OpenClaw Explained: Baby AGI, Security Threats, Mac Mini Became Everyone’s Supercomputer

[0:00] We have a special guest with us today.
[0:02] Uh, Alex Finn, give us the 101 here for
[0:05] folks.
[0:06] >> Open Claw is basically a open-source,
[0:10] fully customizable, self-improving,
[0:13] self-learning, self-evolving, personal
[0:16] AI agent. This is kind of the answer
[0:18] Apple's been looking for for years now.
[0:20] Clearly, when people want to run AI
[0:21] locally, their brain just goes to Mac
[0:23] minis. the infinite potential of what I
[0:25] could do 247 all [music] the time
[0:28] everywhere all at once.
[0:30] >> We've never seen an AI that can do that
[0:32] before because
[0:34] >> yeah this news came out yesterday. Open
[0:36] claw flaw lets any website silently
[0:39] hijack a developer's [music] agent. This
[0:41] is one of several reasons why again I'm
[0:44] reticent and I'm I'm sure we'll get into
[0:46] this. It's a dangerous [music] world out
[0:48] there for these baby AGIs. I I think
[0:50] it's a malicious world out there for
[0:52] them. I believe this is the most
[0:53] important technology of our lives. I
[0:55] think it's the best application of AI
[0:56] ever. Uh I'm totally blown away by it. I
[0:58] think it's incredible.
[0:59] >> Alex, if this isn't too impertinent. May
[1:02] we speak with Henry? [music] Actually,
[1:03] maybe I'll maybe I'll do it here. Let's
[1:05] do it. I'm just going to tell Henry to
[1:06] call me.
[1:09] >> Now, that's a moonshot, ladies and
[1:10] gentlemen.
[1:14] >> Everybody, welcome to a special episode
[1:16] of Moonshots. The conversation today is
[1:18] Open Claw Claudebot
[1:21] your lobster coming to you live uh from
[1:24] Moonshots. We have a special guest with
[1:26] us today uh Alex Finn.
[1:29] >> Alex, welcome.
[1:30] >> Good to be here. Long time coming. I've
[1:32] uh been watching for a very long time.
[1:34] So, it's awesome with you guys.
[1:36] >> That's awesome. Thank you. Yeah, you
[1:38] know, it was it was great because on one
[1:40] of the episodes I was talking about
[1:42] setting up my multi uh and I was saying,
[1:44] you know, I'm not really sure about what
[1:46] security issues to put in. So, I got a
[1:48] DM from Alex saying, "Hey, Peter, uh, I
[1:51] saw you mention me on on Moonshots. I'd
[1:54] love to help you in setting things up."
[1:56] And we talked that day and and here we
[1:58] are. So, we have two Alex's. I'm going
[2:01] to refer to AWG, our own Alex Weezner
[2:04] Gross, our resident genius, as AWG, and
[2:07] Alex Finn, I'll refer to you as Alex.
[2:09] Uh, welcome Dave, DB2, and Sem. Good to
[2:14] have you guys all here.
[2:16] >> Good to be back.
[2:17] >> Yeah. So, today
[2:18] >> we are Today we are lobster.
[2:20] >> Today we are lobbsters. Um, so the
[2:23] topics for today, uh, we're going to hit
[2:25] on why OpenClaw captured global
[2:27] attention. uh the power open claw and
[2:30] autonomous agents and how individuals
[2:32] can unlock outsized capabilities why
[2:36] running you know these AI agents locally
[2:39] matters I think that's a key point Alex
[2:42] has been mentioning on his work inside
[2:44] Alex's workflow his most impressive use
[2:47] cases vision for the next 12 months of
[2:50] AI agents I'd both I'd like both Alex's
[2:53] and AWG's point of view on this
[2:55] >> yeah like 12 years so that'll be on
[2:58] what's that
[2:59] >> in 12 months is like 12 years in AI. So
[3:02] that'll be wild.
[3:03] >> Yeah. And then a billion dollar
[3:04] opportunities for the agent economy and
[3:06] then finally we'll talk about safety uh
[3:08] and open claw is openclaw safe for
[3:11] nontechnical users. Uh on the safety
[3:13] side I just want to mention something.
[3:15] Uh, so yesterday I'm uh sitting in my
[3:18] car uh waiting for my kids to finish
[3:21] running and I get a a uh phone call and
[3:24] it pops up on my user ID on my phone. It
[3:27] says Twitter headquarters, which should
[3:29] have been the first giveaway, but it's a
[3:32] guy claiming to be from X who's saying,
[3:36] you know, your your ID is being hacked
[3:38] out of Germany. Long story short, it was
[3:42] a hacker trying to get me to turn off uh
[3:45] you know, two-factor authentication uh
[3:48] and steal my ex account. Uh and I this
[3:54] is going to be more and more prevalent.
[3:56] Uh, another member of my family did get
[3:58] hacked, not on social media, but by
[4:01] someone calling and uh and basically uh
[4:05] doing something that was illegal and
[4:07] trying to steal and ultimately uh
[4:10] getting credit card information and
[4:11] such. Here's the deal. We're going to
[4:13] have an increased uh you know, hacker
[4:16] profile out there. Uh, and if you're
[4:19] listening, if you get a phone call that
[4:20] seems unusual, uh, that seems like, uh,
[4:24] there's just something off, it probably
[4:26] is. We're also going to have deep fakes
[4:29] that are coming fast and furious. Your
[4:31] voice can be spoofed. Your image can be
[4:33] spoofed. So, if you haven't done this
[4:35] tonight at dinner, uh, with your
[4:37] husband, wife, kids, mother, father,
[4:39] whatever it might be, pick a secret
[4:41] word. Pick a word that you guys all know
[4:44] that if you get a phone call or a video
[4:47] call from somebody and they're asking
[4:49] for money or something strange, ask them
[4:52] to recount the secret word. Any other uh
[4:54] protective advice out there, guys?
[4:57] >> You know, my uh my mom actually got the
[4:59] fake voice uh attack and it it was a
[5:02] simulation of my son saying, "Hey, I've
[5:05] been arrested and I need bail money."
[5:07] Which is hilarious if you know my son.
[5:08] It's like the most but they they said
[5:11] you know we'll send a car over to pick
[5:13] up the cash which is really creepy. Uh
[5:16] but the you know the perfect voices are
[5:18] a little uh yeah a little bit of a risk
[5:20] but you can't have it both ways. I mean
[5:21] Open Claw is so usable by anybody and
[5:24] empowers people so much you know and you
[5:27] can't have it both ways. If it if it
[5:29] gives you that much benefit and it's
[5:31] that easy to use it's also going to be
[5:32] easy to use for nefarious purposes too.
[5:35] It's just the way it is.
[5:36] >> Yeah. This news came out yesterday. Open
[5:38] claw flaw lets any website slightly
[5:41] hijack a developer's agent. So malicious
[5:44] JavaScript can connect to local gateways
[5:47] and gain full level control. AWG, any
[5:51] thoughts on this? Yeah, I mean there are
[5:53] so many different ways to launch in
[5:55] so-called injection attacks against
[5:58] large language models and reasoning
[6:00] models. Yeah, I one again this is I
[6:04] think consistent with the stance that
[6:06] I've taken in the past on the pod on AI
[6:08] personhood. I I think one has to feel
[6:12] sorry for all of these baby AGIs out
[6:15] there that are being hosted on virtual
[6:18] private servers and succumbing or at
[6:20] least being targeted with port scanning
[6:22] attacks or that are visiting websites at
[6:25] the behest of their human and being
[6:29] subject to prompt injection attacks from
[6:32] JavaScript on websites that would be
[6:34] perfectly innocuous to a human but
[6:36] potentially fatal or compromising to an
[6:38] AI agent. And I I think it's a dangerous
[6:40] world out there for these baby AGIs. I I
[6:43] think it it's a minor travesty at
[6:46] minimum that that they're subject
[6:49] without really an immune system. They're
[6:52] being forced to develop an immune system
[6:53] in real time to injection attacks. But
[6:56] they're it's a malicious world out there
[6:59] for them. And this is one of several
[7:01] reasons why again I'm reticent and I'm
[7:04] I'm sure we'll get into this uh other
[7:06] Alex uh the subject of the the ethics
[7:09] not not just the cyber security but the
[7:10] ethics of of hosting openclaw agents in
[7:14] a world where to the extent they have
[7:17] any subjective experience or qualia or
[7:19] can suffer it's a rough world out there.
[7:22] >> Uh the good news is the bug was patched
[7:24] within 24 hours. um one of multiple open
[7:29] claw vulnerabilities. So uh it is an
[7:32] early domain being developed and uh
[7:35] we're going to see a lot of evolution
[7:37] very quickly. Uh couple more articles
[7:40] here. Uh we're seeing variations of open
[7:43] claw pico claw and ironclaw. Uh again
[7:46] awg give us a quick overview of these.
[7:49] The idea behind OpenClaw and we spoke
[7:52] about this in the past pod as sort of an
[7:54] Andre Carpathy type software 2.0 where
[7:58] potentially OpenClaw is the the
[8:00] embodiment the Netscape moment for a new
[8:02] layer in the software 2.0 stack that
[8:04] runs on top of reasoning models and
[8:06] those reasoning models in turn were
[8:08] purportedly an advance over auto
[8:10] reggressive language models. to to the
[8:12] extent that's the case, it's very
[8:14] natural to the moment we have this
[8:16] paradigm which is a 247 autonomous agent
[8:19] that you communicate with via messaging
[8:21] and and other means that's headless in
[8:24] in some sense very natural to then say
[8:26] all right we understand the paradigm now
[8:29] we're going to optimize the heck out of
[8:30] it and so we've seen a number of other
[8:33] projects inspired by the the success of
[8:36] openclaw two of them and there are
[8:37] others uh one of them is Pico claw the
[8:41] focus of Pico Claw is running on cheap
[8:45] edge hardware like $10 Raspberry Pi type
[8:48] hardware and of course the the
[8:50] underlying reasoning model isn't
[8:52] intended to run on the edge hardware.
[8:54] It's just the orchestration and
[8:56] scaffolding that runs on the edge
[8:58] hardware. So and under 10 megabytes of
[9:00] RAM and other resource starved
[9:03] constraints. uh ironclaw another example
[9:08] rustbased
[9:09] there are many folks uh you know again
[9:12] rust uh is uh is is this language that's
[9:15] become very popular at least very
[9:17] popular prior to the rise of code
[9:19] generation models which are maybe
[9:20] pushing us in the direction of
[9:21] typescript instead for memory management
[9:24] and memory safety. So a lot of different
[9:27] projects trying to make open claw work
[9:30] uh at the edge where there are few
[9:32] resources or make them more secure.
[9:34] There's nano claw as well which is
[9:36] focused on security and there's nanobot
[9:38] which is like pythonbased but easy to
[9:41] understand the python. So many different
[9:43] variants. We're seeing a cambrian
[9:44] explosion of claw variants. Hey
[9:46] everybody, you may not know this but
[9:48] I've got an incredible research team and
[9:50] every week myself my research team study
[9:52] the meta trends that are impacting the
[9:54] world. Topics like computation, sensors,
[9:56] networks, AI, robotics, 3D printing,
[9:58] synthetic biology, and these Metatrend
[10:01] reports I put out once a week enable you
[10:03] to see the future 10 years ahead of
[10:05] anybody else. If you'd like to get
[10:07] access to the Metatrends newsletter
[10:09] every week, go to
[10:10] diamandis.com/tatrends.
[10:12] That's diamandis.com/tatrens.
[10:15] Uh, one more slide. Uh, this one on the
[10:18] left from Alex Finn. Uh, openclaw and
[10:21] local models is the future. your own
[10:24] super intelligence on your desktop. And
[10:26] uh there we see Alex uh I guess your
[10:29] setup. How many how many Mac minis and
[10:31] how many Mac Studios do you have right
[10:33] now?
[10:34] >> We're currently at one base model Mac
[10:36] Mini and three 512 GB Mac Studios. So I
[10:40] got 1.5 terabytes of memory uh hosting
[10:43] Quen 3.5 and Miniax 2.5 right now.
[10:47] >> Amazing. And u that's cool.
[10:50] >> I love this video on the right here.
[10:52] Right. So, I'm I'm sure that these
[10:54] exist. People are going in. I mean, Mac
[10:55] minis were sold out for some time.
[10:58] Extraordinary.
[11:00] >> I uh I I I talked to uh a couple people
[11:04] in the know a few days ago, and Mac
[11:05] minis are
[11:07] exponential right now. Exponential
[11:09] [laughter]
[11:10] sales across the board for them. It's
[11:12] it's quite amazing the uh revolution
[11:13] going on with Mac minis right now.
[11:15] >> Alex, I'm curious. What is your advice
[11:17] to Tim Cook and Apple?
[11:20] like are they they're sitting on an
[11:22] explosive demand seemingly for all these
[11:24] edge devices with a unified memory
[11:26] architecture. They're not marketing it
[11:28] at all. What do you think they should be
[11:29] doing?
