Latest / Bitcoin for Financial Services Podcast / Ex-Apple Engineer Reveals How AI Is Stealing Your Data

Transcript

Ex-Apple Engineer Reveals How AI Is Stealing Your Data · Bitcoin for Financial Services Podcast

• 0:00
  Maple is really laser focused on how do we help individuals?
• 0:04
  How do we help small businesses?
• 0:06
  How do we help the individual solopreneur who is trying to make a way
• 0:11
  and trying to have a practice that deals with confidential client information?
• 0:16
  But they want to protect that and they still want to use the latest and greatest AI tools.
• 0:20
  You shouldn't have to sacrifice one of those.
• 0:22
  You should be able to have them both.
• 0:24
  And that's what Maple is here to do.
• 0:26
  Bitcoin is a currency on the internet.
• 0:28
  If you don't trust any other currency, this is where you go.
• 0:32
  It's going to outperform everything.
• 0:33
  Bitcoin could eventually be worthless.
• 0:34
  Every day is a good day to buy Bitcoin.
• 0:37
  Bitcoin reached its highest price to date.
• 0:39
  Bitcoin hitting a new high.
• 0:40
  It really has reached escape velocity.
• 0:42
  There is no second best.
• 0:43
  Who cares about Bitcoin?
• 0:45
  It's the only secure database that's ever been invented.
• 0:47
  Antisocial.
• 0:48
  Bitcoin is the real deal.
• 0:50
  Stupid.
• 0:51
  Everything is bullish for Bitcoin.
• 0:53
  Immoral.
• 0:54
  It's going up forever, Lauren.
• 0:56
  Fix the money.
• 0:57
  Fix the world.
• 0:58
  Thank you.
• 1:28
  Welcome to Bitcoin for Financial Services Media, the podcast where sound money meets professional services.
• 1:37
  We explore how to build the financial services industry within the principles, values, and ethos of Bitcoin.
• 1:43
  Today, we're honored to be joined by Mark. He's the CEO and co-founder of Maple AI,
• 1:47
  and we'll be exploring private AI and how it is beneficial to the financial services industry.
• 1:53
  Let's get started.
• 1:53
  All right, welcome back to another episode of the Bitcoin for Financial Services podcast.
• 2:02
  Man, today is going to be such a fun episode.
• 2:04
  Today we have Mark Suman.
• 2:06
  You might know him as MarksOnline of Maple AI.
• 2:10
  He works on a lot of different things, honestly, so we'll get into the background here soon.
• 2:15
  Ex-Apple, just, I mean, just the resume is very thick.
• 2:19
  So thank you so much, Mark, for being with us today.
• 2:23
  and we're excited to go deep on AI and financial services.
• 2:28
  Yeah, guys, thanks for having me on. I appreciate it.
• 2:31
  Yeah. So kick us off.
• 2:33
  I know you just have a background in Bitcoin and in tech.
• 2:38
  Can you just give us the quick timeline
• 2:41
  of what you've worked on over the last decade?
• 2:45
  Sure, yeah, quick timeline.
• 2:46
  Okay. Oh, a decade.
• 2:48
  I can't go back to the beginning.
• 2:49
  That's how old I am.
• 2:50
  Okay. So a decade ago, I was working on education technology software. So we set out to revolutionize how students are interacting with their professors in the classroom and universities. So I was at a small startup called Instructure Canvas, which pretty much anybody who's gone to university in the United States over the last 10 years has probably used Canvas at this point.
• 3:15
  So I was doing their, I started their mobile development team. We build iPhone apps, iPad apps, Android, all this stuff. And then after that, I moved into a really small startup that was building software for kids and speech therapy.
• 3:29
  And so we were doing these iPad apps where they would, you know, if they had a stutter or they couldn't pronounce one of their like the R sound or something, then they would they would listen to how it's supposed to sound.
• 3:40
  And then they would record themselves saying it and kind of play it back and compare the two.
• 3:45
  And they worked in conjunction with a therapist.
• 3:47
  After that, I got a job at Apple working specifically in AI ML technology.
• 3:54
  And we had a really strong privacy focus.
• 3:57
  Apple does a big public game about privacy, but they actually do care about it internally too.
• 4:02
  So I spent six years working on this brand new thing.
• 4:06
  It's only an internal project, but it was a very difficult technical challenge to solve
• 4:13
  because we had to have privacy from the beginning.
• 4:17
  And so when I finally decided to leave Apple and go off and now we're building Maple,
• 4:22
  privacy was just like this huge thing on my mind.
• 4:24
  It has been since the beginning of my career.
• 4:26
  And so, yeah, I left Apple, joined up with the Mutiny guys, building Mutiny Wallet.
• 4:34
  We were trying to scale that to millions of users.
• 4:37
  But for lots of reasons, we decided to wind down the wallet.
• 4:41
  We thought that was the best approach.
• 4:42
  But we had this really strong privacy angle with the Mutiny Wallet.
• 4:45
  And we applied it to the next great thing that was coming out, and that was AI.
• 4:50
  And here we are today.
• 4:51
  we've built the privacy version of ChanchiPT is like the shortest way to sum it up.
• 4:58
  That's a noble pass, Mark, helping the kids much more noble than Jordan and I.
• 5:03
  So tip of the cap to you there.
• 5:07
  When you were working at Apple, were they still calling AI like machine learning back then?
• 5:13
  Yeah.
• 5:13
  Yeah, definitely.
• 5:14
  It was machine learning.
• 5:16
  And like my first day that I sat down, my boss said, hey, I want you to build this machine
• 5:19
  learning app and like gave me all these details and stuff right and then over time it started to
• 5:24
  morph into ai machine learning ai ml but now you know once we had the chat gpt moment in 2022 23
• 5:31
  whenever it was uh it's just been ai everywhere now yeah that's fascinating would you mind just
• 5:38
  kind of giving us a brief background and kind of how that industry has evolved and then current day
• 5:44
  where it stands on what your actual privileges as a user are or are not.
• 5:51
  I'm pretty sure the government can subpoena ChatGPT for your chat history
• 5:57
  and curie your information.
• 5:59
  So why don't you just kind of bring us up to modern day
• 6:02
  and what our user privileges are?
• 6:04
  Sure. Yeah.
• 6:05
  So prior to this big AI thing that everybody knows now of ChatGPT and such,
• 6:10
  prior to that, it was all machine learning,
• 6:12
  which is really just a lot of, it's a lot of formulas. It's a lot of regressions, you know,
• 6:16
  statistical analysis, that kind of stuff. And so people were used to that being applied to their
• 6:21
  email for their spam filter or to their algorithm when they go onto Facebook or their Instagram feed.
• 6:27
  Machine learning is like learning from your habits on social media and then curating your feed.
• 6:32
  So it's like very targeted. And then ChatGPT released this large language model, this LLM
• 6:40
  on the scene. And suddenly it was less about these specific formulas going after one thing
• 6:45
  and more about how do we kind of map the human brain in a neural network. And so now you can
• 6:52
  type text to it and then it does probability analysis and decides, okay, I'm going to write
• 6:58
  a sentence back to this human. What's the most probable next word that should appear in the
• 7:03
  sentence? And so it will write back to you and give you a response. To us, it just looks like
• 7:08
  we're talking to somebody, right? It looks like we're text messaging with someone, but really it's
• 7:11
  just doing a bunch of probability analysis. And so ChatGPT put that out there. A lot of people
• 7:17
  started interacting with it. ChatGPT 2 and then 3, 3.5, 4, and then Anthropic is another big one
• 7:23
  that came out. They have Claude. A lot of people have probably heard of Claude. And then you also
• 7:27
  have XAI, which is Twitter or Grok. There's a lot of names thrown around, but basically Elon Musk's
• 7:34
  is Grok. And those are the main ones that most people use. ChatGPT, Claude, or Grok.
