Key Takeaways


Alex Shetski frames AI as an overhyped but useful tool with broken unit economics; each new user adds inference cost, funded by a “circular” VC pipeline that routes money back to Amazon, Google, and Nvidia. LLM progress has plateaued, pushing the market toward an oligopoly while open-source lags at scale. He argues the real opportunity is a new six-layer internet stack that adds native payments (Bitcoin) and identity/social graph (Nostr) to the web, with AI as an effort multiplier: small teams can now ship global, low-fee, payment-embedded products fast. Reflecting digital burnout and content “slop,” his startup Satlantis is pivoting from social feeds to a Bitcoin-native events app, solving real meet-space problems (global payouts, low fees, portable identity, discovery via social graph) instead of competing for attention.

Best Quotes

“AI isn’t a second coming of God, it’s a tool, like the tractor or the PC.”

“Every new AI user actually costs you.”

“The money going to startups ends up back with the same LPs, Amazon, Google, Nvidia.”

“Bitcoin dying would be catastrophic for civilization. Nostr dying wouldn’t matter much.”

“We’re not a Bitcoin company. We’re an events company that happens to use Bitcoin and Nostr.”

“The next 20 years will be a renaissance in consumer apps.”

“We don’t need another social app. We need tools that get people off their screens.”

Conclusion

The episode marks a pragmatic turn: skip AI grandiosity and Web2 clone wars, and instead build real products on an open stack where Bitcoin handles global payments, Nostr provides portable identity, and AI multiplies effort. As users tire of algorithmic outrage and infinite content, the winning apps will re-humanize tech, creating markets that didn’t exist (like Bitcoin-native events) and channeling digital networks into physical community, sovereignty, and durable value.

Timestamps

0:00 - Intro

0:37 - Musings on AI

6:20 - The AI business model

14:33 - Broad application on AI

20:52 - Bitkey & Obscura

22:36 - Importance of Nostr, Satlantis

34:16 - Carving a new market

43:15 - Crowdhealth & Unchained

45:03 - The Cold Start Problem

53:13 - New layers in the design space

59:00 - Improvments to meetups

1:04:00 - Everything app

1:07:14 - Unplugging and passing the torch

1:19:54 - Head down and build

Transcript

(00:00) that a lot of the money was actually coming through Amazon or Google or Nvidia or something like that. That was sort of like either the LPS or behind the LPS. It was kind of like the wrong kind of circular economy. It was like laundering money. But the concept of general intelligence is just infinite.

(00:18) Now we're at a new stage and now you have small startups again can now do payments at scale with Bitcoin. Hopefully do a social graph at scale with something like Nostra and can build faster, better, quicker and go to market faster, better, quicker with AI tools. The cost of Bitcoin dying and failing is catastrophic. \[Music\] Stand at the precipice. Gold running to 4,000.

(00:43) Bitcoin cooling off after hitting a new all-time high. The debasement trade is on. AI is taking over the world. Or is it? Sitting down with Alex Gretzky to solve all the world's problems again. It's been a while, dude. Yeah, we we solved the world problems like what was it a year ago, 18 months ago. Now we got new problems to deal with. So here we are again.

(01:07) Yeah, a lot has changed and we were just riffing on AI. Is it hype, overhype? Is there something there? Is the way it's being applied, the way it will ultimately be applied um in terms of bringing the most utility and productivity to the world. Where are we in the cycle as somebody who has been toying around with it for quite a long time now? Yeah, I think I don't think my opinion's changed since we we finished up and open sourced the spiritual stuff.

(01:38) Um, my opinion remains that it's a tool. Uh, it's very much like the tractor or uh, you know, the the PC, the laptop, whatever. It's um I think where I've always diverged with a lot of the overhype um or hype in general is that it's some sort of you know second coming of God or um some sort of sentient being that's going to take over the world and all this sort of stuff.

(02:02) I've always felt like that was, you know, the nerd's wet dream, you know, and like overblown uh kind of like esqueologgical um desire to see the circuits take over the world and to prove all the um the religious people wrong, that you know, God can be created in the in the circuit.