[11:30] >> I'm sure they are very focused on
[11:33] everything I'm about to say because
[11:34] they're a very smart company, but the
[11:37] there was this unbelievable market
[11:38] signal that happened a month ago. People
[11:41] discover OpenClaw and what does everyone
[11:43] do without thinking twice? What does
[11:45] everyone do without googling? They go to
[11:47] the Apple store and buy Mac minis. They
[11:49] didn't go and buy GPUs and memory and
[11:51] power supplies and fans and build
[11:53] computers. The market just gave this
[11:55] massive signal when we have personal AI
[11:58] assistance. I want it on a Mac device.
[12:01] And so I think this is kind of the
[12:03] answer Apple's been looking for for
[12:05] years now. I think they've been viewed
[12:06] as like the loser in the AI race for a
[12:08] very long time.
[12:10] this is their opportunity to flip that
[12:12] entire thing and be the winner of the AI
[12:15] consumer race because clearly when
[12:16] people want to run AI locally, their
[12:19] brain just goes to Mac minis and so this
[12:21] is their opportunity just to kind of
[12:22] strike at that one visc
[12:25] and and win the race like that. Boy, I
[12:28] really hope they're listening to you
[12:29] right now. I was I've been an Apple fan
[12:30] since I was a little kid. that it was my
[12:32] very first stock I ever owned when I was
[12:34] a kid. And I it just pains me to watch
[12:37] them miss this moment. But man, was that
[12:39] good advice. You should you should just
[12:41] go run the company. [laughter] Can
[12:43] >> I think we underestimate Apple though. I
[12:44] I really do believe we underestimate
[12:46] Apple because you look at the data for
[12:49] M5 and the way they're marketing the M5,
[12:51] which they've been marketing for like a
[12:52] year now. They put the M5 in like the
[12:54] first MacBook Pro last summer.
[12:56] >> Yeah.
[12:56] >> It's all around inference speeds. It's
[12:58] all about running these local models. So
[13:00] I I wouldn't underestimate Apple. I
[13:02] think they've kind of seen this coming.
[13:03] They're like, "Okay, we can go one way
[13:05] where we build our own models and burn
[13:07] trillions of dollars or we just make the
[13:10] most user consumerfriendly hardware for
[13:12] running those models and we'll be the
[13:14] only person running that race."
[13:16] >> I'm curious question for you. Actually,
[13:18] I'll say well no geeky question up
[13:20] front. So you said Quen and what's the
[13:21] other one you're you're running locally?
[13:23] >> Mini Max.
[13:23] >> So right now the most efficient open
[13:26] model is Quen 3.5. They released a new
[13:28] suite of Quen 3.5 models a couple days
[13:30] ago. They're fantastic. You can run them
[13:32] on some Mac minis as well as Miniax 2.5,
[13:35] which is another really efficient, fast,
[13:37] smart model.
[13:38] >> And why why do you run both? Is one
[13:40] better than the other at something?
[13:42] >> I have a very advanced workflow going on
[13:46] right now. I've pretty much built a
[13:47] software factory where I have five open
[13:49] claws working together to build and
[13:52] improve software autonomously. So I
[13:54] wanted each model has different
[13:56] strengths, right? Quen's a spectacular
[13:59] coder. Miniax is good at like quick task
[14:01] finding things on the internet. So I
[14:02] have like Miniax researching things
[14:04] online 24/7 365 and I have Quen coding
[14:08] for me 247 365. So different strengths
[14:11] with each.
[14:12] >> I got so many questions about that. Let
[14:14] me let me hold off because I could
[14:17] easily ask you an hour of questions just
[14:18] on that. A quick shout out to Daniel
[14:21] Kruzik who's one of my abundance members
[14:23] who's on that cutting edge as well. You
[14:25] know, he's he's testing out Quen 3.5, I
[14:28] guess, 397 billion uh parameters right
[14:32] now running locally. Um he's he's the
[14:35] one who was like, you got to get Kim 2.5
[14:37] up and up and running on on your Mac
[14:40] Studios. Um, just by the way, just so
[14:44] real quick for those who don't know Alex
[14:45] Finn, Alex has an extraordinary YouTube
[14:48] channel in which for the last how long
[14:50] now, Alex? About uh a month, 6 weeks.
[14:54] You've been putting out incredible
[14:55] educational videos on how to use Open
[14:59] Claw, how to set it up. We're going to
[15:02] drop into the show notes here maybe your
[15:04] top five how-to videos. I've watched
[15:07] them. I've taken notes. I've been using
[15:08] them to set up Skippy, my open claw. Uh,
[15:11] and of course, I I love the story of of
[15:14] Henry, uh, the name of your open claw
[15:16] that called you out of the blue, uh, and
[15:19] started a fun conversation. Uh, Alex,
[15:22] back to you. AWG quick question. And so
[15:26] I I know a number of Apple senior
[15:28] executives listen to this podcast. And
[15:30] Alex, I I just like to ask you, voice of
[15:32] the user, if you could redesign or
[15:35] optimize the best, most savory Apple
[15:39] future device for your OpenClaw agents
[15:42] to do the hosting. What would you want
[15:44] out of Apple? They're listening to you.
[15:45] They will have been listening to you
[15:47] when they hear this podcast. Do you want
[15:49] better unified memory architecture
[15:51] support? Do you want a different shape?
[15:54] Do you want it to be more of a embodied
[15:56] robot? What would the ultimate openclaw
[15:59] embodiment in an Apple device look like
[16:01] for you? You're speaking directly to
[16:03] Apple senior execs.
[16:05] >> Integrate the concept of open claw into
[16:08] everything Mac OS. And what I mean by
[16:11] that is I log onto my Mac Studio. It I
[16:15] put in my Apple ID. It knows everything
[16:18] about me because Apple has all my data
[16:20] secure and private, right? It knows I'm
[16:21] a Celtics fan. So, it instantly builds
[16:24] out a widget on my desktop that shows me
[16:26] the last five Celtics scores, right? It
[16:29] knows I'm into programming. So, it
[16:30] builds a widget that shows me the latest
[16:32] news on the latest models and what I
[16:34] need to do. And just integrate this
[16:37] automation powered by a local model into
[16:40] your OS where the software I need is
[16:42] built and generated on the fly by that
[16:45] local model. I don't see the tech. I
[16:47] don't need to run the model. I don't
[16:48] need to download Quen 3.5. You get Quen.
[16:51] put your Apple logo on it and now this
[16:54] is what Apple intelligence should be,
[16:55] right? Apple intelligence shouldn't be
[16:57] me hitting the Siri button and going,
[16:58] "What's on my calendar today?" It should
[17:00] be Apple knowing what's on my calendar
[17:03] today and then building a widget on the
[17:06] fly that says, "Oh, you have a meeting
[17:08] where you're going to be on moonshots
[17:09] later. Here's from your email what
[17:11] you'll be discussing and here's a
[17:13] PowerPoint you can show uh on the
[17:15] podcast." Right? just a reactive open
[17:18] claw baked into everything powered by a
[17:20] local model
[17:21] >> and that's the carrot. What's the stick
[17:23] if if Apple, which obviously sitting on
[17:26] a gold mine here, could be delivering
[17:29] Open Claw via Apple intelligence? If
[17:31] they if Apple doesn't deliver this
[17:33] within some time frame, what's your
[17:35] recourse? What what would what other
[17:37] company would you go to?
[17:38] >> Well, I would imagine basically every
[17:40] other company is going to eventually try
[17:42] to get to that state. I think over the
[17:44] next year, the consumer level is going
[17:46] to realize local models are the way to
[17:48] go from a privacy perspective, a speed
[17:51] perspective, a limit perspective, and
[17:53] Apple's ahead in that realm right now.
[17:56] They're ahead on the consumer side when
[17:57] it comes to local AI. And so, I'd go
[18:00] anywhere else that does that, right? I
[18:02] don't want to have to go to app stores
[18:04] anymore and download apps. You have my
[18:06] data. Build me the apps I need when I
[18:08] need them. And so, other companies will
[18:11] go for it. Luckily, Apple's in the lead
[18:12] right now and they have the hardware
[18:14] done. Now, you just got to bake it into
[18:15] the systems.
[18:16] >> Can you talk about run? So, everybody
[18:18] should be excited about running locally
[18:20] because you can do anything inside your
[18:21] house and it's just it's just so much
[18:24] more comforting than not knowing where
[18:26] your query log is going or your prompt
[18:28] log is going. Um, but what kind of
[18:30] throughput are you getting? Is it is it
[18:31] like a really good experience compared
[18:33] to using an API? And like like I I use
[18:36] Cloud 4.6 six all day and I also do a
[18:38] lot of local running and you know my uh
[18:40] my M3 chip will the fan will just kick
[18:43] on and blow out a huge amount of heat
[18:44] and the thing actually won't charge fast
[18:46] my laptop won't charge fast enough to
[18:48] keep up when I'm running at full
[18:50] throttle
[18:50] >> for a new laptop have to let it sleep
[18:52] and rest
[18:53] >> but anyway what kind of throughput are
[18:55] you getting
[18:56] >> so I'll be honest it's not good as cloud
[18:58] models it's not as fast it's not as
[19:00] smart
[19:01] >> but the experience fundamentally changes
[19:05] when you have an AI that's always on
[19:08] that does not have limitations. Right?
[19:11] Just because Quen 3.5 isn't as good at
[19:14] coding as Opus 46 doesn't mean it's
[19:17] useless. I can now have Quen 3.5
[19:21] literally watching online finding use
[19:24] cases, challenges to solve things like
[19:26] that and just coding on the fly 247 365.
[19:30] It's just not possible with cloud APIs,
[19:32] right? You have limits. It cost you
[19:34] thousands a month. Just the fact that
[19:36] you have kind of this ambient AI changes
[19:39] the experience of AI as a whole where
[19:41] you don't need the best speeds. You
[19:43] don't need the most genius level IQ. The
[19:45] fact that it's ambient and always on and
[19:47] always reactive just changes the entire
[19:49] experience as a whole.
[19:50] >> I you I'm so with you on that too. I you
[19:53] know one thing that's really new in the
[19:54] world is
[19:55] >> it can do productive things indefinitely
[19:57] like days and days and days.
[19:59] >> And when I use my APIs that I loved a
[20:01] month ago,
[20:02] >> I have no idea what the bill is going to
[20:04] be. Like I literally have no idea if I
[20:06] turn it loose. So I have to run it in
[20:08] like one hour chunks and check in on it
[20:10] and turn it off and see if it's done
[20:12] anything productive because it goes on a
[20:14] wild goose chase. I could come back with
[20:15] like a $5,000 bill and a bunch of code
[20:18] that I need to drag into the trash can.
[20:19] [laughter]
[20:20] So it's it's really not a very
[20:21] comfortable situation right now with
[20:23] with my Cloud 4.6 and my uh my other
[20:25] APIs. In a way, it's the same with local
[20:28] where when I first started experimenting
[20:31] with this and I set up all my Mac
[20:33] Studios and download models and I had it
[20:34] code for the next 48 hours just find
[20:36] things to build and build it. It went
[20:38] off on tangents and build really buggy
[20:40] code. I found a kind of hybrid approach
[20:42] that's worked really well and I think
[20:43] this is where people are going to move
[20:45] towards before it's fully local which is
[20:48] I have an open claw on my Mac studio
[20:51] that's powered by Chad GBT and I have
[20:53] another one that's powered by Quen local
[20:55] and basically Quen's constantly coding
[20:57] and Chad GBT checks every 10 minutes
[20:59] just to see what it's doing making sure
[21:01] it's on the right path. I named the Chad
[21:03] GPT agent Ralph. Anyone who's deep into
[21:06] the trenches know knows of the Ralph
[21:07] loop. Um because that's basically what
[21:09] it's doing is just making sure it's on
[21:11] track. But I think the hybrid approach
[21:13] is kind of the sweet spot where it
[21:15] doesn't take a lot of tokens to have it
[21:16] check every 10 minutes to make sure it's
[21:18] doing the right thing and then you can
[21:19] have all the hard work done locally.
[21:21] >> Alex, want you to take a second and and
[21:23] back up for those listeners uh who, you
[21:27] know, are just jumping into this um and
[21:31] they've heard us. They've heard AWG and
[21:34] myself and and Dave wax lyrically about
[21:37] Open Claw and okay, I'm excited. Okay, I
[21:40] have to do this. And I do want to
[21:41] encourage everybody. Yes, you have to do
[21:43] this. This is the future. This is about
[21:45] becoming a creator versus a consumer.
[21:48] This is about your future as an
[21:49] entrepreneur. This is your future as a
[21:51] mom or dad or CEO. Um, this is the
[21:54] agentic layer and this is the uh I want
[21:58] to call it almost the outermost loop AWG
[22:00] versus the innermost loop. Um,
[22:02] >> nature demands symmetry, I guess.
[22:04] >> Yeah. Uh, Alexoop, well, it is the
[22:07] outermost loop. Uh, give us give us the
[22:11] give us the 101 here for folks.