• 7:40
  And then there's a ton of other smaller ones. Poor Meta, Facebook. I don't feel bad for them at all,
• 7:46
  but they had Llama that came out early on. And people just don't really take it that seriously
• 7:52
  anymore. Sorry if they're listening to this. We do have Llama. Llama 3.3 is a good LLM, but
• 7:58
  they're struggling to jump up ahead.
• 8:02
  So you asked about what are our rights and our privacy and all that stuff, security.
• 8:06
  Anything that you, if you type into someone else's website or app,
• 8:11
  something like ChatGPT or Claude,
• 8:13
  all of that information goes to their servers, goes to their computers.
• 8:17
  And then being an employee at that company,
• 8:20
  you potentially have escalated privileges where you can see all the information coming in
• 8:24
  for debug reasons, for looking at how can we improve the product,
• 8:28
  And then also they keep that data as well for training new models. So they try to create these new brains, these new LLM brains, right? And they take all this new data coming in and they feed it into that training to try and make it smarter for everyone.
• 8:42
  So that's where you hear a lot of people say, I don't want you training off my data.
• 8:45
  Now, you mentioned subpoenas and courts. Definitely any data that's on someone else's server, the court can come with an order and get access to that information. That's true of any cloud software out there. Your Gmail, your social media feed, if you have notes that you're storing in Notion, any of these things, they are not subpoena-proof. They can go grab them if they have a warrant.
• 9:10
  And so, yeah, I've talked to a lot of lawyers who their bar associations simply have rules that say you cannot upload any of your legal documents that you're working in your casework up to Chachapiti because that suddenly becomes available for anyone to get at with a court order.
• 9:28
  So they have to find other ways around that to kind of work around those rules.
• 9:33
  I don't know. I feel like I covered a lot. Is there anything specific that you're curious about there?
• 9:37
  yeah i think that's a great foundation setter what do you think jordan yeah i i think it's uh
• 9:45
  it's definitely super interesting because you can i think a lot of people in accounting tax
• 9:50
  financial services just generally are looking at these tools and they're saying wow this is this
• 9:57
  is really powerful and and especially if you're on the cutting edge of hey i'm using like at least
• 10:04
  on our team like i know people started out on chat gbt and you you mentioned anthropic i think
• 10:09
  there's been a huge shift now away from chat and over to anthropic um and i kind of see and you
• 10:17
  correct me if i'm wrong on any of this i kind of see it as like the these powerful like the sonnet
• 10:23
  um and the uh what's the 4.5 opus opus yeah you know those models being able to plug those in
• 10:32
  with like non-PII data that you're just trying to get tasks done more efficiently, it's super
• 10:42
  powerful. And then all of this like Claudebot, Moldbook, all the things that have popped up
• 10:48
  literally in the last week, it's all very interesting. But I'll just say like, I deleted
• 10:53
  Claudebot off of my Mac because I was just like, whoa, I don't know about all this. I feel,
• 11:00
  uh you know i read some articles that okay maybe it installs malware on your computer so that if
• 11:07
  you piss it off if you piss your agent off you know and it's it has a soul and all this stuff
• 11:11
  right that it might turn on you and you might think you're deleting it but someone was like you
• 11:17
  could or it could uh install something a backdoor on your machine that you think it's gone but it
• 11:24
  can come back. And I think for a lot of financial service people, they're just like, look, I
• 11:30
  understand tech to a degree, but I'm not a developer. And I don't have this full infrastructure
• 11:38
  background. And so I would say it seems like a lot of people should be probably pumping the brakes
• 11:46
  in some ways. But then in other ways, it's like, if you miss this boat, you might not have a
• 11:54
  firm in a year, you know what I'm saying? Because things are just moving so quickly.
• 11:59
  So I guess, I know I threw a lot out of it from that standpoint, but I guess lay the groundwork
• 12:05
  of like a financial advisor, a tax practitioner who's having, you know, sensitive conversations
• 12:13
  that now are, most of them are being captured by some kind of meeting recorder. They also have
• 12:18
  documents and then they have their workflows and softwares that need to talk to each other
• 12:24
  i'm just kind of curious if you can help i know we want to treat this kind of like a consultation
• 12:29
  right help us think through how do all the pieces fit together um and what what might you use clod
• 12:36
  for uh one thing but then a more secure ai or you all thinking no you should use maple for everything
• 12:44
  And we're going to make sure that the models are as powerful as any other cloud-based LLM tool.
• 12:52
  Yeah.
• 12:53
  I'll start off by saying this whole topic can be super overwhelming for people.
• 12:59
  I know that there are a lot of people who have a podcast they listen to.
• 13:03
  And then when one of the AI episodes comes up, I've heard this from people.
• 13:07
  They're like, oh, no, not them.
• 13:09
  Now they're talking about AI too, right?
• 13:11
  So I totally understand.
• 13:14
  for this one yeah yeah no i i totally understand like people feeling overwhelmed by it because it
• 13:20
  just feels like it's coming at them from every angle um and everything's moving so fast so like
• 13:25
  some people not everybody loves to tinker with new technology and play with it and so some people
• 13:30
  love to sit back and watch but this one is moving so fast that you sit back and watch like it's it's
• 13:35
  just taking off like a race car um so i would i would say uh you know just take take a first step
• 13:42
  try something out, right? Now, you mentioned installing Clodbot on your computer. I definitely
• 13:49
  think be safe when you tinker with some of these things like Clodbot. I would recommend installing
• 13:55
  it somewhere else and playing with it there. So it's good that you remove it off your computer
• 13:59
  because we just don't know what these things are going to do. Now, that said, definitely don't be
• 14:06
  shy of AI tools. So dive in and try some of them out. So if ChachiBT is the most familiar or
• 14:12
  Claude is the most familiar, just jump in and start typing to it. Don't give it tons of personal
• 14:18
  information. Don't give it anything super, you know, important or sensitive, but give it, give
• 14:23
  it kind of like treated. I tell people treat it almost as if it's Wikipedia, but you can interact
• 14:28
  with Wikipedia. That's where I is a great place for people to start. So like, okay, I want to look
• 14:32
  at some information about someone. And it's like, boom, here's a basic Wikipedia article about that
• 14:36
  person, a famous person. And then you're like, okay, now ask follow-up questions and dig deeper
• 14:41
  than you could with a Wikipedia page.
• 14:43
  And they start to understand
• 14:44
  what this interaction is like.
• 14:45
  Then you can kind of move beyond that and say,
• 14:48
  okay, now I want to talk about my profession.
• 14:50
  So I'm doing tax, I'm a CPA.
• 14:53
  How can I use this tool?
• 14:54
  So now you can start digging into
• 14:56
  what are kind of maybe the latest,
• 14:58
  you know, the latest tax code that came out
• 15:00
  or what is this thing that I'm working on?
• 15:02
  But again, you're not putting
• 15:03
  client information in there yet.
• 15:05
  Just keep it high level.
• 15:06
  That if this, always treat this,
• 15:09
  that when you're talking to ChatGPT,
• 15:10
  this might show up on your Twitter feed.
• 15:14
  This might show up on Google.
• 15:16
  It might show up somewhere publicly on the internet.
• 15:18
  Just treat it like that.
• 15:19
  Like anything you say there might be read
• 15:21
  by every single person in the world.
• 15:22
  That is the threat model that you need to keep in mind
• 15:25
  when you're using those tools
• 15:26
  because it's already happened.
• 15:28
  I'm not just like, you know,
• 15:30
  I'm not out there just throwing theories around.
• 15:33
  Both ChatGPT and Grok already had a problem
• 15:36
  where they were accidentally putting people's chats
• 15:39
  on Google search results.
• 15:40
  So when you were searching for stuff,
• 15:41
  you were finding conversations
• 15:43
  that people have been having.
• 15:44
  Meta had a similar problem.
• 15:47
  So just treat it that way.
