Bittensor’s (alleged) $10M rug pull (feat. Mark Jeffrey) | E2275

Summary

This Monday’s TWIST is a deep dive into the most disruptive event in the Bittensor (TAO) ecosystem to date: the alleged $10 million rug pull by Sam Dare, the founder of Templar (Subnet 3) and CEO of Covenant AI. Mark Jeffrey — a 30-year internet OG, Stillcore Capital partner, and one of TWIST’s go-to crypto explainers — joins to walk through what happened on-chain. Sam allegedly dumped 37,000 TAO (~$10M) of his subnet 3 holdings into the on-chain Uniswap pool, walked away from Covenant AI’s three subnets (Templar, Basilica, Grail), and posted a long “decentralization activism” letter as justification. Mark, Lon, and Alex find that letter unconvincing, and the on-chain action looks far more like a classic exit scam than a principled stand. The price of TAO dropped 25% in the immediate aftermath, but the ecosystem rallied — subnet owners across Bittensor posted “I am Spartacus”-style declarations of loyalty and Const (Constantine Steeves, Bittensor’s founder) responded with a thoughtful rebuttal and a fresh proposal to make subnet ownership contingent on locking the largest token stake for the longest period, so that founders are forced to demonstrate “capital conviction.”

The second half of the episode goes wide on the kinds of subnets that make Bittensor genuinely interesting. Ken Miyachi (BitMind, Subnet 34) explains his deepfake-detection / proof-of-human system, where one set of miners is paid to detect AI-generated content and a separate adversarial set is paid to fool them — a “steel on steel” loop that updates daily as new generative models like nano-banana drop. He raised $3M from Arch Capital, Canonical Crypto, Mechanism Capital, and Blockchain Builders Fund. Will Squires and Stefan Cruz of Macrocosmos (subnets 1, 9, 13) walk through IOTA, a “Train at Home” application that strings together idle consumer compute and interruptible data-center GPUs into a globally distributed training cluster — the goal being to train ChatGPT-class frontier models out of “ground beef” rather than “ribeye steak.”

Calacanis frames the whole conversation around an unlock that’s bigger than the drama: subnets give unaccredited investors public-market-style liquid exposure to extremely early-stage AI startups, and — in his view — that participation is what makes AI politically tolerable as it eats more of the economy. The episode closes with an off-duty riff on Bieber’s stripped-down YouTube-themed Coachella set vs. Sabrina Carpenter’s pop spectacle, Bruce Springsteen on Broadway, a Chappelle late-night crowd-work show, and the Staples Baddie — the TikTok creator at Staples who sold Calacanis on the Zebra G-750 pen.

Highlights

Bittensor as “AI Linux”

“It sort of reminded me of the OS wars when you had OS/2, you had Microsoft NT, you had Sun Solaris, but who won? It was Linux, right? Linux powers 90% of our world today. And what I saw with Bittensor were sort of the early moments of what looked like an AI Linux. OpenAI, but like what OpenAI should have been, truly open AI.” — Mark Jeffrey, 7:21

yt-dlp --download-sections "*7:21-8:30" "https://www.youtube.com/watch?v=sdhuUZbx5Sk" --force-keyframes-at-cuts --merge-output-format mp4 -o "twist-bittensor-ai-linux.mp4"

What Sam Dare actually did on-chain

“Sam Dare, he dumped 37,000 TAO on his own subnet holders. It’s about $10 million in money and he sort of walked away from the entire project. It was Covenant AI, the designers of that 70 billion parameter model… He sold 1% of the alpha holdings on three Covenant subnets that were near 100%. … The price of TAO dropped about 25% in the immediate aftermath.” — Lon Harris, 12:55

yt-dlp --download-sections "*12:55-14:30" "https://www.youtube.com/watch?v=sdhuUZbx5Sk" --force-keyframes-at-cuts --merge-output-format mp4 -o "twist-sam-dare-rug-pull-recap.mp4"

”I am Spartacus” — the ecosystem’s rebuttal

“It’s basically the Bittensor version of I am Spartacus, right? Everyone is standing up and saying no, no, none of this stuff is true. And Const gave a very, I thought, thoughtful reply to the letter that Sam posted as you know his rage letter.” — Mark Jeffrey, 20:45

yt-dlp --download-sections "*20:24-21:30" "https://www.youtube.com/watch?v=sdhuUZbx5Sk" --force-keyframes-at-cuts --merge-output-format mp4 -o "twist-bittensor-i-am-spartacus.mp4"

Owner lock-ups: capital conviction as the new ownership rule

“If you’re operating the subnet, then your tokens are locked… Const released a new one this morning where basically the ownership of the subnet is always up for grabs based on who locks the most tokens for the longest period of time… So the chain will reward capital conviction in a subnet. So it is possible you’ll lose your subnet to somebody that believes in it more than you.” — Mark Jeffrey, 23:14

yt-dlp --download-sections "*22:52-24:00" "https://www.youtube.com/watch?v=sdhuUZbx5Sk" --force-keyframes-at-cuts --merge-output-format mp4 -o "twist-bittensor-owner-lockup-proposal.mp4"