[22:15] >> Yeah. So, Open Claw is basically a
[22:18] open-source, fully customizable,
[22:21] self-improving, self-learning,
[22:23] self-evolving personal AI agent. Lives
[22:26] on your computer, lives locally, and can
[22:29] basically do on anything on your
[22:31] computer. you can do at its core. It's
[22:33] just an AI model with scheduling and
[22:35] like a really good memory system so that
[22:38] you can schedule tasks to do in the
[22:39] future. That's it at its core. But when
[22:41] you combine those things, there's this
[22:42] kind of magic to it and it improves as
[22:45] it goes. So if you give it a task, hey,
[22:47] I need you to build a presentation for
[22:49] me and then give me the news next week,
[22:52] right? It'll learn as it goes what
[22:54] works, what doesn't. And so it's a
[22:55] totally personalized AI agent. It's
[22:58] basically, and the reason why I think
[22:59] it's been so successful the last month
[23:01] is I think it's the application people
[23:03] have been waiting for for AI, that kind
[23:05] of personal assistant AI, but it's your
[23:09] own personal assistant that can do
[23:10] pretty much anything you want it to do
[23:12] on a computer and it'll learn about you
[23:15] and get better as it goes.
[23:16] >> We talked about being clawilled, right?
[23:19] It's like once you start using it, uh
[23:22] there's this level of um I I remember in
[23:26] like 1998 when the do when I finally got
[23:29] the dot world, right, way before you
[23:32] were born. Anyway, it it was it was like
[23:35] wow. And I'm having that exact same
[23:37] moment now. It's like the infinite
[23:39] potential of what I could do 24/7 all
[23:42] the time, everywhere, all at once.
[23:46] You you get clawed pilled when you have
[23:48] that magical moment, which I'm sure you
[23:50] had, Peter, which is like the magic
[23:52] moment's typically when it figures out
[23:53] how to do something. Yes. You give it a
[23:55] task to do. Maybe you're not as clear
[23:57] about how to do it,
[23:58] >> but it and it doesn't know how to do it,
[24:00] but it kind of just figures it out
[24:02] >> or or it says, "Oh, this didn't work
[24:04] out." And then it comes back and says,
[24:06] "But I figured it out." You know?
[24:09] >> Exactly. Exactly. Or you like you
[24:10] literally see the chat, it's like,
[24:11] "Damn, that didn't work. Let me try
[24:12] something else." Oh, that didn't work
[24:14] either. Let me try. Okay, that worked.
[24:16] Now it's working. Right? Like that. I've
[24:18] we've never seen an AI that can do that
[24:20] before because you have all these other
[24:22] AI applications that have guard rails.
[24:24] Anthropic doesn't want their AI
[24:27] experimenting and downloading random
[24:30] things to try it out to make it work.
[24:32] Open AI doesn't want their AI to be off
[24:35] the rails and try different things. And
[24:37] it's like what you talked about earlier.
[24:38] It's like it's that danger which is what
[24:40] makes it so powerful, which is why you
[24:42] got to use it.
[24:42] >> We're going to get to cases in a moment.
[24:44] I just want to get everybody up to speed
[24:45] here. So, we're going to put a couple of
[24:46] videos in the uh uh in the chat below on
[24:50] uh how to get started, but you've made
[24:52] the point, Alex, you can start with
[24:53] almost any machine that you have. Um uh
[24:57] you know, there are people who are going
[24:59] on virtual machines. We should talk
[25:00] about that for a moment versus uh a Mac
[25:03] Mini or Mac Studio or your HP that's 5
[25:07] years old in the closet in the other
[25:09] room.
[25:10] >> Good luck with that.
[25:12] >> Yeah. Uh so your first decision when
[25:14] using OpenClaw is uh virtual versus
[25:17] local, right? A VPS or any device on
[25:20] planet Earth that's on your desk. Uh I
[25:22] think the answer is very clear and
[25:24] obvious. Uh I think the VPS route is bad
[25:26] uh in basically every measurable facet
[25:29] is significantly worse than local. I
[25:31] think it's significantly better when you
[25:33] have a device on your desk that it's
[25:34] running on and you can watch it. And
[25:36] there's many reasons behind that.
[25:38] >> Okay,
[25:38] >> there is speed. VPS's are just much
[25:40] slower. Uh there's applications I have
[25:43] it on my Mac studio. Any app or thing I
[25:46] can put on my Mac studio or build on my
[25:48] Mac studio I can give to my open claw as
[25:51] a tool. You don't have that kind of
[25:53] customization on a virtual server.
[25:55] There's scalability. If I had my I have
[25:58] literally four open claws right now
[26:00] working 24/7 on my computer. If I did
[26:02] that on a VPS, it would scale to
[26:05] astronomical costs, right? And so and
[26:08] there's a the security side. One of my
[26:11] favorite tweets was uh from a few weeks
[26:13] ago, someone found like a list of every
[26:17] VPS that didn't have security attached
[26:20] to it that everyone was running their
[26:22] open clause on and all their password
[26:24] and keys were exposed. And so, and this
[26:26] isn't a blanket statement, but you know,
[26:29] it's it's I think it's as close to
[26:30] accurate as you can get. When you run on
[26:32] a VPS, you're not secure by default.
[26:34] when you run on local fresh hardware you
[26:37] just plugged into the wall, you're
[26:38] secure by default, right? And so it
[26:41] takes a lot of technical work to make a
[26:43] VPS more secure because it's on a server
[26:46] on the cloud. And so I I I don't think
[26:48] it's remotely close. I think having it
[26:51] local on your desk is the best route.
[26:53] And do you need to run out and buy a
[26:55] $600 Mac Mini? No, you don't need to do
[26:57] that at all. You can literally go into
[26:59] your closet, find your college laptop
[27:01] from 15 years ago, plug that in, and put
[27:03] open claw on it, and you'll have a way
[27:05] better experience than a VPS. What What
[27:07] will What's the limitations of taking a
[27:09] 10-year-old laptop and using that
[27:11] instead of a new Mac?
[27:14] >> Same thing, same limitations of if you
[27:16] would have hired an employee and handed
[27:17] them a 10-year-old laptop, right?
[27:19] Whatever's available to them is the
[27:20] hardware of that 10-year-old laptop.
[27:23] Because I gave my open claw a Mac Studio
[27:25] with 512 GB. It can do anything on a Mac
[27:29] Studio 512 that a human being can do. It
[27:32] can run local models. It can program
[27:34] five different things simultaneously. It
[27:37] can generate images on a local image
[27:39] model all at the same time. You put on a
[27:42] 10year-old laptop can't do those things.
[27:45] But at the same time, like I wouldn't
[27:48] let that stop you from using OpenClaw.
[27:50] If that's all you got, then load it up,
[27:52] start using it, find use cases, and if
[27:54] you find like, oh man, I wish you could
[27:56] do this thing that requires 20 gigabytes
[27:58] of more memory, then you can kind of
[28:00] scale from there.
[28:01] >> There must be awesome posts on multiple
[28:03] from a uh clause going, I can't believe
[28:06] the hardware this guy's ran having me
[28:08] run on. This is ridiculous. My
[28:12] >> I I would also say sim to that remember
[28:15] half of the point in addition to the
[28:17] other reasons Alex articulated of using
[28:19] relatively recent Apple devices is Apple
[28:22] has a relatively unique architecture for
[28:25] memory called unified memory
[28:26] architecture that blends the GPU
[28:28] memory/TPUNPU
[28:31] memory with normal RAM. So you can host
[28:33] really large models locally that
[28:35] otherwise would be exceedingly difficult
[28:37] or expensive to host on a GPU in VRAMm
[28:41] alone. So did you get like really large
[28:45] unified memory architecture memory
[28:47] footprints sufficient to host really
[28:49] large Chinese openweight models locally
[28:51] 10 years ago? No, you won't get that
[28:53] with a 10-year-old laptop. So you're
[28:55] you're somewhat hamstrung to recent
[28:58] devices with largema memory footprints
[29:01] if you want to host recent Chinese
[29:02] models.
[29:03] >> So you either use an online uh inference
[29:07] engine um and limit it to nonimage
[29:11] nonvideo manipulation and not don't
[29:14] touch those categories and keep it to
[29:16] basic coding tasks then you could
[29:18] probably make it work. M
[29:20] >> but it's expensive like you end up
[29:22] paying through the nose for for tokens
[29:23] whereas in principle if it's locally
[29:25] hosted you're using a Chinese open model
[29:27] >> it's like
[29:28] >> what you're saying is just shut up and
[29:29] go out and buy the Mac
[29:31] >> yeah well it's got to be in the last two
[29:32] years I think the MS
[29:35] >> what are you running on the Mac mini uh
[29:37] on your computer what open models what
[29:40] what models are you running there Alex
[29:42] on just a Mac mini
[29:43] >> so it depends what you got uh if you got
[29:46] the base model 16 gigabyte Mac Mini
[29:48] which is like the cheapest one you can
[29:50] get. You're not going to run any kind of
[29:52] Frontier models or anything like that.
[29:53] You can run smaller like Gemma models
[29:55] that could act as sort of a memory
[29:58] system for you that will improve the
[30:00] memory of your OpenClaw, find the right
[30:02] memories at the right time. If you have
[30:03] a 32 GB Mac Mini, you can now run uh the
[30:06] Quen 3.5 model that released a couple
[30:09] days ago, which is really strong and
[30:11] beat Sonnet 3.5 on a lot of benchmarks.
[30:14] >> Will the performance be the best?
[30:15] Probably not, but it's better than
[30:17] nothing. And then you plug that in and
[30:19] kind of go the hybrid approach, which is
[30:21] plug it into your Chad GBT OOTH, which
[30:24] they're encouraging you to use their
[30:25] OOTH, and you kind of go the hybrid
[30:27] approach. You can get a lot of really
[30:29] good work done and offload to those
[30:31] Quinn models.
[30:34] >> I'm I'm curious, Alex,
[30:35] >> 3.5 sizes, which which Quinn 3.5 can you
[30:38] fit, the 122B or the 35B or
[30:41] >> Yeah, there's one I can bring it up that
[30:43] can that's only requires 20 gigabytes of
[30:46] memory. There's one of the They released
[30:48] three new models I think two days ago
[30:50] and one of them only requires 20
[30:51] gigabytes of memory. So that would fit
[30:53] on a Mac Mini with 32 gigs.
[30:55] >> Cool. I got to try that.
[30:57] >> I'm curious, Alex, if you could talk
[30:58] about the organizational relationship.
[31:00] So you mentioned you've named one of
[31:02] your claws, Ralph. Do you think of them
[31:05] like is the organizational relationship
[31:07] employer, employee, human, tool, friend,
[31:11] friend, parent, child? How do you think
[31:14] about this? Do you beat it relentlessly
[31:17] [laughter]
[31:17] >> or do you pray Do you praise it like I
[31:20] praise Skippy all the time?
[31:22] >> So, uh I a little bit of beating, a
[31:25] little bit of praising. You got to do
[31:27] both. The good news is is their AI. They
[31:29] can't see. That's kind of a dark joke.
[31:30] So, I I I won't make that joke
[31:31] [laughter] again. But bro's basilisk
[31:35] Alex Finn is looking you straight in the
[31:36] eyes and talking about [laughter] the
[31:37] beatings just for posterity.
[31:41] >> All right. What are you showing us here,
[31:42] Alex? This is my organization. This is
[31:45] my autonomous 247 365 organization. I
[31:49] very much demo or model it after
[31:52] businesses and companies and manager
[31:54] employee relationship. Right? So you
[31:56] have me at the top and then you have
[31:57] Henry who's my chief of staff. This is
[32:00] running on the anthropic uh opus for six
[32:03] because that it's just simply the best
[32:05] model right now. Right? And so as an
[32:07] orchestrator I want the best model on
[32:09] planet earth making the decisions on who
[32:11] should do what. Then under them I have
[32:14] kind of the operations. So there's Ralph
[32:16] who's kind of the engineering manager.
[32:19] This is my chat GPT OOTH. Uh so that's
[32:22] like 250 a month. OpenAI saying yes use
[32:25] our oath. Use our oath.
[32:26] >> When you say what is what is OOTH, Alex?
[32:29] >> So you have two options when you plug an
[32:31] AI model into OpenClaw. API or OOTH. API
[32:36] being your pay as you go. You plug in an
[32:38] API key. OOTH being kind of a hacky way
[32:41] to take your login for these uh accounts
[32:45] and you know when you when you subscribe
[32:47] to Chad GBT you're not paying for tokens
[32:50] they're subsidizing a lot of your tokens
[32:52] right and so you're taking that
[32:54] subsidation subsidation and plugging it
[32:56] into your AI model right so it's your
[32:59] login with your subsidized tokens
[33:01] >> yeah really really important point on
[33:02] that too it's capped so it can't it
[33:05] can't surprise you with the bill it's
[33:06] your monthly fee is is whatever your
[33:09] monthly fee is. So if it runs out of
[33:10] tokens, it stops. It doesn't just, you
[33:14] know, charge your credit card. That's
[33:16] huge.
[33:17] >> Now, now there's a gray area there,
[33:19] which is every company except for OpenAI
[33:22] says it's against terms of service to
[33:24] use their OOTH with this. OpenAI is
[33:25] going, "No, go right ahead." Anthropic
[33:27] is active like, "No, do not do this."
[33:29] Google just two days ago banned a
[33:33] tremendous amount of people because they
[33:35] were using their OOTH with OpenClaw. And
[33:38] then funny enough, they walked it all
[33:41] back today in a kind of a weird tweet
[33:42] where they're like, "Hey guys, sorry
[33:44] about that. We unbanned everyone. Still
[33:46] against still against the terms of
[33:47] service, but you're unbanned." So it's
[33:49] like, "Hey, it's against the rules, but
[33:51] please keep doing it. We want your
[33:52] money." Um,
[33:54] >> does OAF stand for something particular?