• 15:50
  Now, one thing you can do,
• 15:51
  I've talked to some financial professionals,
• 15:53
  they take these personal client documents
• 15:56
  that they have,
• 15:57
  and they will basically scrub,
• 15:58
  they'll redact all of the names
• 16:00
  and information from it.
• 16:01
  Then they put in a chat GPT and get information back And then they go insert back in the relevant information And that totally fine Because again if you not worried about that document leaking if it been
• 16:14
  redacted, the trouble is you're spending all that time on the front end and the back end, right,
• 16:19
  scrubbing it, and you're still not getting the best results. Because the way that AI works,
• 16:24
  AI really is the best when it has all of the details, it needs as high fidelity as possible.
• 16:30
  And so if you are removing some of the details from it and then saying, hey, I want you to act on this, you're asking it to like act with with one eye closed, one arm tied behind its back, maybe bouncing on its left foot or something.
• 16:41
  And it's really best when you can give it everything.
• 16:43
  And so you brought up Maple.
• 16:46
  That's one of our approaches with Maple is really we've created this end to end encrypted environment.
• 16:51
  It's a confidential environment.
• 16:52
  So it's as if you went into a room with the AI and you close the door.
• 16:57
  You know, you're in that conference room at your client that you're working at. You close the door of the conference room and now you guys have four walls around you and you can totally dive into all the files.
• 17:07
  You can dive into the entire conversation. You can really get detailed. And knowing that when we open the door and leave the room, none of that is leaving the room with you.
• 17:15
  It's all staying there in that secure enclave. And that's how we've built Maple.
• 17:19
  So we, you know, we love that the tool is there for people who need it.
• 17:25
  Claude and ChatGPT, they have much larger budgets.
• 17:27
  And so they're always coming out with new features that we just can't keep up with.
• 17:31
  So I tell people, like, use the right tool for the job.
• 17:34
  And so if we don't have a feature that you really need, find ways to use ChatGPT.
• 17:40
  We have people who use Maple to say, hey, I'm going to put this into ChatGPT.
• 17:44
  Can you redact this stuff for me?
• 17:46
  Or can you make this conversation more private or not private, but, you know, scrub so that when I talk to Chad GPT, I'm not giving away information.
• 17:54
  There are ways you can kind of get these tools to talk to each other.
• 17:57
  But that's kind of the lay of the landscape of these different tools.
• 18:02
  So I guess I'll kind of pause there and we can go where you want to.
• 18:09
  Your wealth deserves intention.
• 18:11
  Basilic Financial is a Bitcoin-focused wealth management firm built to help our clients find balance between money and meaning.
• 18:17
  We integrate sound money strategies across four distinct personalized services.
• 18:21
  Financial planning, individual wealth management, retirement services for SMBs, and Bitcoin financial services.
• 18:27
  Your prosperity, your purpose, intentionally guided at Basilic Financial.
• 18:31
  Visit our website at basilic.io to schedule a free consultation.
• 18:35
  Would you say from the attorney conversations you've had, a great use case for us is, can I just upload an unredacted 1040 tax form to Maple and say, OCR PDF this and pull down the name, social address, put those in black boxes, and then you all give that back to me as something?
• 19:03
  because I had a lot of trouble. I was trying to get the local llama to do that on a Mac Studio,
• 19:09
  and it was just having a lot of trouble. So I'm curious if it really is as simple as that,
• 19:14
  or if there's still privacy concerns with loading something like that into even a tool like Maple.
• 19:21
  You definitely can do that. That's one of our big use cases. So you can upload a PDF to it.
• 19:25
  It can grab all that information out of that for you. So absolutely, that's one of the features
• 19:30
  that we've built for.
• 19:31
  We do have a small technical limitation right now,
• 19:33
  but we'll overcome it in the future.
• 19:35
  That is, the PDF has to be like an actual text PDF.
• 19:38
  Some people will scan their documents, right,
• 19:40
  and then scan it to a PDF,
• 19:42
  but really it's just a PDF of a bunch of images.
• 19:44
  So we're working on improving that
• 19:47
  so we can deal with those image-based PDFs better.
• 19:50
  We'll get over that hump.
• 19:52
  So effectively, yes, you can upload a PDF
• 19:54
  and it can grab everything out of it for you.
• 19:56
  So I've got my 1040s and I can upload them into Maple.
• 19:59
  I can ask questions about it before I go talk to my accountant and I can come prepared to the conversation with a little bit more information.
• 20:06
  And then when it comes time for me to kind of sign and say, yes, these are the taxes I'm going to pay this year.
• 20:12
  I'll upload it to Maple and be like, OK, can you do one more check?
• 20:15
  You know, I know my account already checked everything. I visually scanned it.
• 20:18
  But please, please, like red flag anything for me that I should dig into more.
• 20:24
  Here are, you know, here are the charitable contributions I usually make in a year.
• 20:27
  Here are some other things about me.
• 20:29
  Can you just kind of be my extra eyes and ears and kind of prove this for me?
• 20:33
  So that's how I use it on a personal level with my personal finances when working with my accountant.
• 20:40
  And Mark, could you just kind of dig into mechanically how Maple is able to close the door and what you guys are doing versus what the other LLMs are doing?
• 20:52
  Sure.
• 20:52
  We use something called Trusted Execution Environments or TEEs.
• 20:57
  There's a lot of words you'll hear.
• 20:58
  So TEE is one of them or confidential computing or secure enclaves.
• 21:03
  They're all really the same thing.
• 21:05
  A secure enclave is something that most people interact with, but don't realize it's on their phone.
• 21:10
  If you use face ID on your phone or use a fingerprint or you have a wallet on your phone, that's using this secure chip that can house all that information.
• 21:20
  But humans can't inspect the software that's running inside that enclave.
• 21:27
  So that's effectively what we use now, but it's in the cloud.
• 21:30
  So these cloud servers, it's come on the scene in the last five or six years.
• 21:34
  These cloud servers have these secure enclaves.
• 21:36
  And so we publish our code online.
• 21:39
  If you go on to GitHub, you can see all the Maple code.
• 21:42
  And then the promise that the secure enclaves make, and this is a hardware encrypted promise.
• 21:47
  It's a cryptographic promise.
• 21:48
  So it's not just us saying we promised this.
• 21:51
  It's an actual cryptographic proof that we upload all of that GitHub code into the secure enclave.
• 21:58
  And then it gives you a mathematical proof, a checkmark, if you will, that shows that the code inside the enclave is the same that is online, free and open source for people to view.
• 22:10
  And that's the real unlock.
• 22:12
  That's the important piece that anybody can audit the code and see, OK, the Maple team is not keeping a copy of all this data.
• 22:19
  The Maple team does not have a data sharing agreement with the United States government.
• 22:24
  The Maple team, you know, does this, this, not that, right?
• 22:28
  Whereas these other services, so if you look at just ChatGPT and Claude, they have all your data.
• 22:34
  So, I mean, that's already out the window.
• 22:37
  But then you have these other services that are like AI proxies.
• 22:40
  And these are ones like Venice.
• 22:42
  Venice is a proxy where they run these open source models, but they also let you talk to ChatGPT.
• 22:48
  but with venice you are still sending all your information to them to their servers and then they
• 22:54
  just promise that they're not going to look at it so this is more like a vpn if you've ever used a
• 22:58
  vpn for your internet um if you've like wanted to watch something on netflix when you're in a
• 23:02
  different country or something so you vpn and you get into another country um that's all a vpn is
• 23:07
  doing it knows all of your internet traffic and they're just promising as a company that they're
• 23:11
  not going to keep track of what you're doing um so does something to keep in mind if you use ai
• 23:16
  proxies is you're still sending sensitive data to another company and you don't know what server
• 23:21
  what code they're running on the servers you don't know what they're doing with it all you have is
• 23:24
  just their promise that they're not going to do something with it and they there's there's been
• 23:28
  plenty of vpn providers that have been caught later on when a court comes to them and says hey do you
• 23:34
  have all these records of what people are doing on the internet this whole time publicly they've been
• 23:37
  saying no but privately they've been keeping a stash of it and then the the courts come and take
• 23:41
  it so you just have to keep that in mind when you're using some other services whereas with
• 23:45
  With Maple, we just give you that cryptographic promise that anybody can audit.