BitMind’s adversarial deepfake-detection loop

“If you did this in a traditional way, what you’d have to do, wait for [a new model] to come out, generate a bunch of that data, retrain your detection model. And what we have is an open decentralized permissionless system where people are generating data for us all the time and people are also coming up with new detection algorithms, retraining quickly, and so we can actually update our system on a daily, weekly, at a very fast cadence to always stay ahead of the newest generation models.” — Ken Miyachi (BitMind, via the Subnet 34 segment), 27:00

yt-dlp --download-sections "*26:27-28:00" "https://www.youtube.com/watch?v=sdhuUZbx5Sk" --force-keyframes-at-cuts --merge-output-format mp4 -o "bitmind-steel-on-steel-deepfake-loop.mp4"

”We make AI models out of ground beef”

“Everybody loves ribeye steak, but actually meatloaf’s what you want when you’re comfortable… we take bits of compute from here or there, we perform some magic in the background, and we make meatloaf. … So the way IOTA works is the sort of minimum compute unit we can use is, um, at the moment the size of a MacBook and for about 30 minutes, and we can actually extract value from that.” — Will Squires (Macrocosmos, via the IOTA segment), 45:54

yt-dlp --download-sections "*45:54-48:00" "https://www.youtube.com/watch?v=sdhuUZbx5Sk" --force-keyframes-at-cuts --merge-output-format mp4 -o "iota-meatloaf-vs-ribeye-training.mp4"

Key Points

• The cold-open thesis (0:15) - Mark Jeffrey: “the incentivization alignment engine works spectacularly well, but then it fell out of alignment when there was sudden success” — the rug pull is what extreme founder upside looks like without lock-ups
• Mark’s 30-year resume (0:52) - Built The Palace virtual community in the 90s; early Bitcoin OG; now running Stillcore Capital, an unaccredited-friendly TAO + subnet fund
• Why Bittensor over generic crypto (2:42) - “I’ve never seen an ecosystem where there was more early signal than the Bittensor ecosystem”; ~50% of teams have credible product hypotheses
• Uniswap as the architectural ancestor (4:04) - Permissionless listing + community-supplied liquidity; Mark uses it to explain why a subnet owner can dump tokens at-will into the on-chain pool
• The “AI Linux” framing (7:21) - Bittensor as the open-source counterweight to the five companies that own centralized AI
• Stillcore’s portfolio so far (10:29) - 8 subnet investments including Ridges, Shutz, Hippias, Vidio, Score; tilted toward product-market-fit + revenue, with research bets like Nova (pharmaceutical molecule hunting) as moonshots
• What “rug pull” actually means here (11:44) - Subnet owners are emitted a large stake by the chain; they have information asymmetry and unlocked tokens, so they can flood the on-chain Uniswap before anyone reacts
• The Sam Dare timeline (12:55) - 37,000 TAO (~$10M) dumped across Templar, Basilica, Grail; 1% of the alpha holdings on three near-100%-owned Covenant subnets; TAO -25% on the news
• Sam was previously on TWIST (14:25) - The hosts say in retrospect he came across as “squirrely”; “an edge to him… a little unpredictable”; answers about future plans were “not as crisp”
• Const = Constantine Steeves (15:17) - Bittensor co-founder; built the original Templar subnet for Sam, who used to work at the Open Tensor Foundation under him
• The Carl Rinsch / Netflix analogy (17:44) - Director who took Netflix’s sci-fi budget, bought crypto and luxury mattresses, and is now serving jail time — a reminder that founder absconding happens in every industry, not just crypto
• What Sam should have done (21:23) - Negotiate an amicable exit with the Open Tensor Foundation, hand over the subnet private key, keep a negotiated allocation
• Owner lock-ups as the smart-contract fix (23:14) - Const’s morning-of proposal: subnet ownership goes to whoever locks the most tokens for the longest period; capital conviction becomes price discovery
• BitMind / Subnet 34 (24:43) - Ken Miyachi’s deepfake-detection company; consumer drag-and-drop tool, BitMind Bot (Grok-style on X), enterprise APIs for media + government evidence verification
• Three deepfake use cases that surprise people (28:24) - Newsroom verification, business interview infiltration (“North Koreans… grifters trying to get multiple jobs”), and AI-generated evidence in legal/government cases
• BitMind’s $3M seed (33:00) - Arch Capital, Canonical Crypto, Mechanism Capital, Blockchain Builders Fund
• Token vs. equity in Bittensor (33:32) - Subnet token = “always-liquid Bitcoin-like ownership in the product engine”; equity = locked-up ownership of corporate IP/revenue. Different optionality, can stack
• Trump Accounts and the SEC accreditation test (39:32) - Calacanis cites Brad Gerstner / Michael Dell tailwinds and Paul Atkins’s plan to let people test into accredited status as parallel unlocks
• “The most ferocious form of capitalism ever invented” (41:21) - Mark on Bittensor: contests within contests within contests; no middle management because mediocrity gets emitted nothing
• Macrocosmos runs subnets 1, 9, and 13 (42:07) - Apex (1), IOTA (9), Daddy Universe / Gravity (13); previously ran 5 but downsized
• Why distributed training is the moonshot (42:24) - Frontier-model training is becoming “prohibitively expensive”; data centers the size of downtown Manhattan; Const’s original Bittensor inspiration was distributed training
• The math of distributed frontier training (44:08) - $200M compute run / 10,000 contributors = $20K each; SETI@home for AI
• IOTA install: two clicks (45:54) - Download a binary at iota.macrokosmos.ai, attach a wallet, machine becomes a training node when idle (10 PM-6 AM); was the first ready-packaged miner app on Bittensor
• Interruptible compute as the unlock (54:00) - Macrocosmos validated with 7-8 data-center clients that asynchronous, interruptible 20-minute slots at 10 cents on the dollar are what unlocks unsold capacity
• Macrocosmos is bootstrapped (55:55) - Stefan built the first Bittensor subnet ever; no outside venture capital
• The owner-lock-up takeaway from Stefan (57:27) - “More of a fiduciary duty to the people that believe in you… a measure of intention, a measure of conviction”
• Off-duty: Bieber vs. Sabrina at Coachella (62:24) - Bieber browsed YouTube on stage and sang over his old MVs; Sabrina drove a car off the set; Mark’s analogy is Bruce Springsteen on Broadway — meta-commentary only works once you have 20 years of catalogue
• The Staples Baddie pen recommendation (74:09) - Calacanis bought the Zebra G-750 ($14 on Amazon) because Kaden Roland, a Staples employee with 582K TikTok followers, recommended it — “she should be getting paid a million dollars a year”