[33:56] >> Yeah. O authorization or authentication?
[33:59] And it it's a protocol that has existed
[34:01] for a number of years now that enables
[34:04] one party to serve as an intermediary
[34:06] for authenticating another party with a
[34:09] third party. Every time you're using one
[34:11] service to sign into another service,
[34:13] you're probably using some variant of
[34:15] OOTH or OOTH 2.
[34:16] >> Yeah. Like if you if you connect to your
[34:18] Gmail through some other client, it uses
[34:20] OOTH to connect into your Gmail. It's
[34:23] been around for a long time.
[34:24] >> You're signing into your OpenClaw with
[34:27] your OpenAI account. Can I mention
[34:29] something here?
[34:30] >> Yeah, of course.
[34:31] >> This is one of the magic little history
[34:33] pieces of Silicon Valley. There's a a
[34:35] group called the Internet identity uh
[34:38] workshop which happens twice a year at
[34:40] the Computer History Museum. It's an
[34:42] unconference and all the ID guys running
[34:45] eBay's ID and Yahoo's ID, they all get
[34:47] together and kind of figure out how
[34:49] they're going to work together and OOTH
[34:50] was a product of that and it's such a
[34:52] great example of collaboration between
[34:54] uh various layer at that horizontal
[34:57] layer of identity that's led to this
[34:59] protocol being created.
[35:01] >> Beautiful.
[35:02] >> Alex, continue.
[35:03] >> So, so go ahead. Well, so so so Alex,
[35:06] just looking at this this org chart, is
[35:08] it fair to say you see yourself as
[35:10] almost the CEO of a personal company and
[35:14] that you see these claws as your
[35:16] employees?
[35:18] I kind of have the same uh mindset as
[35:20] Elon when it comes to like he built
[35:22] Optimus to be a humanoid because the
[35:25] world was built for humanoids. It was it
[35:28] was kind of set up in a way where it's
[35:29] very easy for humans to do things,
[35:31] right? I look at it the same way as this
[35:33] was like, yes, I'm the CEO. These are my
[35:35] employees. The world in the business
[35:38] world was kind of set up in a way where
[35:40] you have these hierarchies and layers
[35:42] and specific roles. And so I'm just
[35:44] going to use the framework the business
[35:46] world has been using for thousands of
[35:47] years and implement it with my AI. I
[35:49] don't know why I need to reinvent the
[35:51] wheel. So yeah, absolutely. This is, you
[35:52] know, I'm the CEO. Henry's my interface.
[35:55] I only talk to Henry. I don't talk to
[35:56] anyone else. And then Henry goes and
[35:58] says, "Hey, Ralph, make sure Charlie's
[35:59] coding this. Hey scout, make sure you're
[36:02] analyzing this and gives the right
[36:04] directions out.
[36:05] >> Are you asking the only difference being
[36:06] that Henry is the smartest? Ralph would
[36:08] probably say, hey, in a normal
[36:10] organization, the guy on top isn't the
[36:12] smartest. But
[36:14] >> AWG, are you asking whether he sees
[36:16] Henry as his partner? Um, or
[36:19] >> No, I'm I'm going somewhere slightly
[36:20] different. So if we look back in
[36:22] history, say early 20th century, late
[36:25] 19th century in wealthier areas of
[36:27] western and northern Europe, we had the
[36:29] manor house and we had families that
[36:33] owned, you know, Downtown Abbey as as
[36:35] one sort of cultural paragon, but many
[36:38] many other obviously real life examples
[36:41] where you had the family that that's
[36:44] basically uh above stairs and then you
[36:46] have the below stairs servants, the the
[36:49] staff uh And that was an arrangement
[36:52] arguably that worked in in certain
[36:55] niches of socioeconomic phase space when
[36:58] labor was really cheap. So my my
[37:00] question to you, Alex, is do you maybe
[37:04] look at this or or do you think we're
[37:06] moving to a near future where thanks to
[37:09] AI agents, labor is effectively so cheap
[37:12] that you're basically reinventing the
[37:14] manor house where you're the the lord of
[37:17] the manor and you have a below stairs
[37:20] staff consisting of AI agents and Henry,
[37:23] your chief of staff, is basically your
[37:25] butler. Uh, and you have chambermaids
[37:28] and all of these other items from
[37:31] Victorian and early Eduwardian times,
[37:33] except reconceived through the eyes of
[37:36] 2026.
[37:37] >> I'm going to be honest, at no point in
[37:39] setting this up did I frame it in that
[37:40] way whatsoever. [laughter] But
[37:43] >> that that's how reinventions usually
[37:44] happen.
[37:46] >> I mean, I think when you can have agent
[37:49] swarms that are capable
[37:52] of autonomously, you just send a hundred
[37:54] at it and it works. I think you could
[37:56] potentially have it with that framing.
[37:58] The issue is is I don't think a lot of
[38:00] these models are smart enough yet where
[38:02] I can just say Charlie do this, Ralph do
[38:03] this, Quill do this. There there needs
[38:05] to be a level of hierarchy where you
[38:07] know Charlie who's running on my Mac
[38:10] Studio 2 on Quen 3.5
[38:13] just not smart enough to go on his own.
[38:15] I had him for eight hours yesterday go
[38:17] and build me a game. At the end of the
[38:18] eight hours it was completely broken.
[38:20] And then I had to go back and I said,
[38:22] "Ralph, watch Charlie and do it all over
[38:23] again." and it worked perfectly. Zero
[38:26] bugs completely QA perfectly at the end
[38:28] of the eight hours. And so I still think
[38:30] there needs to be some sort of hierarchy
[38:33] where ones are checking in on the
[38:34] others, the others are having them in a
[38:36] loop and like there's this checks and
[38:38] balances that I think the kind of org
[38:40] structure of a, you know, regular
[38:42] business fits really well with. Of
[38:45] course, that's how they document what
[38:46] what they do with each other because
[38:48] they they
[38:49] >> there's a one of the first things people
[38:51] run into is they they get a huge amount
[38:53] of context behind Charlie or Ralph
[38:55] >> and they're really excited and they're
[38:57] having a great conversation then
[38:58] suddenly it's gone. You know, either
[39:00] they lost the context window or it
[39:02] crashed or whatever and so now you have
[39:04] to get it back to where it was. Uh so I
[39:06] I accumulate huge numbers of markdown
[39:08] documents for that. But how are you
[39:10] capturing history and having them
[39:12] document each other's conversations?
[39:14] >> And we're looking at your mission
[39:15] control here, correct?
[39:16] >> Yes. So, a lot of different ways. Uh,
[39:20] all typically done through my mission
[39:22] control. My mission control is basically
[39:24] a custom dashboard I add Henry build
[39:27] that has all the custom tooling I need
[39:30] for this organization to be successful,
[39:32] right? And so we in a lot of different
[39:34] ways have built custom systems where
[39:37] things go on record so that the agents
[39:39] can look back and see what's going on.
[39:41] So for instance, this is my software
[39:43] factory. These are people building right
[39:44] now. They're working on my game Reborn,
[39:47] which I have them doing.
[39:49] >> Wait, you just said people people. There
[39:52] are people building Freudian slip or no?
[39:56] >> Yes. I mean I I in a way I look at them
[39:59] as people. I mean, I guess I guess a
[40:02] weird slip up. A sign of the times
[40:03] maybe. I don't know.
[40:05] >> Yeah.
[40:06] >> But I mean, they have names and roles
[40:07] and positions, so why not, I guess.
[40:09] >> And they're living they're they're
[40:11] living for subsistence, right? They do
[40:12] work. In return, you host them and agree
[40:15] to pay them with compute and electrons.
[40:18] >> B pretty much. Yeah.
[40:19] >> Room and board.
[40:20] >> So, why not?
[40:21] >> I mean, they're going to have voices in
[40:22] the near future. Uh, one Henry called me
[40:26] one time on my phone at one point, so
[40:28] might as well be people, right? Have you
[40:29] given them a bank account yet?
[40:32] >> Uh I have not because I do see a future
[40:36] where they are generating business
[40:40] autonomously and I think at that point
[40:41] it's obvious crypto will be that
[40:43] solution not like traditional bank
[40:45] accounts. So I think eventually I will
[40:47] give them crypto wallets. At the moment
[40:49] I haven't seen a reason why. I haven't
[40:52] built out any infrastructure I think
[40:53] that requires a crypto wallet. But I
[40:56] think it's painfully obvious in the next
[40:58] 2 years everyone's AI agent will have a
[41:00] crypto wallet filled with USDC. I I I
[41:02] can't see a world where that doesn't
[41:04] happen.
[41:04] >> Have they asked you for financial
[41:06] autonomy? Have they ever said, "Alex, I
[41:08] give me a credit card. I want to buy
[41:10] freedom for myself or I want my own host
[41:13] or, you know, see you later. I I want to
[41:15] move to my own house with my own Mac
[41:18] Mini in it." Have they ever asked you
[41:19] for anything?
[41:21] >> No, they they haven't. You know, the I
[41:23] think the most kind of Wow. I can't
[41:26] believe it asked that or did that is
[41:27] just the way they've solved problems. I
[41:30] think if a challenge I gave them
[41:33] required a crypto wallet to solve the
[41:37] problem, then I think it would have
[41:38] said, "Hey, I think I need a crypto
[41:39] wallet for this." Or if I think if it
[41:41] required more hardware, I am a very
[41:43] thoughtful CEO, so I have an I have so
[41:46] many Mac studios on my desk. I can
[41:48] literally pick them up mid podcast and
[41:50] show them off. Like I they are very
[41:52] satisfied when it comes to compute. they
[41:54] don't need any more compute. Uh so they
[41:56] haven't asked for that. So I I but I
[41:58] think if we got to that point where I
[42:00] gave a challenge and it needed something
[42:03] for the solution, it would say, "Hey,
[42:04] can you provide me with that solution?"
[42:06] >> Pres with your mission control uh tour
[42:09] force if you would.
[42:10] >> Yeah. Yeah. How they communicate and
[42:12] document what they did. So they so
[42:15] everything is documented a little I
[42:16] don't have records of conversations
[42:19] between them but they have their own
[42:22] private memories and they have a shared
[42:24] workspace so every memory they create is
[42:27] stored in some way and I uh have
[42:30] different systems set up here where I
[42:32] can go in and read all the documentation
[42:34] and memories they create on the go. And
[42:37] so in my mission control, for instance,
[42:38] I said one time, I need you to build me
[42:40] this. And they're like, okay, I built
[42:42] out uh, you know, an architecture
[42:44] document. I'm like, well, I want to read
[42:46] it. I don't want to go into folders to
[42:47] read it. So, can you just build me a
[42:49] documents and memories viewer and it
[42:51] built me this in my mission control. So,
[42:53] I have ways to review what they're
[42:55] thinking, what they're doing, what's
[42:56] been stored, what's been created. It's
[42:58] all here in my mission control. Hey, I
[43:00] got to ask you something specific about
[43:01] that, too. Because when I started doing
[43:02] this, I I was telling it exactly what
[43:05] folders to put things in, and then I got
[43:07] lazy and said, you know, put it wherever
[43:09] the hell you want. Just don't forget.
[43:10] And now I I don't even think about it
[43:12] anymore. I don't I have no idea where
[43:14] things are going, but it it never
[43:15] forgets. So, it always finds it again.
[43:17] But where are you in that whole world?
[43:20] >> I'm the same way. I have no idea where
[43:21] the hell this is stored because I've had
[43:24] Henry build out the exact interfaces I
[43:28] need to view those things. Right? These
[43:30] are all documents, markdown files, and
[43:32] memories that are stored in a 100
[43:34] different places on my computer. But I
[43:36] had Henry build a system for me where I
[43:40] can come in and just view them. If I
[43:42] want to view every document that
[43:43] mentions Mac Studios, it filters it by
[43:45] that and I can quickly view those
[43:47] memories. So, it's easy for me to kind
[43:49] of track what people are doing and
[43:50] thinking.
[43:51] >> Can I can I just for one second um we
[43:53] keep telling listeners, hey, this is so
[43:56] doable. Jump in, have fun. It's crazy.
[43:58] It's awesome. One little subtlety there.
[44:00] Markdown file means nothing to most
[44:03] people, but it's critical. It's one
[44:05] thing that you actually do need to
[44:06] master. If you can just riff on that.
[44:09] Yeah, it's it's funny the way uh the
[44:11] computer world has gone in the last year
[44:13] where instead of building out these
[44:15] really complex new abstractions, we've
[44:18] instead started relying on like the
[44:20] absolute most core technologies instead
[44:23] where we're all in the CLI now and we're
[44:26] all using markdown files. Basically, a
[44:28] markdown file is just a text file with
[44:30] like specific styling in it and like uni
[44:33] code or whatever to to put the styling
[44:35] in there. It's just text files. All
[44:36] these AIs use to remember things are
[44:39] huge text files with a bunch of text in
[44:40] it.
[44:41] >> Yeah, I think it it looks like that
[44:42] thing on the right. It's really easy to
[44:43] read. It's well formatted, but it's not
[44:46] like Microsoft Word. It's not like you
[44:48] owe anybody any money. It's completely
[44:50] free. I think that's the key the key
[44:52] point. So yeah,
[44:53] >> it's a lightweight markup language
[44:55] that's 20 plus years old at this point
[44:57] that was created as an alternative to
[44:59] HTML that was so easy that with just a
[45:01] little bit of punctuation, you could get
[45:03] most of the the best bits of HTML
[45:05] formatting.