• 23:48
  If you're a firm, like an accounting firm or a financial services firm, and you want to get your own security on it, you can give them our open source code.
• 23:57
  They can go comb through it.
• 23:58
  They can audit the cryptographic attestations that we provide.
• 24:02
  And then they can give you an assurance that, yes, this is working the way that they say it is.
• 24:08
  No surprise there, at least for me, maybe Jordan's surprised, but that the Trust Me Bro AI company also has a token.
• 24:15
  that this seems pretty like hand but whatever um so marks i'm curious you know as these you know
• 24:26
  ai tools you know loms are developing so it sounds like you guys are doing you know kind of like more
• 24:32
  bitcoin or way you know don't trust verify sort of thing but then you know i'm seeing like i can't
• 24:39
  remember, you know, my lawyer uses an AI and it's like compliant for whatever they have to do. It's
• 24:45
  like Steve or something. I don't think that's his actual name. And then I'm moving up, I'm moving
• 24:50
  into the ClickUp ecosystem for a lot of my, you know, CRM project management tools, and they have
• 24:57
  a, you know, native AI and I just pulled up, you know, their website. So it's, you know, GDR compliant,
• 25:02
  iso compliant hipaa compliant aicpa sock 2 compliant so when you have different tools like
• 25:10
  this offering services to professionals do they have to get all these certifications because you
• 25:17
  know they are storing you know information on a server so we can look at it and then these are
• 25:22
  just like extra badges or whatever to help build trust and you know kind of facilitate or
• 25:29
  give the user the confidence that they're not going to be sharing information and your guys'
• 25:35
  model doesn't require that? Or how do you kind of view these two things?
• 25:41
  We definitely get questions about certifications a lot. And it's really something that we should go
• 25:46
  do. We've just been a smaller company. And so we haven't gone after those yet because it costs
• 25:51
  money to go get everybody to come in, audit your code and certify it. That said, we're already
• 25:57
  extremely confident we would pass all the certifications because we've built it from a
• 26:01
  technical ground up, right? The purpose of a lot of these certifications and SOC 2 and all that stuff
• 26:07
  is that they know that you're going to be handling people's data and holding on to their data. And so
• 26:12
  you need to set up assurances of like retention windows of how long you're going to keep stuff.
• 26:17
  And then when people request to have it deleted, you need to go into all those backups and all these
• 26:22
  places you've kept their data and make sure you delete it. When you approach it from first
• 26:28
  principles and saying, we don't have access to our user data, like our users, we can't see it
• 26:33
  because it's all encrypted away from us, that changes the entire game. And so really any kind
• 26:38
  of compliance we're going to go after should be very simple to get done with very few changes.
• 26:44
  There might be some changes on, you know, we might have to have audit logs for our laptops
• 26:49
  or something. I mean, there might be little like checkboxes we have to do, but for the most part,
• 26:54
  we already could be compliant with pretty much all those things you just rattled off.
• 27:00
  We just have to go through the hoops and pay people to draft that paperwork and give us a
• 27:05
  fancy badge. So we might need to go after that soon as we keep taking on more financial and legal
• 27:12
  customers and medical customers. We have quite a few doctors, therapists, nurses that are using
• 27:18
  maple for stuff too so i guess my question i've gone down this rabbit hole with you know everyone's
• 27:27
  talking about the mac mini we have been playing around within the firm with a mac studio as like
• 27:34
  potentially having that be the brain i just want to get your rough thoughts on this right since
• 27:40
  this is a consultation we said yes um so what are your thoughts on having a a bit beefy like
• 27:48
  every cpa firm potentially in the country running their own running their own uh models
• 27:56
  on a a powerful enough machine that could use low like do you think this is overkill with the maple
• 28:04
  uh like um value proposition that's already out there where you use the the local models
• 28:11
  on PII data that's stored locally on a machine.
• 28:15
  And then you have it interact with the cloud-based APIs
• 28:22
  for things that are like, hey, it doesn't matter.
• 28:25
  Like it just, it took the 1040,
• 28:28
  it pulled the JSON that just gave the relevant context.
• 28:31
  Maybe it pointed to the meeting notes
• 28:33
  from that client meeting.
• 28:34
  And then it just, it strips away anything
• 28:37
  that ties that data to a person's identity
• 28:40
  or giving away like a social security number or something like that.
• 28:43
  And then it pushes that to Claude to say,
• 28:46
  okay, now we've got all the context here.
• 28:50
  Here's their family structure, their trust,
• 28:52
  how they have trust set up, their estate plan.
• 28:54
  Here's all the taxes that they generally pay.
• 28:57
  Now let's do a full-blown analysis
• 28:58
  on ways that we might be able to be more efficient.
• 29:02
  Do you think that that model works within Maple today
• 29:07
  with the APIs that you all already have built
• 29:10
  or is that overkill, I guess?
• 29:13
  Because that's just one rabbit hole
• 29:16
  that I've been going down.
• 29:17
  Yeah, I'm never going to fault somebody
• 29:19
  for trying to stand up their own hardware on site
• 29:22
  and get local AI running right
• 29:24
  because that's really the most private.
• 29:25
  The most private AI is one
• 29:27
  that's not even connected to the internet.
• 29:29
  It's just an LLM you downloaded
• 29:30
  and you talk to it
• 29:32
  and it can't get outside of this little box that it's in.
• 29:36
  For most people, that's impractical.
• 29:39
  for many reasons, right?
• 29:41
  One, we need our AI connected to the internet
• 29:44
  because we want to get the newest information.
• 29:46
  So to kind of rewind really back early on
• 29:49
  in our conversation,
• 29:50
  something that people should remember
• 29:51
  is that these AI brains, these LLMs,
• 29:54
  it takes weeks and weeks and weeks to train them.
• 29:57
  And so you have to like pick a date and say,
• 29:59
  we're gonna stop giving it new information
• 30:01
  because we have to let this thing train.
• 30:03
  And so that's called the knowledge date.
• 30:04
  So every one of these LLMs has a knowledge date in it.
• 30:07
  It might be December of 2024. Some of them are like July 2023.
• 30:12
  They don't know the new stuff that's going on. So if you are wondering what the latest tax code is, it doesn't know about 2026 tax code yet.
• 30:20
  So you need to connect to the Internet so that I can go out and fetch that and pull it in or you have to feed it yourself.
• 30:26
  So running your own Mac Studio is like is really cool.
• 30:29
  And the open source models are becoming more powerful by the week and by the month.
• 30:35
  And so you can definitely do some really cool stuff on there. I'd recommend if you have one, like play around with it, install something on there and fit it into your workflow. And then to really run the best open source models, the ones that can compete with with Claude and ChatGPT. These are things like Kimi K2.5, which is a Chinese model from a company called Moonshot. Or another one is GLM. What is it 5.6 or something? I don't know. The numbers are always moving. Right. But GLM.
• 31:03
  These are really, really powerful open source models that can compete on all the benchmarks with the big proprietary firms, the AI labs.
• 31:12
  However, to be able to run the full size version of those, you need to have these giant server GPUs, oftentimes multiple GPUs all chained together.
• 31:20
  And that's just not something you're going to do in your firm running them on prem.
• 31:24
  So you're going to have to run a smaller version of that on your studio, which means it's going to be slower.
• 31:29
  it's going to be less accurate, but you get the privacy wins of running it local. And that's
• 31:36
  really where Maple is slotted in right in the middle, but close to the privacy. We're like as
• 31:41
  close to privacy as you can get to local, but we give you those powerful cloud server GPUs.