(02:29) So, I've always felt like there was a there was a strange um sort of desire for that that was completely logical. And we we might have discussed this last time. It's been so long, but um I know I've rifted with someone about the fact that intelligence or the the concept of general intelligence is just infinite. And what we have with the breakthrough with LLMs has been specialized intelligence around the reproduction of patterns um linguistically first and now uh in applications like video but by no means um is this anywhere near a general intelligence nor is it anywhere near

(03:02) sentience which is you know further down the track. So there was um I shared a tweet with you now from um from Devon Ericson and a couple other people who sort of like pointing at this which is hey this is a a great tool uh it can be used effectively. Um although as with all tools it's quite hard to be used effectively. Most people just use it poorly and they just get slop.

(03:27) And as we're seeing the whole world is sort of or the whole online world is just being filled with more and more and more slop. Um, but if used effectively, um, you can create quite cool things with it. Um, I know, for example, I'm I'm a father now to a newborn and, uh, the little guy was, um, was sneezing the other day and the first thing we did was we just asked Chip, hey, you know, what should we do? Should we get like some saline solution or anything like that? What like what do we do? And we got some good feedback.

(03:58) So, so for like for that kind of stuff, it's a super breakthrough. Um, also for things like the video that you guys did for um for the opportunity cost calculator or the opportunity cost app, like that was \[ \\ \] brilliant, man. Um, so so for that stuff fundamentally it's it's great.

(04:17) But does that mean that this is um going to end all the jobs and replace all the human beings and give us some sort of sentient being? \[ \\ \] no. like that is just and I think people have sort of slowly come to that conclusion themselves instinctively cuz you know you had GPT3 GPT4 and people like oh yeah by the time we get to GPT5 this thing's going to be like \[ \\ \] god and GPT6 it's going to be running the world and here we are GPT5 is not really any different than four um you know we we sort of had that jump between three and four but since then you know

(04:48) everything's kind of tapered off um and all of the different models are kind of you know the same the big guys whether it's Grock GPT Gemini they all sort of like function the same so I feel like we've hit a ceiling there um so yeah that's my that's sort of my thoughts off the cuff yeah the I completely agree cuz it's funny making these videos it's people and we've had a bunch of people come to us like how are you doing this like it like what what is the secret sauce it's like there is no secret sauce like it actually does take hard work like the amount of iterations you have to run through with the

(05:27) generations because you understand that the LLM are actually not smart and do not they're not intelligent. You have to literally handhold them and format things in a certain way. You have to bring the context of a cinematographer and a director and understand how to prompt that. Like it's not easy.

(05:47) We have a bunch of people in our circle saying a we're going to go do this and it's we're going to create similar memes and short films, whatever you want to call them. And it's like, all right, good luck. Cuz it takes us like a week. The last one we just did took 55 hours of work from storyboarding, scripting, and then iterating on generations.

(06:10) And it is simply just a tool that allows us and enables us to do things that would have cost us tens of thousands of dollars and now they cost $5,000. um in labor and minimal cost of credits right now, which they're good. I mean, you pulled up another tweet from from Zoomer. It's like, how sustainable, yes, is the business model for these large language models with how much they're charging customers right now? I think we're probably in a very uh unique and anomalous point in the cost curve for the end customer right now where they're just essentially

(06:46) um trying to subsidize this with loss leading by making tokens super cheap when that we know that the power that uh the cost of the power it takes to actually produce the outputs is far exceeds and hacking the hardware as well far exceeds the what they're charging us. This is this is such a good point.

(07:08) So, actually I um around around the time when we were sort of trying to commercialize Spirit Satoshi 2 years ago now, um this hit me like a ton of bricks as a business owner because we had spent all of this time um training and building and refining and fine-tuning this the the spiritual associ we had finally got to a good enough model that we thought we could charge for it.

(07:36) And when we went to uh some Bitcoin companies to kind of sell it as a assistant, but an assistant that actually understands Bitcoin, um they were only willing to pay like 20 bucks a month or something or maybe 50 bucks a month for it. And when we did the