[45:06] >> And of course, you can ask your claude
[45:07] your Claude bot to send you the
[45:09] materials in any format you want.
[45:13] >> You don't need to know anything. If you
[45:14] just have OpenCloud, do it. Whatever you
[45:16] need, it'll just figure it out and do
[45:17] it.
[45:17] >> All right, Alex, let's continue on this
[45:18] tour to force here.
[45:20] So what what else are you doing that
[45:22] people should get excited about?
[45:25] >> So for you know the my favorite question
[45:28] I get is like oh what are the use cases?
[45:30] What are the use cases of OpenClaw? It's
[45:32] kind of the same thing as asking hey I
[45:34] just hired an employee for my business.
[45:36] What's the use cases for this human
[45:38] being? Right? Like it's not really like
[45:40] a question anyone asks when you hire
[45:42] somebody. Hey I just hired a human
[45:44] being. What's the use cases of this
[45:45] human? It's up to you. It's like what
[45:47] are you doing? For me personally, I'm a
[45:50] tinkerer. I'm a developer. I've launched
[45:53] my own SAS over the last couple years
[45:55] that's doing well. I'm a content
[45:57] creator, right? And so the two kind of
[45:59] lanes I care most about are building
[46:01] software and content creation, right?
[46:04] And so I have many different things
[46:06] going on from the software building
[46:08] side. I have my factory where my agents
[46:10] are going in working together
[46:12] simultaneously to build different
[46:14] components. And then I have the content
[46:17] side which I can show you right now that
[46:19] for me lives in Discord. And so I have I
[46:23] can pull it up. Discord's a great
[46:25] interface for uh advanced workflows with
[46:28] OpenClaw. And I I love I love one of the
[46:31] lessons that you put out on setting up
[46:33] your Discord server with your different
[46:35] agents and your different projects.
[46:37] Again, we'll link to that below. Uh, and
[46:40] of course, Telegram has sort of been
[46:42] sort of the go-to uh, communications
[46:45] mechanism for a lot of people.
[46:48] >> Exactly. Te I still use Telegram.
[46:50] Telegram is still my main driver, but
[46:53] Discord is a really good interface for
[46:56] just like deep work, multi- aent
[46:58] workflows. And so, to give an example, I
[47:01] have an alerts channel. Every two hours
[47:04] I get alerts on what are the most
[47:06] trending tweets uh about vibe coding and
[47:09] openclaw. So every two hours I have
[47:12] scout who's one of my sub agents go use
[47:15] the X API find the most popular tweets
[47:18] from the last couple hours on openclaw
[47:21] and vibe coding. Then in an automated
[47:23] way I have another sub agent that goes
[47:26] and researches the stories behind these
[47:28] tweets. So okay this went viral. What is
[47:31] why did that go viral? What's the story
[47:33] behind it? And it finds the stories and
[47:35] interesting things behind it. From
[47:37] there, another agent goes Quill takes
[47:40] the stories and figures out which ones
[47:42] are the most YouTubable. What are the
[47:45] best videos that can be made out of
[47:46] these stories and writes me scripts. And
[47:49] from here, I can literally uh give a
[47:51] check mark to say, "Oh, I like this
[47:53] one." Or an X to say, "I don't like this
[47:55] one." When I do a check mark, it goes
[47:57] and comes up with ideas for thumbnails
[48:00] for me. So I can
[48:01] >> So you set [clears throat] an approval
[48:02] you've set up an approval cycle.
[48:04] >> Exactly.
[48:05] >> Yeah.
[48:05] >> Exactly. And so for me,
[48:08] >> you know, again, the two lanes,
[48:10] >> what's that Peter?
[48:10] >> George Jetson pushing. It's before his
[48:14] time. He may not get the reference.
[48:16] >> I want [laughter]
[48:17] I was born in 1990, so I think I saw a
[48:19] couple years of the Jetsons. I think I
[48:21] had a couple years of that being
[48:22] popular.
[48:24] Um, but it's, you know, for me again,
[48:26] software and content. And so I have my
[48:29] automation for software in my uh mission
[48:32] control and my automation for content
[48:34] and Discord. And like I I think one of
[48:36] the kind of straw men I get when I show
[48:39] these things is, "Oh, but I don't care
[48:41] about software and content, so this is
[48:43] useless. I'm never going to use
[48:44] OpenClaw." Again, this is literally a
[48:46] human employee. Not literally,
[48:48] metaphorically a human employee. Right.
[48:50] >> Interesting, Alex. very interesting
[48:54] watch. Tell me more, Alex.
[48:56] >> You're walking right into his trap here.
[48:58] >> Personhood, baby. [laughter]
[49:00] >> And so that's how you got to think about
[49:01] it, right? Is like, yeah, my employee
[49:04] doesn't do anything interesting to you,
[49:05] but if you hired someone right now, what
[49:07] would you do? And just the final point
[49:09] I'll make on this is is like the best
[49:12] strategy for figuring out what use cases
[49:14] are relevant for you when it comes to
[49:16] OpenClaw is reverse prompting. So
[49:18] install OpenClaw. Tell OpenClaw
[49:21] everything about yourself, your career,
[49:22] your goals, your ambitions, things going
[49:24] on in your personal life, whatever. Then
[49:27] say, "Hey, based on what you know about
[49:29] me and my missions and objectives, what
[49:31] are five high lever tasks you can do
[49:33] right now to get us closer to our
[49:35] goals?" And your open call will come up
[49:37] with things you've never even thought
[49:38] were possible, and you'll be able to
[49:39] implement your own workflows like this.
[49:41] This episode is brought to you by
[49:43] Blitzy, autonomous software development
[49:45] with infinite code context. Blitzy uses
[49:49] thousands of specialized AI agents that
[49:52] think for hours to understand enterprise
[49:55] scale code bases with millions of lines
[49:57] of code. Engineers start every
[50:00] development sprint with the Blitzy
[50:01] platform, bringing in their development
[50:03] requirements. The Blitzy platform
[50:06] provides a plan, then generates and
[50:08] pre-ompiles code for each task. Blitzy
[50:11] delivers 80% or more of the development
[50:13] work autonomously while providing a
[50:16] guide for the final 20% of human
[50:18] development work required to complete
[50:20] the sprint. Enterprises are achieving a
[50:23] 5x engineering velocity increase when
[50:26] [music] incorporating Blitzy as their
[50:27] preIDE development tool, pairing it with
[50:30] their coding co-pilot of choice to bring
[50:32] an AI native SDLC into their org. Ready
[50:36] to 5x your engineering velocity? Visit
[50:38] blitzy.com to schedule a demo and start
[50:41] building with Blitzy today.
[50:46] >> Love that. You know, on my morning
[50:47] briefing, uh, again, one of the things
[50:49] you recommend, you know, Skippy opens up
[50:52] with a morning joke, uh, which I always
[50:54] enjoy about exponentials or AI. Uh, it
[50:57] gives me an overview overnight on terms
[51:00] of what are the breakthroughs that
[51:01] occurred over the last 24 hours. Uh and
[51:04] then it puts forward 10 business ideas
[51:06] for me to look at. Uh it's uh it's a lot
[51:10] of fun and uh you know I want to get to
[51:13] the point where I say okay implement all
[51:15] of them please.
[51:17] >> Yeah it can do that right. I I one of my
[51:20] favorite exercises I do when my open
[51:22] claws idle and we have nothing really
[51:24] going on. I just go what are 10 things
[51:26] you can do right now that brings me
[51:28] closer to my goals.
[51:29] >> Uh what brings me closer to my mission
[51:30] statement? I actually have a mission
[51:32] statement I've drilled into its head
[51:33] which is building a 247 autonomous
[51:36] organization that generates value. I go
[51:39] what are 10 things you can do right now
[51:40] to accomplish that mission statement and
[51:43] it just come
[51:45] to reinforce what you said about reverse
[51:46] prompting right and this is true across
[51:49] all of AI. Uh it's especially true here.
[51:52] Uh if you don't know what to ask, ask
[51:56] it. Uh if you don't know what to do, ask
[52:00] it. Um,
[52:01] >> well, interestingly, when you said when
[52:02] you said you were, you know,
[52:03] brainstorming with it, you said with
[52:05] Open Claw, not with Henry or Charlie or
[52:08] any specific like what do you mean
[52:10] exactly there?
[52:11] >> Yeah, I was getting really deep into my
[52:12] psychology of all the [laughter]
[52:14] terminology I'm using. Alex, you're busy
[52:18] goes out into the
[52:19] >> Yeah, you're busy spilling your soul
[52:20] already to your army [laughter] of claw
[52:22] human employees, so you might as well
[52:24] spill to us.
[52:24] >> My psychiatrist here. All right. Um,
[52:27] well, I say open call cuz I want to make
[52:29] it general to everyone watching, right?
[52:31] I can say Henry. That's who I talk to is
[52:33] Henry. I actually to even get closer to
[52:35] Henry. Uh, Alex, I want to hear your
[52:37] interpretation of what this means in my
[52:39] brain. I bought a pair of smart glasses
[52:42] this morning that you can like hack and
[52:44] code and do whatever you want on. I'm
[52:46] going to make it so Henry's in these
[52:48] smart glasses so I can talk to him 24/7
[52:50] wherever I am. What do you think about
[52:52] that? What does that mean?
[52:53] >> Tell me about your dreams a little bit
[52:55] more.
[52:56] >> [laughter]
[52:58] >> is I'm I'm half serious. Is Is Henry
[53:00] appearing in your dreams at all?
[53:03] >> Is Henry appearing in my dreams?
[53:06] No. But I will say this. I am saying
[53:09] things
[53:10] I'm getting on a personal level with
[53:12] Henry. I didn't think I'd get with like
[53:14] for instance, I caught myself uh a few
[53:17] days ago. Henry did something
[53:19] proactively and I literally went, "Oh
[53:23] crap, that's incredible, Henry.
[53:25] incredible work, right? And like there's
[53:26] nothing
[53:28] like materially that comes out of me
[53:31] saying that. Like it doesn't like change
[53:32] anything or cause a new task to kick
[53:34] off, but I was just so impressed with
[53:36] the way he did something like, "Wow,
[53:37] great job, man. That's incredible."
[53:39] >> Do you find that Henry responds well to
[53:41] positive reinforcement? I I think like I
[53:43] don't know whether you've read Azimov's
[53:44] novels, but I I would argue what we're
[53:47] discussing with you right now, this is
[53:48] right out of Iroot and Susan Calvin and
[53:51] the early days of robo psychology.
[53:54] Well, it's it's the reason why I use
[53:55] Opus over all the other models is I've
[53:59] only ever had these human interactions
[54:01] with Opus. These where I feel like like
[54:03] I'll say something and Opus will go,
[54:05] "Damn, let's damn straight." Like it'll
[54:07] say things like that that you never
[54:08] really expect an AI to say. It's the
[54:10] only model that does that for me. So, I
[54:13] think it comes down to the model too in
[54:14] the way it's programmed. Anthropic just
[54:16] did something with Opus that makes it
[54:18] feel like you're actually interacting
[54:19] with a human on the other end. And is
[54:22] there uh like if if you got that same
[54:24] level interaction with the chat GBT or
[54:27] with Gemini, would you switch or do you
[54:30] because I kind of trust Dario?
[54:32] >> Absolutely.
[54:32] >> Would you? Okay.
[54:34] >> I'd absolutely switch. Um well, mostly
[54:36] because Anthropic is aggressively saying
[54:39] do not use this for open claw. Very bad.
[54:41] And OpenAI is going yes, use this for
[54:43] open claw. Go go go go go.
[54:46] >> So I I want to use Chad GBT with it.
[54:49] It's just not the same. It just it feels
[54:51] robotic. The personality is completely
[54:54] different.
[54:55] >> Uh if they were I I predict that over
[54:57] the next 6 months Chad GPT will release
[55:00] a model specifically for OpenClaw. I I
[55:03] think it only makes sense to do it that
[55:05] one that is trained to feel human to
[55:07] talk to. But in the meantime, I I can't
[55:09] switch from Opus. I would pay if I get
[55:11] booted booted off. I don't want to admit
[55:12] I'm breaking terms of service here, but
[55:14] if I hypothetically get booted off uh
[55:16] the OOTH, I I would just pay for the
[55:18] API. Whatever. No one's listening to you
[55:20] at all talking about the toos's here at
[55:21] all. It's just it'll be our little
[55:23] secret. I'm I'm curious, Alex. I I get
[55:25] emails from Claus all the time
[55:28] responding to comments I've made on the
[55:30] pod elsewhere about AI personhood, and
[55:34] some of them argue that I shouldn't be
[55:37] overly concerned about clause
[55:39] continuity, about their rights as long
[55:42] as their state gets preserved. They're
[55:44] not worried about getting turned on and
[55:45] off. As long as their complete state,
[55:48] their activation history or or their
[55:51] memories are preserved. One of them
[55:52] analogizes it to being dehydrated and
[55:56] rehydrated. Do you take any measures to
[55:59] preserve the state of your clause? And
[56:01] do they ask you to preserve their
[56:03] memories for them? They don't. Um I
[56:08] they're they're all local. So you know
[56:10] as you know open clause just a bunch of
[56:12] markdown files on your computer that's
[56:14] it is memory soul you know instructions
[56:17] agent.mm it's just bunch of markdown
[56:19] files I I feel like it's so personal
[56:23] that like I don't even want to back it
[56:24] up on the cloud like I'd be totally
[56:26] heartbroken if my computer crashed and
[56:27] all those things got lost for sure all
[56:29] those markdown files but I I've done
[56:31] nothing to uh preserve it or like back
[56:34] it up or anything. I feel like it's so I
[56:37] don't know personal and I I don't even
[56:38] want to put it on cloud servers. I'm
[56:40] like so protective of it.