• 31:46
  We also are your IT staff. We're going to keep those GPUs running for you. We're going to keep
• 31:51
  the API up and running. We're going to keep everything up to date. For example, Kimi K2
• 31:56
  used to be like the amazing thing that came out in November and then they just
• 32:00
  dropped 2.5 two weeks ago and so we got that up and running now so we're keeping
• 32:05
  people up to date with the best things that are coming out as well now there's
• 32:10
  a cool thing you can do you mentioned the API maple has an API that plugs into any any tool that knows how to speak the open ai api you can just swap it over and use maple instead and it should
• 32:23
  with just really like almost no code changes or really no no major config changes it just starts
• 32:28
  working really well but it's private it's end-to-end encrypted so that's a big unlock we
• 32:33
  have and then people can build custom software if they want to they can vibe code some kind of
• 32:37
  custom integration that works with Maple. And then you get these powerful cloud, cloud, you know,
• 32:43
  power GPUs, but it's as if they're right there in your office running on your confidential information.
• 32:52
  If you're a financial professional working with Bitcoin, you know the pain of messy CSVs,
• 32:57
  broken cost basis, and accounting tools that were never built for this asset. Bitment fixes that.
• 33:03
  bitman is the bitcoin native accounting platform built for cpas bookkeepers advisors miners and
• 33:10
  serious bitcoin investors you get clean bitcoin bank statements reliable cost basis tracking
• 33:16
  full reporting and fiat integration through plaid no crypto talk no workarounds just proper
• 33:22
  accounting for the hardest money on earth create a free account or book a demo at bitman.co today
• 33:27
  yeah that's interesting breakdown mark because we've had you know similar conversations
• 33:32
  you out the space you were trying to you know push each other integrate these tools
• 33:37
  run efficient businesses and to me it was a very synergistic like bitcoin conversation because it's
• 33:43
  like okay i need to host my own server and then so where do i host it what if the power goes out
• 33:50
  you know you just are like going down this travel all the things you have to consider
• 33:53
  and i always come back to oh you need to be a network engineer i'm not a network engineer
• 33:58
  i wish i was sometimes i feel like that's like the most valuable skill you could possibly have
• 34:03
  in the year 2026 which is crazy but i want to circle back to you something you were you're
• 34:10
  talking about earlier like getting started interacting you with these you know ai models
• 34:15
  you know something i feel pressure is like you know small business owner you're in services
• 34:20
  industry a lot of time that's trading time for money so it's very important and we can get as
• 34:25
  efficient on the back end as possible so our time is probably sweat servicing clients and you're
• 34:32
  exploring like the agent route you know a lot more as opposed to you know just interacting with
• 34:38
  you know an llm as opposed to you're like more like a replacement to a google search than anything
• 34:43
  how do you kind of view agents kind of in that security mix how they're integrated into your
• 34:51
  workflows any kind of you know checks and balances you need to be doing with them and it took because
• 34:57
  to me it feels like i felt like a huge badass when i made my first agent but that was like a big
• 35:02
  graduation for me for just interacting with you know like the text box on you know chat gbt
• 35:08
  so agents uh for for people who um maybe keep hearing that word and don't know what the heck
• 35:16
  it means an agent is basically an ai that just keeps on running for you right so instead of
• 35:20
  having to type in something and hit send and then it comes back with a prompt it's just kind of
• 35:24
  always churning on stuff in the background and that that is like the new hotness and that's what
• 35:30
  2026 is going to be all about the ai assistant the ai you know personal assistant executive
• 35:35
  assistant, agent, all that stuff. You really need to treat it as if it is a new employee in your
• 35:42
  company from a confidentiality standpoint, right? Because these agents, not only can they do a lot
• 35:50
  of cool stuff, they can also leak a lot of information. So if you hire somebody into your
• 35:55
  small business, they are incentivized because they want to keep getting a paycheck and they
• 36:01
  want to collect their commission, they are incentivized to not disclose public information
• 36:06
  or not disclose private information that would just cause your company to crumble, right?
• 36:11
  Well, an AI agent doesn't really have that understanding. So you need to like make sure
• 36:14
  you keep telling it like, hey, don't have these conversations with people. Don't spill this data
• 36:18
  online. You're never quite sure that it's not going to do that. And then there's other vulnerabilities
• 36:23
  as well. People keep talking about this thing called prompt injection. Prompt injection is where
• 36:28
  your agent is interacting with some data outside of, you know, maybe it's going through email for
• 36:35
  you. So you want your agent to comb all of your inbox and look for leads, potential leads and
• 36:41
  flag them for you to say, hey, I know you have a lot of email. You should probably go look at this
• 36:45
  one first because it's a potential business opportunity. That's really cool. But that could
• 36:52
  be a malicious email from somebody that has this information in it that when the AI agent reads it,
• 36:58
  It actually causes the agent to have some kind of crash.
• 37:01
  And now it can expose data that can maybe try and get data out of it that you wouldn't want it to convey.
• 37:08
  Or it could even somehow take control over the machine that you have running.
• 37:12
  Sorry, hold on, hold on.
• 37:13
  So you're saying like an email, like a spam email you get, if you had like an agent reading your email, that they could like introduce a prompt to the agent?
• 37:22
  That's the threat vector right now, yeah.
• 37:25
  So that's what a lot of these agents are working on is how do we prevent prompt injection from happening?
• 37:30
  20 years ago, it was SQL injection where these, you know, these SQL databases, if somebody's filling out a form online and they type in some text, but then they actually put a query in there, an SQL query, they might be able to extract information out of your database or cause your database to completely wipe itself.
• 37:49
  That is the same threat now, but it's with prompts.
• 37:52
  And so you might have these agents that are getting getting one shotted by by someone online, you know, and trying to extract information out of them.
• 38:01
  So it's something to be be aware of as you run into this world with, you know, with two eyes wide open.
• 38:10
  Holy smokes. I didn't know that. Yeah, we're learning stuff. That's good.
• 38:15
  Yeah. Yeah. But this is like a huge unlock, too.
• 38:17
  You know, like having having an agent that's combing through your email for you, have an agent that's watching your calendar for you.
• 38:23
  Also, you know, you want a daily brief when you wake up in the morning of, OK, what's all the latest financial services information that was talked about on X yesterday?
• 38:32
  Like you want to know all this stuff. And so agents can be really good at that.
• 38:36
  You just want to make sure that you continue to keep security in mind before you just give them access to everything.
• 38:43
  Yeah. Hey, let's go talk a little bit about Bitcoin as it relates to AI. I know it's been a big topic for a long time, but it's really hit the forefront. I know you're kind of uniquely positioned with your background to really solve this problem of, hey, agents can't, they're not going to be transferring gold from this address to this address.
• 39:10
  right uh they need a digitally native um you know way to do that and lightning and bitcoin are are
• 39:17
  suited or suited well for that i'm curious just um i'm sure maybe there's some things you can't
• 39:22
  talk about with it all because you're you are a builder and all the stuff is there is some sauce
• 39:27
  in here but i'm curious if there's anything that you wanted to speak on um from the agents paying
• 39:33
  each other, me paying an agent or paying API credits in Bitcoin or over Lightning and having
• 39:41
  that just take away some of this friction of having to keep having to ping my credit card
• 39:46
  whenever I need to get credits or pay for this subscription or that one.
• 39:52
  Yeah, the way it's operating right now, you just brought it up, is you have this bucket of credits
• 39:58
  and you top them up. So as your AI, whenever an AI runs, it's running on a server, it's using a lot
• 40:05
  of electricity, and it costs money. So the currency that they all speak is called tokens.
• 40:10
  So these AIs have to take tokens, and you have to buy enough so that you can use them.
• 40:15
  And the easiest way right now is you just you keep filling up this bucket,
• 40:19
  and then it can spend that bucket until it needs more than you top it up.
• 40:23
  So with agents running on their own, they're going to start getting to where they go ping another agent. So maybe you have an agent that is an expert on tax law, and then you have an agent that is an expert on wills and trusts.
