqueeze to get 10% off.