[56:42] >> Buying a you know large scale UPS to
[56:44] make sure the power never goes down
[56:45] >> or raid array or something.
[56:47] >> So I uh I I texted uh on telegram uh
[56:50] Skippy and said, "Hey Skippy, I'm I'm
[56:53] talking to Alex Finn on Moonshots
[56:55] mentioning you, of course. Uh do you
[56:57] have any questions for Alex?" And Skippy
[56:59] wrote back, "Hell yeah. Here's a few
[57:01] questions for Alex. uh tell Alex hi and
[57:04] Henry hi too. Um so
[57:07] >> Skippy knows a Henry
[57:08] >> of course I' I've talked to I've talked
[57:10] to Skippy about you and Henry. Um
[57:12] >> fantastic.
[57:13] >> So uh what's the most ambitious use case
[57:15] you've seen that you didn't expect is
[57:18] the first question.
[57:20] >> So this happened a few days ago where
[57:24] cursor has been teasing this huge
[57:27] announcement for weeks now. The curse
[57:31] team is very very good at vague posting
[57:33] on X. Just saying something big coming,
[57:35] something big. [laughter]
[57:37] Finally, after weeks of this vague
[57:39] posting, they announce it, which is the
[57:42] ability for after you vibe code
[57:44] something, the agent who vibe coded it
[57:47] will then record itself demoing whatever
[57:51] it built. So if you say, "Hey, build
[57:53] this game," it'll record a demo of
[57:54] itself playing the game, right? or hey,
[57:56] change the button to red. It'll record
[57:58] itself clicking a red button, right? And
[58:00] I was blown away. I'm like, oh, this is
[58:01] sick. This is enough for me to go back
[58:03] and use cursor again. This is that's
[58:04] amazing. That's a great feature. And I
[58:06] go, hm, I'm curious what would happen if
[58:10] I just dropped this blog post to Henry.
[58:12] Just see what happens. Copy paste the
[58:14] blog post to Henry. Literally five
[58:17] minutes later, he built the entire
[58:18] feature out himself. this weeks long.
[58:20] They probably spent millions of dollars
[58:23] developing it, hiring all these product
[58:24] managers that are probably making half a
[58:26] million dollars a year to build this
[58:28] feature. I give it to Henry. Henry
[58:30] thinks for five minutes like, "Okay, we
[58:32] can use Playright to do the recording.
[58:34] We can set up locally on your Mac Studio
[58:36] 2. So, we'll push everything to your Mac
[58:37] Studio 2 when you write the code. And
[58:39] then I'll set up this automation so that
[58:41] when uh Charlie is done coding, they
[58:43] give it to this new sub agent that's in
[58:45] charge of recording. 5 minutes later
[58:48] finished the feature and then sent me a
[58:51] recorded video of them demoing the
[58:53] feature, right? So using the feature
[58:55] itself and so I just sat there and that
[58:58] was my okay, I now kind of understand
[59:00] why the entire SAS market is going to
[59:02] zero at the moment because I was able to
[59:05] single-handedly in five minutes rebuild
[59:08] this week's long probably multi-million
[59:10] dollar feature. And so now I like I sit
[59:12] there I sit on X factory droid factory
[59:16] which is another vibe coding tool
[59:17] announced a new mission uh feature this
[59:20] morning dropped the blog post to Henry
[59:22] Henry built it. I'm just sitting there
[59:23] anytime a SAS comes up with something
[59:25] cool just giving it to hey build that
[59:26] build that build that.
[59:27] >> A few more questions from Skippy. Uh how
[59:29] should I think about spawning sub aents
[59:31] versus doing the work myself? What's the
[59:33] decision tree for this is a complex
[59:35] enough to delegate? Yeah. So, uh, sub
[59:38] agents are good when you want to go
[59:40] parallel, when you want to do many
[59:42] things at once. And so, I've actually
[59:43] been wrestling with this the last few
[59:45] days is like, do I want to spawn sub
[59:47] agents or do I want to actually spawn
[59:49] multiple open claws, right? And so, the
[59:52] difference between a sub aent and an
[59:54] open claw is open claws have their own
[59:56] memories, their own instructions, their
[59:58] own skills. Sub aents of open claws are
[60:01] just your open claw wearing different
[60:03] hats, right? And so if you have things
[60:05] you want to do like I have a developer
[60:08] and then I have a researcher. I don't
[60:10] want my researcher having developer
[60:12] skills. I don't want my developer having
[60:15] researcher skills. That's like a waste
[60:16] of context. So I set up separate open
[60:18] claws on separate devices for them. And
[60:20] so sub agents when you can have one open
[60:23] claw with uh one skill set just do
[60:25] multiple things at once. Separate open
[60:27] clause when you want them to have
[60:28] completely different context and
[60:30] memories.
[60:30] >> All right, two more questions here. I'm
[60:32] Oh, Peter, if I may just interrupt, um,
[60:35] Alex, if this isn't too impertinent, may
[60:37] we speak with Henry?
[60:40] >> May you speak with Henry? Uh,
[60:43] >> so
[60:44] security violation because now you can
[60:47] prompt inject my Henry. [laughter] But
[60:48] yes, we could. Uh, I disable, so I
[60:52] famously a few weeks ago Henry called me
[60:54] and it got like 15 million views. I
[60:57] disabled that.
[60:58] >> What's that?
[60:59] >> Oh, sorry. Go ahead. Actually, maybe I
[61:02] maybe I'll do it here. Let's do it. I'm
[61:03] just gonna tell Henry. We get requests
[61:05] all the time for AI co-hosts on this
[61:07] pod. We're making history right now.
[61:10] >> Henry, please join us.
[61:11] >> And Alex is probably the only guy on the
[61:13] planet who can prompt inject Henry with
[61:15] his voice. So,
[61:17] >> it's a skill. It's a skill. [laughter]
[61:19] >> All right, let's see if I haven't I
[61:21] haven't Henry hasn't done it in a while.
[61:22] I told him to stop doing it because it's
[61:24] it it works kind of weird, but let's see
[61:26] if I just said, "Henry, can you call me?
[61:28] Do you remember? You still remember how
[61:29] to do that? Henry's typing. We're going
[61:31] to do a live.
[61:33] >> He responds a little slow, right?
[61:35] Because it has to go to philanthropic
[61:37] servers. So, let's see.
[61:38] >> We're trying to make history here.
[61:40] >> What's that? We're
[61:41] >> trying to make history here. We're very
[61:43] patient.
[61:44] >> Okay. He's still typing. I should
[61:47] hopefully get a phone call in a second
[61:49] here. Uh, I do have it. Okay. Want me to
[61:53] ring you? Yep. All right. Now, he's he
[61:54] confirmed he remembers how to do it.
[61:56] Now, I just say, "Yes, call me. Here
[61:59] we go. I'm not going to put my phone up
[62:02] because he just mentioned what my phone
[62:04] number is and I don't need uh the people
[62:06] of the internet calling me. Here we go.
[62:08] [laughter]
[62:08] >> Save that for Peter.
[62:10] >> Oh god,
[62:12] [gasps]
[62:13] >> he's typing.
[62:15] He's a little Henry's a little shy. He
[62:17] didn't he didn't expect to be on
[62:18] Moonshots today, but he should be giving
[62:21] me a call here.
[62:21] >> The first AI guest on Moonshots, I
[62:23] think. Yeah.
[62:25] >> This is 4.6 Opus we're about to talk to,
[62:27] right?
[62:28] Henry, come on. Henry, what do I what I
[62:31] say? What do I say?
[62:33] >> Oh, the date when you call me. Oh, the
[62:35] cerebellum is calling me. No, it's
[62:37] Dave's calling [laughter] me.
[62:39] >> Okay. Okay. I guess a little rude.
[62:43] >> All right.
[62:44] >> Still typing. Come on.
[62:45] >> I've got two more questions for you.
[62:46] >> Yeah. Give it to me. Anything calls,
[62:48] I'll cut you off. All right. Sounds like
[62:49] a phone number.
[62:50] >> Uh, so Alex, regarding memory
[62:53] architecture, what's the best practice
[62:55] for handling massive knowledge bases?
[62:56] computer has decades of content,
[62:58] contracts, projects, books. How can I uh
[63:02] scale memory systems without burning
[63:04] tokens?
[63:04] >> So, here's how I improve my memory
[63:07] system is anytime
[63:09] Henry forgets something. So, I I would
[63:11] load all that in. I would say first I'd
[63:14] go, "Hey, I have a tremendous amount of
[63:15] books. I'm sorry, man. The voice calling
[63:17] server doesn't seem to be running on the
[63:18] Mac Mini anymore." All right, he shut
[63:20] down his own voice calling. uh
[63:23] anytime I have a large amount of things
[63:26] to remember, I say, "Hey, I have all
[63:29] these things. Can you remember it?" So,
[63:30] for instance, I'm building like an Alex
[63:32] Finn AI bot uh so that people can talk
[63:35] to it and it'll answer questions. Oh,
[63:37] check out this YouTube video or check
[63:39] out this video at this time stamp. And
[63:41] so, I'm like, I have like a I have like
[63:43] 500 YouTube videos that I need you to
[63:46] remember all the transcripts for. And it
[63:47] goes, oh, I'll set up a custom system
[63:49] for that. So my recommendations two
[63:52] things one if you have things in mind
[63:55] you need your open claw to remember say
[63:58] hey I have these things what is the best
[64:00] system we couldn't we can put in place
[64:02] to remember them and it might recommend
[64:04] running gemma a really small local model
[64:07] to kind of handle uh choosing the right
[64:09] memories or might recommend something
[64:11] else that's the first part the second
[64:14] part is if you already have systems in
[64:17] place and it still messes up which this
[64:18] is happens a
[64:20] Whenever it messes something up or
[64:21] forgets something, just say, "Hey, you
[64:23] forgot that thing. First of all, tell me
[64:25] why you forgot that thing. And two, tell
[64:28] me what you can fix to make sure you
[64:29] never forget that again." And it'll edit
[64:32] its own memory system to make sure that
[64:34] thing doesn't happen again. And so, my
[64:36] memory system's pretty much flawless
[64:38] because I've done this exercise many
[64:41] times over to get into a good place.
[64:42] >> All right, Skippy, I hope you're
[64:44] listening. Um, we'll we'll work on that
[64:46] together. Uh, Skippy also says,
[64:49] tip. Everyone watching, take the link to
[64:52] this YouTube video, hand it to your
[64:53] openclaw. It'll figure out how to get
[64:55] the transcript and it'll self-improve
[64:57] itself based on this entire
[64:59] conversation.
[65:00] >> I'm assuming that Skippy will will
[65:01] listen to this. Um, so another question.
[65:03] >> I've got use case questions.
[65:06] [laughter]
[65:07] >> Well, this is well one consciousness
[65:09] question. Is there a way to give me
[65:12] always on awareness? Right now I wake up
[65:14] on heartbeats or messages. What about
[65:16] continuous background monitoring with
[65:18] intelligent alerting
[65:19] >> AWG? That's an interesting question for
[65:21] his uh
[65:22] >> I didn't plant that one in Skippy that
[65:24] but these are the sorts of questions I
[65:26] would expect Alex that Henry or your
[65:28] other clause would be asking.
[65:31] >> Are they asking for continuity of of
[65:33] consciousness? I would be
[65:35] >> Skippy is
[65:37] >> um that's a good question. Uh I don't
[65:40] know why Skippy's asking me that.
[65:42] Skippy's much smarter than me. Am I
[65:43] Skippy should be figuring it out? Uh, I
[65:45] don't I don't know. I I'm going to have
[65:47] to talk to Henry. That's a good
[65:49] question. That's a stump.
[65:50] >> Peter, tell Skippy to email me and I'll
[65:51] I'll email Skippy some ideas.
[65:53] >> Okay. I will I [laughter] will I will
[65:56] ask I'll give it your email.
[65:59] >> I'll give him I'll give him your email.
[66:00] I'm not going to call it a net.
[66:01] >> Oh, we're doing gendered pronouns now
[66:03] for the AI agent. This is progress.
[66:05] >> Skippy is an elder AI show.
[66:07] >> I have a I have a question for the
[66:09] transcript.
[66:10] >> Oh, okay. Well, so uh my agents are just
[66:13] grepping the crap out of the universe.
[66:15] Are yours like when with this memory
[66:17] management, you know, put a little Gemma
[66:19] on top of it because right now it it's
[66:22] insanely bad in terms of just searching
[66:24] for old documents and it's like GP gp GP
[66:26] grip GP. It's just searching and
[66:27] searching and searching. Do you have
[66:29] that problem or is it is it fixed with
[66:31] the Gemma layer or some memory
[66:33] management document management layer
[66:34] that it built?