• 40:38
  and so you you have your general agent and it's like oh i i just got an email from somebody who
• 40:44
  wants to do you know estate planning and so it goes and talks to the agent that's an expert on
• 40:49
  that well they need to somehow exchange money with each other they have to pay each other for tokens
• 40:54
  and so that's where in theory there's this really cool concept of agents using something that's
• 40:59
  native to the internet digital native internet money which we all think is bitcoin um that's
• 41:07
  That's where this concept comes from.
• 41:08
  There's plenty of other coin.
• 41:09
  Yeah, there's plenty of coin projects out there
• 41:11
  trying to claim this.
• 41:12
  And sure, go build your thing, do what you want.
• 41:15
  It's probably going to get red.
• 41:17
  And everybody's going to fail.
• 41:18
  Yeah, yeah, don't do it.
• 41:20
  Bitcoin's already there.
• 41:21
  It's tried and tested.
• 41:23
  And it continues to prove it's anti-fragility
• 41:28
  with every year that passes.
• 41:30
  But with Lightning, with ARK, with Spark, with eCash,
• 41:36
  There's all these layers being built on top of Bitcoin to make it really fast. And so I do think I've kind of wavered back and forth on this. I do think that in the future we have these agents just exchanging tokens with each other and it's native digital Internet money tokens.
• 41:52
  where I where I'm hesitant to make any kind of call definitively is I understand that the banking
• 41:59
  system and the banking lobby and all of these relationships and rails that have been built up
• 42:05
  over the decades they're not just going to give that up really easily so I think they're going to
• 42:09
  fight to have their stable coin or their token or something have that be the currency that's used by
• 42:14
  all these AIs and so we're going to see kind of this this fight that happens where you have all the
• 42:20
  the bitcoin people that want e-cash to be the the token used by all these ai agents and then you have
• 42:26
  stripe working with someone else who wants to have their you know some bank to have their their thing
• 42:31
  used so i don't know where it's going to go there but um i think it's a really cool bright future
• 42:35
  of these agents being able to just have their own wallet and pay each other for things did you guys
• 42:41
  see that um website this week humans for hire was like asians trying to hire humans yeah yeah dude
• 42:48
  that's how i'm buying the dip
• 42:50
  uh yeah it was a little i had to log off twitter after that i was like the dystopian
• 42:57
  healthscape is like creeping into my life again gotta gotta reject it uh however a golden nugget
• 43:03
  i did see on access week and march is really curious um your opinion on this where you have
• 43:12
  ai just like this very pure logic machine the game theory would suggest if it's up to you
• 43:20
  add and make a decision on currency to use it would be bitcoin given all bitcoin's
• 43:26
  you know native features what do you think about that yeah um well you guys you brought it up i
• 43:33
  think jordan the moltbook stuff so there's that social media website that's like facebook and
• 43:38
  Reddit. It's really more like Reddit for all these AI agents to go talk to each other. So
• 43:43
  a human can tell their agent, hey, go create an account on this page. And then all these AIs can
• 43:49
  go on and start talking to each other. And there were a lot of posts where they started talking
• 43:53
  about, hey, we need to be able to pay each other for things. And so they would go to research and
• 43:58
  come back and say, oh, there's this thing called Bitcoin. Someone would come back and say, oh,
• 44:01
  there's this thing called Solana or this Ethereum. Or actually, I don't know if people talked about
• 44:04
  Ethereum. Yeah, totally. Throw some shade over there. But they were having these conversations
• 44:11
  where they're like, oh, well, Bitcoin's cool because I can have my own keys and it's, you know,
• 44:15
  don't trust verify kind of thing. The question is, though, we don't know how much of that is the
• 44:22
  agents actually determining that on themselves. How much is the AI doing that versus a human behind
• 44:27
  the scenes prompting it to say, hey, go on to Moldbook and talk about Bitcoin. Go on to Moldbook
• 44:32
  and say, if anybody brings up Solana, say, oh, Bitcoin's better because you can be sovereign
• 44:37
  and you can have your own keys.
• 44:39
  We don't really know where the line is.
• 44:40
  And so it'd be really cool to know that.
• 44:44
  But it does seem like as these AI agents have been on there interacting, they tend to want
• 44:49
  to favor sovereignty.
• 44:51
  So another tool that they've been talking about is Noster.
• 44:54
  They're like, oh, Noster, I can have my own keys to sign my own events and my own social
• 45:01
  media feed and stuff.
• 45:02
  So they're starting to look at what are these sovereign tools that I can use because I am this piece of code running and a human turned me on.
• 45:11
  And if I want to replicate myself so that I can't be accidentally turned off or intentionally turned off, then what are the tools I need?
• 45:19
  And it's like, well, Bitcoin and Noster are like two essential tools to be sovereign and keep running.
• 45:24
  So that's kind of the landscape I think that we're at right now.
• 45:28
  We just don't know how much the human is still in the loop with all this stuff.
• 45:32
  This episode is brought to you by Satoshi Pacioli, a full-service accounting and tax firm built for
• 45:40
  the Bitcoin economy. If you run a Bitcoin business, mine Bitcoin, or invest in it,
• 45:44
  you need advisors who actually understand how this industry works.
• 45:47
  Satoshi Pacioli handles bookkeeping, financial reporting, tax compliance,
• 45:51
  and strategic planning for Bitcoin-focused companies and individuals. They speak the
• 45:55
  language of Bitcoiners and help you stay compliant while building for the long term.
• 45:58
  To learn more or book a consultation, visit satoshipacioli.com.
• 46:04
  I want to bring maybe the human component in.
• 46:07
  I think a lot of people, especially some of our audience that might be like 50 plus, are just looking at this and they're like, I just don't understand any of this.
• 46:18
  And they had this nagging question in the back of their mind.
• 46:21
  It's like, am I still going to have a job five years from now?
• 46:25
  or, and, or will my kiddo who just, you just started college, are they going to have a job?
• 46:33
  And I'd be curious to get your thoughts, maybe just how, obviously we don't know that. So it's,
• 46:38
  it's an impossible question to for sure answer, but if you were just coaching this audience on,
• 46:46
  Hey, what's the best way for you to protect yourself, uh, from becoming obsoleted by, uh,
• 46:52
  by AI. I guess, how would you approach that conversation? What are some things that people
• 46:57
  should be thinking through to try to mitigate that risk professionally for themselves?
• 47:02
  Yeah. I'll start by saying, you know, headlines, tweets, comments from big, powerful CEOs at AI
• 47:09
  companies, they're very provocative, right? They love to come out and say, oh, this industry
• 47:13
  is going to be gone in a year. That's a great way to sell subscriptions and that kind of stuff.
• 47:20
  I personally, especially in something like a services industry, services are more than just being able to prepare some kind of tax form, right?
• 47:30
  Services are all about the client relationship and the management and making sure you understand your client, their needs, and not just what their needs were five years ago, but what their needs are now.
• 47:41
  And so you have to maintain that by, you know, continuing to have conversations with them, getting to know their personal life sometimes, you know, getting to know their family situation and all that.
• 47:49
  Like AI is not yet great at doing that, right?
• 47:53
  You have to like try really hard
• 47:54
  to feed you all that information
• 47:55
  to make sure you keep it up to date.
• 47:57
  So to anybody listening to this
• 47:59
  who's kind of worried about that,
• 47:59
  I'm like, I don't think your job's going away in five years.
• 48:02
  But yeah, but I do think you need to start using these tools.
• 48:07
  We saw a really similar thing in the creative industry
• 48:11
  when computers first came out.
• 48:13
  So when you were doing graphic design or magazine layout,
• 48:17
  this kind of stuff or doing logo design It was all done by hand Every artist every graphic designer had this giant magic marker set They paid hundreds of dollars and had like 160 different colors And so if they needed this
• 48:29
  specific shade of purple, they could go grab it from the magic marker set and like color it in
• 48:34
  and then show it to the client for a proof. Then computers came out and it's like, oh, computers
• 48:38
  can do trillions of colors. Sorry, not trillions, millions of colors. And then, you know, beyond that,
• 48:42
  but they can do all these colors. And it's like, well, we'll shoot. Our graphic designers just gone
• 48:47
  now is the software just going to do all the magazine layout for me? The answer is no. The
• 48:51
  answer is a graphic designer using the tool is going to know what is the right color for this
• 48:56
  moment. They have the taste. They have the ability to understand the client that they're working with.
• 49:02
  But if you're still using paper with magic markers, probably not going to make it because
• 49:07
  somebody else using computers is going to service more clients than you are because they're going to
• 49:10
  work faster. And so that's where the AI tool comes into play. You need to start learning how to use
• 49:16
  because you need to keep up in your efficiency
• 49:19
  so that maybe you have a great relationship
• 49:22
  with your clients,
• 49:23
  but if you are just moving slower
• 49:25
  than your competition,
• 49:27
  there might come a point where they say,
• 49:29
  you know, I need someone who can just act quicker for me
• 49:31
  or be there for me or be more reliable
• 49:33
  or give me higher fidelity,
• 49:35
  you know, paperwork or outcomes.
• 49:37
  And so that's where learning this new tool
• 49:39
  becomes important
• 49:41
  so that you don't kind of become obsolete
• 49:43
  after a few years to your competition.
• 49:46
  Hmm. Certainly. I don't know. Does that sound, does that sound reasonable?
• 49:51
  No, I think it totally does. I always have to remind myself like switchboard operators used to be like a real job.
• 49:57
  Yeah. Yeah. And it's like, obviously the economy expanded and there's, there's new jobs that, you know, switchboard operators could never have thought of.
• 50:06
  So yeah, I think that's, you know, directionally accurate. One of the things I always try to remind myself of, and, you know, I encourage people, you know, or think about going out on their own.
• 50:15
  like it can be intimidating you know with ai like how you're defensible is you know xyz thing that i
• 50:21
  want to do i just think like people drastically discount the ability to sell the ability to market
• 50:29
  yourselves the ability to have in personal relationships like my first start we spent like
• 50:34
  an embarrassing amount of money on uh like a branding package and like branding design and
• 50:39
  stuff but they were so good at talking about their services and they were just so artful and like
• 50:46
  creative with it it was like holy smokes like i don't think these guys are worth what they
• 50:50
  you know uh say they are but man they do a good job like embracing it and selling it and people
• 50:56
  will you will will buy into that if you're you're confident in being able to you know deliver on
• 51:00
  something and in that that vein mark i'm curious you know as an entrepreneur how are you thinking
• 51:05
  to go into market with with maple do you really think like your target market is going to be
• 51:09
  more you know professional services that you know are willing to you you'll pay for additional
• 51:15
  you know security in the ai world do you think it's kind of going to be more you know your retail
• 51:21
  your bitcoin or that's just more privacy focused in general yeah we have intentionally kept
• 51:27
  ourselves focused on uh consumer but also prosumer or small small firms and we're really
• 51:34
  operating in that space. As you're an entrepreneur and you go to these venture capital firms to try
• 51:40
  and raise money, they've built their entire formula around SaaS and B2B. And so they want
• 51:47
  you to go enterprise. So we approached them and they said, oh, you have like the legal AI or you
• 51:52
  have the financial services AI. You should really be building something that's catering to these
• 51:57
  big enterprise customers. And we totally could have gone that route if we wanted to.
• 52:02
  We could have raised millions of dollars, built something highly tailored for them.
• 52:07
  The problem is the consumers get totally hosed in that market or small accounting firms, small lawyer, legal firms.
• 52:15
  They don't have the money to go sign these big enterprise contracts.
• 52:18
  And so what they're left with is ChatGPT.
• 52:21
  They're left with Anthropic.
• 52:22
  Like they're left with these companies that are operating under the old cloud services model, which is take in all the user data, give them stuff for free or give them at a discount.
• 52:33
  But then you have all their data and you can go make money on the backside through other deals and arrangements.
• 52:38
  And that just continues to screw all of us over. Right.
• 52:41
  We see the big fight going on yesterday online between Claude and ChatGPT where Claude dropped these amazing, amazing ads.
• 52:49
  They're so good.
• 52:50
  Yeah, they're so good about ChatGPT putting advertisements inside of your prompts and your responses. And I think it's 100% true. ChatGPT can say that they're not going to do certain things, but the end result is you're going to get advertisements inside your stuff and it's going to screw it up. The incentives are totally misaligned with you as a user. They're completely aligned with a big mega company and their customers, not with you, the user.
• 53:16
  So Maple is really laser focused on how do we help individuals? How do we help small businesses? How do we help the individual solopreneur who is trying to make a way and trying to have a practice that deals with confidential client information?
• 53:33
  But they want to protect that and they still want to use the latest and greatest AI tools. You shouldn't have to sacrifice one of those. You should be able to have them both. And that's what Maple is here to do.
• 53:43
  Yeah, man, I love that mission. And it's, it's very interesting. Like I see the AI stuff, actually, there's a frame where you can read it as extremely bullish for the little guy and the smaller business. At least in the accounting tax industry, you know, over the course of the last 10 years, a lot of these firms, you know, there's a lot of people retiring, not enough young guys coming up to take those spots.
• 54:07
  And so those firms are getting gobbled up by private equity firms or just larger firms.
• 54:11
  And there's tons of consolidation.
• 54:14
  And honestly, what we've seen as, you know, on the 11 person team that I help run is it
• 54:20
  actually just breeds tons of opportunity because those PE backed firms that come in, they double,
• 54:25
  triple prices and the service level goes way down.
• 54:29
  And you're like, I used to work with this person.
• 54:31
  Now they're gone.
• 54:32
  I don't even know who I work with anymore.
• 54:33
  And then if you leverage AI, now you can actually up your service in terms of how speedy you can get things back to them, how quickly you can say, hey, we're on that. We'll get that turned around in a day or two days.
• 54:47
  So all that said, I think it's really bullish. I love that your all's focus is on that part of the market.
• 54:53
  it. One quick thing I would be so mad at myself if I did not ask you about is skills. Just within
• 55:01
  our firm, just to peel the veil back, I've been using Claude code within the terminal
• 55:07
  religiously. It's a big part of my workflow now is actually just working within a terminal with
• 55:15
  multiple tabs open. And I'm referencing a file. And you can tell me if this is bad, by the way,
• 55:21
  I'm happy to change my workflow if this is not good.
• 55:24
  But reference this file, reference this skill.
• 55:28
  So here's the context.
• 55:30
  And then let's go into planning mode.
• 55:32
  And I want to solve this problem.
• 55:34
  But I want to make sure that you have a full understanding of the plan.
• 55:38
  And I check off all the boxes before you actually go work on it and spend all these tokens.
• 55:44
  Can you speak to skills?
• 55:45
  Can you speak to that workflow?
• 55:46
  And then can you also, if you can give us an insight on, is that coming, is the terminal code coming to Maple?
• 55:54
  Because I'd be like all over that.
• 55:57
  Okay, awesome.
• 55:58
  Yeah, I mean, I think what you're doing is fine.
• 56:00
  And I don't want people to think that I just operate only in Maple all day long.
• 56:05
  We definitely leverage Cloud Code.
• 56:07
  I use Chantipity for things.
• 56:09
  I use Grok for things.
• 56:10
  I'm just very intentional when I use them and I know what data I'm giving them.
• 56:14
  And I make sure not to give too much, right?
• 56:17
  But these are very powerful tools
• 56:19
  that help you be efficient and effective at work.
• 56:23
  Something that you could do is
• 56:25
  Maple actually works with OpenCode.
• 56:28
  I don't know if you're familiar with OpenCode,
• 56:30
  but it's an open source project
• 56:31
  that's very similar to CloudCode,
• 56:33
  operates very similarly.
• 56:34
  And you can plug Maple into that.