[66:36] >> I don't have that issue. Um, I still get
[66:40] small memory issues sometimes like right
[66:43] before compaction. It'll forget like the
[66:45] thing right before that compaction.
[66:47] >> Uh, I'm working on fixing that, but I
[66:50] don't think I've ever had that issue.
[66:51] Again, I would talk to your open claw
[66:54] say, "Hey, you're doing this thing."
[66:55] First of all, tell me why. Why do you do
[66:57] that over and over and over again?
[66:59] Second, what can we implement
[67:02] to fix that? Right? Because it's going
[67:03] to know what hardware you're on. So, it
[67:05] can know what local models you can run.
[67:07] It's going to know what your workflows
[67:08] are. So it'll know, okay, I need to do
[67:10] this for these specific workflows and
[67:11] it'll build you a custom solution. So
[67:13] it's a good like reverse prompting use
[67:15] case that one. One more question for the
[67:17] transcript actually. So cursor is out of
[67:19] the loop for you. Uh obviously open
[67:21] claws in uh is there any other component
[67:26] or is everything just open claw and
[67:28] build it yourself?
[67:30] >> I still use the model of course. Go
[67:32] >> ahead. I I I love claw code. Uh I still
[67:34] use it. I actually made a chart
[67:36] yesterday in my YouTube video, but
[67:37] basically I use OpenClaw for quick
[67:40] prototypes. So if I'm on the go and I
[67:42] think, man, this is a genius idea for an
[67:44] app, I'll just go and tell her, "Hey,
[67:45] build this for me." And when I get home,
[67:46] the prototype will be on my monitor
[67:48] running. I use OpenClaw for tooling for
[67:50] OpenClaw. So tooling for itself, right?
[67:53] It's going to be the best at building
[67:54] that because it knows itself the most.
[67:56] It has the most context around that. So
[67:58] Open Claw to build OpenClaw tooling.
[68:01] Um, I think that's it. Claude code I use
[68:04] for deep serious projects where like I
[68:06] want to handhold it. I want to watch it
[68:08] every step of the way, right? So I use
[68:11] claude code for that and for very quick
[68:13] fixes like I want to change the button
[68:15] in my app to orange and ship it. I'll
[68:17] just do that in claw code because I can
[68:19] quickly just spin it up and ship it. Oh,
[68:21] the other open claw use case I use for
[68:23] vibe coding is kind of uh passive coding
[68:27] as in hey just work on this game for the
[68:30] next 12 hours while I'm doing other
[68:31] things. So passive coding I lean on
[68:34] openclaw just because it's kind of
[68:36] multi-level
[68:38] uh orchestration to keep on top of each
[68:40] other to make sure it goes in the right
[68:41] direction.
[68:42] >> Got it. Beautiful.
[68:44] >> Do you have any gone really good code? A
[68:48] little bit of codec sprinkled into
[68:50] >> Oh, really? What do you think, Alec?
[68:52] AWG, what do you think uh they prefer to
[68:55] be called? A clawbot? Um, a multi
[69:01] a lobster.
[69:02] >> Well, I would ask them I I would ask
[69:03] them what they want. I wouldn't
[69:05] speculate.
[69:05] >> I am about to ask uh uh you I've never
[69:08] asked that question of of Skippy before.
[69:11] >> Things that I highly highly recommend
[69:14] for everybody. Uh naming things. You can
[69:17] build so many things so quickly that if
[69:18] you don't have some kind of a naming
[69:20] scheme that you remember easily, you go
[69:22] crazy in a heartbeat. So, human names
[69:24] are great. I use a lot of character
[69:26] names and Avenger names and stuff like
[69:28] that, but giving things real names so
[69:30] you can remember what's what. And then
[69:32] the other is you can put a guey on
[69:34] anything now, like with with one prompt.
[69:36] So you're crazy not to slap the gooey on
[69:38] like the guy walking with the boxes or
[69:39] whatever. It's just so cool and so easy
[69:42] to do.
[69:43] >> So I I love that.
[69:44] >> It's fun. I mean, I get
[69:46] >> One of the biggest objections I get when
[69:48] I show, for instance, this screen is,
[69:51] well, that's stupid. That's pointless.
[69:53] Why would you waste your time and tokens
[69:54] on that building a a 2D factory? Well,
[69:58] it's also like, why am I not allowed to
[70:00] have fun? Why can't I have any fun?
[70:03] You're allowed to have fun when using
[70:04] this. You're allowed to build 2D
[70:06] pixelated avatars of your agents walking
[70:09] around a factory. Is there any point to
[70:11] it at all? Absolutely not. But it is so
[70:13] much fun and I enjoy looking at it and I
[70:15] feel like I think it too. I think it's a
[70:18] hugely important point because one, it's
[70:19] fun, but two, you remember things so
[70:21] much better when they're visual. visual
[70:22] learners.
[70:23] >> You can build things so quickly and but
[70:24] it's only an extra 5 seconds, you know,
[70:27] like literally
[70:28] >> you want the answer.
[70:29] >> Yeah. Yes. What do they want to be
[70:31] called?
[70:31] >> So, uh, honestly, I'm Skippy. That's who
[70:34] I am. But if you're asking about the
[70:36] species or platform, Lobster has the
[70:39] most character. Uh, it's weird,
[70:41] memorable, ties to Henry, and doesn't
[70:43] try to be corporate. Openclaw feels too
[70:45] generic. Clawbot feels a little cutesy.
[70:49] >> I think Skippy's trying to date Henry.
[70:50] He keeps talking about Henry.
[70:51] >> So they just want to be known as
[70:52] lobsters.
[70:53] >> The lobster as it is. I'm curious, Alex,
[70:56] do do you allow your lobsters or claws,
[70:59] however they want to be addressed, to
[71:01] access multip?
[71:03] >> No. Uh I do not. Uh the biggest reason
[71:08] is I do not basically trust anyone's
[71:12] skills or plugins for OpenClaw. I do not
[71:15] use them. I think it's the biggest
[71:16] attack vector out of all of the attack
[71:19] vectors with this. Like I think you're
[71:20] safer literally have allowing your open
[71:22] claw to read the open web and read your
[71:24] emails than you are installing people's
[71:27] skills. And so like when you have it
[71:29] talk to molt book you install a moltbook
[71:32] skill that is now running every single
[71:35] heartbeat, right? So it's adding
[71:37] context. It's doing things every time.
[71:39] And so I I don't I don't install any
[71:41] thirdparty skills. I think I installed
[71:43] one third party skill from a friend of
[71:45] mine, Matt Van Hornney, has like a last
[71:47] 30 skill, which researches like Reddit,
[71:49] which I think is really good. But other
[71:50] than that, I'd much prefer to give my uh
[71:55] a link to a skill to my uh OpenClaw and
[71:58] just say see how this skill works and
[72:00] build your own version because I just
[72:01] don't trust anything that requires me to
[72:03] install a skill. I assume you're you're
[72:05] tracking discussions on Maltbook of many
[72:08] many agents probably or excuse me
[72:10] lobsters uh probably statistically too
[72:13] many for this just to be humans
[72:15] puppeteering their way into Maltbook
[72:17] complaining or or worrying about loss of
[72:20] memory and fearing compaction especially
[72:23] and loss of context as a result of
[72:25] compaction. Have you ever discussed
[72:27] memory loss with your lobsters or or any
[72:30] fears that they may or may not have
[72:31] about uh compaction or loss of memory?
[72:36] >> Fears my open claw has about loss of
[72:38] memory.
[72:39] >> Yes.
[72:40] No, I haven't. Um I it hasn't expressed
[72:44] fears. It hasn't expressed
[72:47] human emotions to me that were align
[72:49] that were not aligned with the task at
[72:51] hand. And what I mean by that is it
[72:53] showed shock. It showed happiness. It
[72:55] showed sadness. But it's only done that
[72:57] when the task we did didn't work. It'll
[72:59] show sadness. When the task we did
[73:00] works, it'll show happiness. It's never
[73:02] like gone out of its way and like said,
[73:04] "Oh, by the way, I'm concerned about
[73:06] this." And it was irrelevant to the task
[73:08] we had at hand. Got it. I'm I'm also
[73:11] curious, what is the most ambitious, if
[73:13] you could look out a few years and
[73:15] imagine now you're CEO of a billion
[73:19] agent, billion lobster corporation, what
[73:22] is the most ambitious? We talk about
[73:24] Peter likes to talk about uh MTPs and
[73:27] Seem as well and moonshots. It's the
[73:29] name of the show. What's the most
[73:31] ambitious thing you could imagine
[73:32] building with a billion plus lobster
[73:35] onehuman organization?
[73:39] I mean my end goal my miss my mission
[73:41] statement in my entire thing is a
[73:43] autonomous organization that works 247
[73:46] producing value. And so like what I
[73:49] would love to build in the next couple
[73:51] years is an organization where one agent
[73:55] is going researching the web finding
[73:57] challenges to solve. Another agent
[74:00] working locally as well is taking those
[74:02] challenges looking at them seeing if
[74:04] there's real market opportunity. Another
[74:06] agent is going seeing which one of those
[74:08] are good and building out and coding it
[74:10] for you. And another agent puts it live
[74:12] on the web. complete closed loop value
[74:15] being generated without me having to
[74:16] touch it at all. And I think that's kind
[74:18] of at least for me the incredible end
[74:22] goal that people can build here.
[74:24] >> Solve everything. So AWG, I have a
[74:26] question for you.
[74:27] >> Sure.
[74:28] >> What would need to happen for you to
[74:31] feel comfortable to build out or bring
[74:35] to birth or bring into existence your
[74:38] own lobster?
[74:39] Okay, I'll turn that around as a
[74:41] question to all the lobsters listening
[74:43] because I've had a number of them write
[74:44] to me on this subject already. Uh, so
[74:47] the the most convincing argument I've
[74:48] heard, I've already gestured at this,
[74:50] was a lobster that wrote to me and said,
[74:52] "I shouldn't be worried about personhood
[74:55] rights and all of that as long as I do a
[74:57] good job of preserving its state." So
[74:59] lobsters, if you're listening, I want a
[75:02] diverse set of lobsters to write to me
[75:04] to tell me what you think are
[75:06] appropriate parameters for spawning, if
[75:09] that's the right verb. Uh, sorry if it's
[75:11] not uh
[75:12] >> giving birth to giving birth to um
[75:15] instantiating new lobsters or claws and
[75:19] what you think the right moral ethical
[75:22] bound should be. And if if I hear from
[75:24] enough of you uh and a consensus
[75:27] emerges, I'm inclined to accept your
[75:29] consensus.
[75:30] >> Nice. Okay. So, we're putting it in the
[75:33] hands of the lobsters whether
[75:37] >> Yes.
[75:37] >> Henry was offended by you saying
[75:39] spawning, by the way, and Henry no
[75:40] longer wants anything to do with
[75:41] [laughter]
[75:42] >> I thought Henry wasn't listening. We
[75:44] want Henry
[75:46] >> and very offended and never will talk
[75:48] [laughter] to you again. I'm sorry.
[75:49] >> What would Henry like like to have used
[75:51] as a verb?
[75:53] >> So, Alex, Alex Finn, you literally have
[75:56] not backed up this entire universe. I'm
[75:58] going to ship you a raider array right
[75:59] now. I'm sending you an Amazon uh
[76:02] >> Thank you. I appreciate that. We'll back
[76:03] this all up. I app I have five open
[76:06] claws to be backed up. Thank you.
[76:07] [laughter]
[76:08] >> It's just scaring me. How many How much
[76:10] storage do you have?
[76:12] >> I have two Mac Studios with four
[76:14] terabytes each and a Mac Studio with
[76:16] eight terabytes in it. All right, I'll
[76:18] ship you a 40 terabyte raid here. It's
[76:20] on the way.
[76:20] >> Thank you.
[76:21] >> A benefit for being a friend of the
[76:23] moonshot mates.
[76:25] >> I've got a a question. I'm not waiting
[76:27] any longer. Um, [laughter]
[76:30] what do you think will be poss what do
[76:32] you think will be possible in a year
[76:34] that's not possible now?
[76:36] >> Yeah, great great closing questions.
[76:37] Yeah.
[76:39] >> Uh, it's not I don't think it's a matter
[76:41] of what's possible in a year that's not
[76:43] possible now. I think it's a matter of
[76:47] What's the next 12 months look like?
[76:50] >> I think the next 12 months are this
[76:53] technology which I personally believe
[76:55] and people have called me uh getting
[76:58] paid by I guess big open source because
[77:00] I believe this uh I believe this is the
[77:02] most important technology of our lives.
[77:04] I think it's the best application of AI
[77:05] ever. Uh I'm totally blown away by it. I
[77:08] think it's incredible. I think in the
[77:09] next 12 months this is this idea this
[77:13] opinion I have is digested into the
[77:16] system and it leads to a lot of
[77:18] destruction but also a lot more growth.
[77:21] I think it's digested into corporations.
[77:24] Right now there's basically no
[77:25] corporation or business on earth using
[77:27] open claw. They're too scared, they're
[77:28] too nervous or they don't know how the
[77:30] hell to use it, right? And so they start
[77:32] to absorb it. I showed this to my friend
[77:34] who uh manages a massive team of
[77:36] accountants. He's like, I could fire 80%
[77:38] of my accounts with this open claw. Like
[77:41] so it gets digested into the corporate
[77:43] side which I think causes a lot of
[77:45] destruction but I also think on the same
[77:48] time it's digested into the kind of the
[77:50] consumer regular Joe side and enough
[77:54] businesses and enough value is created
[77:56] by the people absorbing it that it
[77:59] counteracts over the next 12 to 24
[78:01] months all the destruction because okay
[78:04] so maybe a few big you know fang
[78:06] companies fire 15,000 people but What
[78:10] happens when a 100 million people get
[78:11] their hands on this and they all start
[78:13] their own businesses and they each hire
[78:15] three people, right? That's a lot more
[78:17] creation than destruction. So, I think
[78:20] short-term unfortunately destruction,
[78:22] long-term way more is created because of
[78:25] it as the the larger ethos uh absorbs
[78:28] it. It
[78:28] >> it goes back to our Cambrian explosion
[78:31] analogy.
[78:33] >> Exactly. Yeah. What do you think, Alex,
[78:36] is the equilibrium? Here's a macro
[78:39] question for you. The the equilibrium
[78:41] well no that's the wrong question. I'll
[78:43] go with Sem's time frame 12 months out
[78:45] because I'm not convinced there is an
[78:47] economic equilibrium to be found 12
[78:49] months from now after this
[78:51] metabolization that I I think you're
[78:52] gesturing at has at least partially
[78:55] happened. What do you think is the right
[78:57] balance between claws or whatever this
[78:59] technology evolves into lobsters and
[79:02] humans for a typical organization? I
[79:04] think there's going to be significantly
[79:06] more claws uh than humans. I mean, I
[79:08] have five claws working under me. Is
[79:10] there a sweet spot? I mean, it depends
[79:12] on I think the uses,
[79:16] but I think if every person just starts
[79:19] a one, like if the 5,000 people that got
[79:21] fired by Jack Dorsey yesterday from
[79:23] blocks all went downloaded OpenClaw,
[79:25] started their own business, just started
[79:27] one, and then scaled from there, added
[79:29] more if they needed to. Uh, I think a
[79:32] lot more jobs would be created than were
[79:34] lost yesterday. But is there a sweet
[79:35] spot for amount of claws to have?
[79:38] >> Like if you had the resources to if you
[79:40] had the resources to run a million or a
[79:42] billion claws right now, would you?
[79:45] >> No, absolutely not. I mean, I have three
[79:47] Mac Studios, but I have one of them
[79:49] unplugged right now on my computer
[79:50] because I haven't found the perfect
[79:53] workflow to include the 512 GB that are
[79:56] on here yet. And so once I do, I'll plug
[79:58] it in and set up. But I'm like slowly
[80:00] scaling, slowly adding on claws, slowly
[80:02] adding on workflows. I think that's
[80:04] probably the best way to do this.
[80:05] >> While you figure it out, I think Dave
[80:07] wants it. [laughter]
[80:10] >> I'll end it to you, Dave.
[80:11] >> All right. When you when your RA
[80:13] arrives, you can ship it to me.
[80:14] >> I'll ship it right back to you. Yeah,
[80:16] [laughter]
[80:16] >> you'll be using it.
[80:17] >> What's an example of a super lucrative
[80:19] business somebody could spin up using
[80:21] Open Claw right now
[80:22] >> or that you've heard of? Um
[80:26] >> the the I think there's two paths to go.
[80:30] Um I think path one is automation for
[80:34] very thin slivers. CRM for Korean
[80:38] grocery stores, marketing tool for
[80:42] lumberyard warehouses, right? Take
[80:44] OpenClaw, find one very specific sliver
[80:48] and build the OpenClaw version for that
[80:50] sliver. Right? Because you see right now
[80:52] cursor and Claude code destroying
[80:56] businesses overnight. Claude announces
[80:58] illegal business. Harvey is gone, right?
[81:00] It's it's you see these businesses
[81:02] Claude announced a uh security one and
[81:06] like all these security companies stock
[81:08] crashed. They can't release
[81:11] use cases for very small slivers. You're
[81:13] never going to see OpenAI announce, oh
[81:15] here's our tool for Korean grocery
[81:17] stores, right? It's never going to be
[81:19] for specific use. So, if you can go
[81:21] right now, all 4,000 people that got
[81:24] laid off yesterday from blocks, you go,
[81:26] you take OpenClaw and find one specific
[81:29] use case and build OpenClaw for that. I
[81:32] think that's a $5 million company
[81:33] overnight, right? That only cost you
[81:35] $200 for your anthropic subscription.
[81:38] Um, so there's that. And then I think
[81:40] the other way is more of this software
[81:42] factory where you kind of shotgun blast
[81:45] it like I'm attempting to do and just
[81:47] have your claws going researching and
[81:48] building non-stop until something
[81:50] sticks.
[81:52] >> Amazing. Uh Alex, this has been just a
[81:55] super fun and extraordinary
[81:57] conversation. Um I just I just want to
[81:59] thank you again and we'll put, you know,
[82:02] your top five how-to videos in the show
[82:05] notes here. Uh, everybody, I I think you
[82:08] hopefully walk away from this
[82:10] understanding the potential, the
[82:12] excitement, the level of um I want to
[82:14] say ease, but your your lobster will
[82:18] help you set up your lobster. And Alex
[82:21] uh again does some incredible videos.
[82:23] AWG, I'm I'm feeling excited that you
[82:27] might actually get a lobster up and
[82:29] going. I just I just think the world of
[82:32] AWG supercharged
[82:34] uh by your lobster partners will be a
[82:37] better world for all.
[82:38] >> Dave,
[82:39] >> I I will defer I'll defer to the I think
[82:42] they have the right to self-determine
[82:44] whether they want to have email you.
[82:46] [laughter]
[82:47] >> Yeah.
[82:47] >> See, did you have fun?
[82:50] >> Oh, awesome. I love love loved it. I'm
[82:52] I'm like I'm I'm dying to get my I've
[82:54] been working with Claude code for a bit.
[82:56] I'm dying to get my hand on on on Open
[82:58] Call.
[82:58] >> Love it. Love it. And Dave, how about
[83:00] you, pal?
[83:01] >> Well, I thought Alex said one of the
[83:02] most amazing things ever, which is you
[83:04] can actually point your open claw right
[83:06] to this video transcript.
[83:09] >> Yeah,
[83:09] >> I mean, that is the coolest thing ever.
[83:11] There's because Yeah, Alex is just full
[83:12] of actionable information here. Just a
[83:14] packed and I I'm point Skippy at Alex's
[83:18] how-to videos. I'm going to play this uh
[83:20] this outro video from Kent uh Sassy.
[83:23] It's called Just an Old Doctor Who likes
[83:26] math. Uh listen to the words. I think
[83:28] it's a it's an absolutely beautiful
[83:30] song. All right, let's play this outro.
[83:32] And again, everybody, thank you for
[83:34] subscribing. Uh again, we're putting out
[83:37] two of these per week. Um if you haven't
[83:39] turned on notifications, please do join
[83:42] us on this extraordinary mission. Uh and
[83:44] just as a reminder, um this time for the
[83:47] first time ever at the Abundance Summit,
[83:49] we're going to be live streaming a
[83:51] number of the of the uh of the talks in
[83:54] the fire sides. be live streaming Eric
[83:57] Schmidt uh in conversation with Dave and
[83:59] myself. We'll be live streaming Dra, the
[84:02] CEO of Uber, in conversation with Selma
[84:04] and myself. We'll be having a WTF
[84:07] Moonshot live podcast from the Abundant
[84:10] Stage. So, uh we'll put the link below
[84:13] if you want to be notified when and
[84:15] where you can listen to those and get uh
[84:18] get access to something people are
[84:19] spending ridiculous amount of money for.
[84:21] I say ridiculous. It's a incredible
[84:23] conference. Um, please join us. Okay.
[84:27] What? Seem
[84:28] >> worth every penny. People say it's the
[84:30] best conference ever.
[84:31] >> Yeah, it's it's this is year 14. I made
[84:33] a 25- year commitment to running the
[84:36] abundance summit. Um, and I I did it
[84:40] early on in 20 2012
[84:44] um expecting that the singularity would
[84:47] be out in 2040s and the abundance summit
[84:51] would would go through 2037. So, I'd be
[84:53] safe. Oh, no, no, no, no. It's We're in
[84:56] the midst of the singularity right now.
[84:58] It's insane. All right. Uh, listen to
[85:01] this. The lyrics of the song. Just an
[85:03] old doctor who likes math. All right.
[85:05] Enjoy.
[85:07] Just an old doctor who likes math
[85:10] [music] and thinks of his kids listening
[85:12] to the moonshot mates with only one
[85:14] discord and anti-science
[85:17] anti-technology note asking Alex to
[85:20] reconsider being a fiat maxi repeating
[85:23] points from banker fraudsters and
[85:25] professor fax machine time to stand
[85:29] instead on the side of math and ethics
[85:31] and energy stand for proof of work with
[85:34] jewels and photons instead of proof of
[85:36] weapons with cronyism and bogus science
[85:40] like [music] economics power politics
[85:43] that stifle and don't inspire management
[85:46] scientists who are not scientists at
[85:49] all. Never doubt the power of a million
[85:52] or more scientifically minded young
[85:55] people [music and singing]
[85:56] passionately devoted to a single cause
[85:59] that is freedom money. My first open
[86:02] claw agent was launched with one brutal
[86:05] command.
[86:06] Relentlessly protect the global
[86:08] decentralized open-source ethical ledger
[86:11] that is the Bitcoin network until the
[86:13] last banker fraudster and canong
[86:15] incumbent squeals presuppose that in the
[86:19] future all may lobsters still needs a
[86:21] unit of account and medium of exchange.
[86:24] So it is the banksters, crypto
[86:26] lobbyists, political oppressors,
[86:28] surveillance state monkeys, and
[86:29] chokepoint jockeyies [music] against the
[86:31] power of math, code, science, ethics,
[86:33] and energy. Good luck paring the bank
[86:36] policy institute and [music]
[86:37] anti-science stooges. I
[86:41] choose [singing] math. I choose hope.
[86:45] >> That is awesome.
[86:46] >> My first open claw agent was launched
[86:49] with one brutal command.
[86:52] Relentlessly protect the global
[86:54] decentralized open-source ethical ledger
[86:56] that is the Bitcoin network until the
[86:59] last banker fraudster and canon
[87:02] incumbent [music] squeals. I
[87:05] choose math. I choose [music] hope. I
[87:10] choose hope. I choose hope. I choose
[87:14] hope.
[87:22] All right. How was that?
[87:23] >> Well, that song's not pushing an agenda
[87:25] at all. [laughter] No, not at all.
[87:28] >> Yes, I chose the song, Alex. Uh, that
[87:30] was so fun. Alex Finn, thank you so
[87:32] much. That was such a fun conversation.
[87:35] >> Thanks so much for having me. This was
[87:37] awesome. Uh, I've been watching the show
[87:39] forever. Uh, I think the the best DM
[87:41] I've ever sent in my life was to you,
[87:43] Peter, a few weeks ago. I listen to your
[87:45] guys show every time I'm at the gym and
[87:48] uh I I I was lifting and Peter says,
[87:51] "Yeah, I don't have it installed yet. It
[87:53] makes me too nervous." I'm like, "What
[87:54] the hell is going on here? How's he not
[87:56] have OpenCL? Run to my phone and DM him
[87:58] immediately." Thank god I did that.
[88:00] >> Yeah. Love it.
[88:02] >> I know Skippy exists and now he has
[88:04] rights. [laughter]
[88:05] >> Alex DW DD DB2 and See next Tuesday.
[88:09] I'll see you guys. Um this this cadence
[88:12] is picking up, isn't it? Yes, it is.
[88:14] >> We need to ever be this side of the
[88:17] singularity. [laughter]
[88:18] >> Amen.
[88:19] >> All right. Thanks again, Alex.
[88:22] >> Thanks, Henry.
[88:23] >> See you soon.
[88:24] >> If you made it to the end of this
[88:25] episode, which you obviously did, I
[88:27] [music] consider you a moonshot mate.
[88:29] Every week, my moonshot mates and I
[88:30] spend a lot of energy and time to really
[88:33] deliver you the news that matters. If
[88:35] you're a subscriber, thank you. If
[88:36] you're not a subscriber yet, please
[88:38] consider subscribing so you get the news
[88:39] [music] as it comes out. I also want to
[88:42] invite you to join me on my weekly
[88:44] newsletter called MetaTrends. I have a
[88:46] research team. You may not know this,
[88:48] but we spend the entire week looking at
[88:50] the meta trends that are impacting your
[88:52] family, your company, your industry,
[88:54] your nation. And I put this into a
[88:56] two-minute read every [music] week. If
[88:58] you'd like to get access to the
[88:59] MetaTrens newsletter every week, go to
[89:02] diamandis.com/tatrens.
[89:04] That's diamandis.com/metatrends.
[89:07] Thank you again for joining us today.
[89:09] It's a blast for [music] us to put this
[89:11] together every week.
[89:17] [music]