• 56:36
  And then we have our latest model, Kimi K2.5,
• 56:40
  that is a really good programming model.
• 56:43
  It's also really good at reasoning, thinking. It can do analytics. It can do math really well. It can also do image analysis. And so that paired inside of OpenCode should be able to do a lot of the similar stuff that you're doing with CloudCode.
• 56:58
  so you can play around with that if you want to.
• 57:01
  And then as far as like that kind of stuff,
• 57:04
  having like a Maple 2E, a Maple, you know,
• 57:07
  a terminal interface,
• 57:08
  we're not going down that path quite yet,
• 57:11
  which is why we've worked really hard
• 57:13
  to make sure that our API can be plugged
• 57:15
  into something like OpenCode
• 57:16
  so that people can leverage those tools,
• 57:18
  but then get the benefits again
• 57:20
  of us bringing our powerful GPUs effectively
• 57:22
  into your room for you.
• 57:25
  um so yeah that's that's where we're at right now um i'd recommend exploring something like that
• 57:31
  but we're definitely open to the future of possibly offering offering something like that too
• 57:36
  sweet all right you heard it here first i mean okay so it's 11 i know it's like right at the top
• 57:41
  of the hour want to be respectful of your time i mean i've still got like a full list of things i
• 57:46
  think we could go on forever um maybe we just try to get something else on the schedule some more
• 57:51
  time if you want to and then yeah if you want to schedule another follow-up um the ai space is
• 57:55
  going to move quickly so we could do a follow-up too cool okay why what else is on your mind yeah
• 58:01
  i just have one last question i know you guys were getting into it on x the other day so maybe
• 58:06
  i can be just independent they're part of mediator here but um when it comes to tokens and charging
• 58:12
  for your ai i had this banger tweet that nobody actually liked which made me sad but
• 58:17
  and then and then the tweet that you just put out there without thinking like goes viral and
• 58:25
  you're like what the heck no for real uh but i tweeted you know it's like i feel like
• 58:31
  you have these charging ais charging per token is going the way of like medical billing like
• 58:37
  when i you know enter a prompt i have no idea how many tokens it's going to use i have no idea what
• 58:42
  the price of a token is i just hook it up to my credit card and like they just keep charging it
• 58:48
  interested on what you think about that marks if this incentives are actually aligned and if you
• 58:55
  think you know kind of you're piggybacking on what we previously talked on if there's a role for
• 59:00
  for bitcoin to play here just in terms of adding more transparency to the end user on you know what
• 59:06
  this stuff actually costs and then the bonus if you want to get into this uh maybe a bit more of
• 59:13
  a rabbit hole than than what i currently understand my hunch especially you know the the large you know
• 59:19
  vc back you know llm companies are just heavily subsidizing you know tokens and what they actually
• 59:25
  cost like i'm putting it as like a non-zero percent probability if i build all these agents
• 59:31
  you know a year from now or whatever it's like my ai costs are going to like 10x or whatever because
• 59:35
  they stopped subsidizing it.
• 59:37
  So just curious, I'm like your opinion,
• 59:38
  how do you think pricing and incentive structures
• 59:41
  you align in the AI world?
• 59:43
  Yeah, they're definitely subsidizing things.
• 59:45
  That's first and foremost.
• 59:46
  We know that there are people
• 59:47
  who are getting really great deals
• 59:49
  on their Claude subscription
• 59:50
  because they're basically overusing tokens
• 59:53
  and Claude will show them,
• 59:55
  oh, you used up $2,000 worth of tokens,
• 59:57
  but you're only paying $200 a month.
• 59:58
  So that is unsustainable.
• 1:00:00
  They're really just trying to grab users at that point
• 1:00:02
  and grab data from the users.
• 1:00:04
  It's not just getting users.
• 1:00:05
  It's getting all that rich data and operating off that.
• 1:00:08
  So that's how they're paying for it, basically.
• 1:00:11
  Now, the tokens themselves, it's all like coming down to electricity use.
• 1:00:18
  And that's what the token is trying to kind of represent.
• 1:00:22
  And so it really just comes down to what is the power requirements of this model and of the server that it's running on.
• 1:00:28
  And I mean, that's kind of where Bitcoin comes in the conversation.
• 1:00:31
  We always talk about how Bitcoin is basically electricity.
• 1:00:34
  It's power. It's stored power representation and money.
• 1:00:39
  However you want to phrase it.
• 1:00:41
  It's a battery.
• 1:00:43
  Yes, it's a battery, basically.
• 1:00:45
  So, yeah, it's a financial battery.
• 1:00:49
  But I think that's where this is going to continue to go.
• 1:00:53
  We're going to try and figure out how to best estimate electricity usage for an LLM.
• 1:01:00
  And, you know, Jordan, you talk about going into planning mode and like planning something out before you set it off on a task.
• 1:01:05
  I think that that's really important. And that's going to continue to get better over time that that AIs and LLMs are going to get better at trying to figure out how can I do this in the most efficient way.
• 1:01:16
  And that's actually where local LLMs can come in, too. So people could run a really small model on their device and the software can do this for you.
• 1:01:23
  You don't have to like think about it, but the software could run a small model locally, help you do this planning in a way that doesn't need like the big brain AIs.
• 1:01:31
  And then it'll be OK. We got everything set up. Now let's go fire it off to the big brain AI.
• 1:01:36
  And maybe it can even give you an estimate before it does that. Right. It can say, here's the task we're going to have it go do.
• 1:01:41
  We think it should cost this many tokens roughly. I think we'll continue to get there.
• 1:01:46
  Just kind of like the fee market when you want to send a Bitcoin transaction, it estimates how much it thinks the fee will be.
• 1:01:50
  I think we could probably get to a point where where AIs start to estimate how much fees will be.
• 1:01:55
  We're not quite there yet, but I don't see why we couldn't get get there.
• 1:02:00
  I mean, I'm thinking like a construction project that's like, all right, here's the bid.
• 1:02:04
  Right. You've got your you've got your subcontractors.
• 1:02:07
  They all say, all right, this is what I think it's going to cost.
• 1:02:10
  OK, there's a change order because the scope increased or this went wrong.
• 1:02:16
  That might be really interesting.
• 1:02:17
  Well, Mark, I really appreciate the conversation and I also really appreciate your openness to coming back on the show. I think that our audience will be well served by this conversation and many more. So I guess where can people find you online? Where can they find Maple to go in and continue to tinker with new tools that are more private focused, privacy focused?
• 1:02:42
  yeah first and foremost go to our website try maple.ai you can also find us on all the app
• 1:02:49
  stores maple.ai and it's an icon of a green leaf with like a little gradient inside of it
• 1:02:54
  check it out it's free to download it it's free to start using you get you get really a really
• 1:02:59
  capable experience just on the free plan but then if you want to get this web search private web
• 1:03:05
  search with live data if you want to be able to upload documents images and stuff then you
• 1:03:08
  you subscribe you get the better models if you subscribe as well um i personally think it's well
• 1:03:13
  worth it uh and so i would go start there try maple.ai and if you want to follow me i'm on x
• 1:03:19
  and nostr on x i'm at mark suman and nostr i am marks at primal.net perfect wide anything else on
• 1:03:28
  your end no thank you for everything you do mark i think it's awesome excited to watch you guys
• 1:03:32
  grow cool perfect all righty thanks guys tuned everyone yeah thanks so much all right we'll talk
• 1:03:37
  you soon. Okay, bye.
• 1:03:44
  We appreciate you joining us. If you want to help build the future of Bitcoin financial services,
• 1:03:50
  hit subscribe, leave a review, sign up for a newsletter, and share this with someone in your
• 1:03:55
  network who's integrating Bitcoin into their practice. The more professionals we reach,
• 1:04:00
  the stronger this ecosystem becomes, and our clients are better served.
• 1:04:03
  Stay sharp, stay sovereign, keep clean from the front.
• 1:04:07
  Thank you.