Mentions

Companies

• Bittensor / Open Tensor Foundation (0:52) - Decentralized AI protocol with subnets, miners, and validators; native token TAO
• Covenant AI (12:55) - Sam Dare’s company; ran Templar (Subnet 3), Basilica, and Grail; trained a 70B-parameter model
• Stillcore Capital (9:19) - Mark Jeffrey’s TAO + subnet fund; ~80% subnet weighting today, moving toward 30/70 TAO/subnet
• Templar (Subnet 3) (12:55) - Bittensor’s flagship distributed-training subnet; the one Sam dumped
• Basilica / Grail (13:00) - Two other Covenant AI subnets affected by the rug pull
• Macrocosmos (24:03) - Will Squires + Stefan Cruz’s company; runs subnets 1 (Apex), 9 (IOTA), 13 (Daddy Universe / Gravity); fully bootstrapped
• BitMind (Subnet 34) (24:32) - Ken Miyachi’s AI-security / deepfake-detection company
• Stillcore subnet investments (10:29) - Ridges, Shutz, Hippias, Vidio, Score, Nova
• Uniswap (2:42) - Decentralized AMM that powers DeFi and that the Bittensor on-chain swap is modeled on
• Coinbase / Binance / Kraken (4:04) - Centralized exchanges referenced in the Uniswap explainer
• Arch Capital, Canonical Crypto, Mechanism Capital, Blockchain Builders Fund (33:00) - BitMind’s $3M seed investors
• Amazon (32:46) - Ken’s prior employer; recommendation systems
• Near (32:46) - Blockchain protocol Ken worked at before BitMind
• CoreWeave (53:01) - GPU infrastructure company referenced as a model for IOTA’s interruptible-compute partnerships
• OpenAI / Anthropic (46:00) - Cited as the inference revenue exemplars; Bieber-style “the five centralized AI companies”
• Plaud (6:00) - Sponsor; plaud.ai/twist; Calacanis bought one for his mom
• Sentry (9:19) - Sponsor; AI-powered debugging agent “Sir”
• Deel (19:23) - Sponsor; international hiring + payroll
• NetSuite by Oracle (30:13) - Sponsor; AI cloud ERP
• Athena (58:07) - Sponsor; AI-powered EAs at $3K/month; athena.com/jcal
• Staples / Office Depot / Walmart / Target (76:34) - Discussed in the Staples Baddie / authentic brand-rep tangent
• Papa John’s (77:41) - Has its own employee evangelist on TikTok
• Hulu / Austin Film Society (73:21) - Where Lon saw Sirât (the off-duty movie pick)

Products & Technologies

• TAO (0:52) - Bittensor’s native token; -25% on the rug-pull news
• Subnets (12:55) - Bittensor’s mini-economies; each runs miners + validators competing for token emissions
• Owner lock-ups (23:14) - Proposed smart-contract change so subnet ownership requires the largest, longest-locked stake
• IOTA / Train at Home (45:54) - Macrocosmos’s distributed-training subnet + binary client at iota.macrokosmos.ai
• BitMind Bot (28:24) - Grok-style “is this real?” assistant on X
• B200 / Blackwell GPUs (48:00) - $250K-per-unit compute Templar relied on; the “ribeye steak”
• H100s (51:31) - Cited as the kind of frontier compute IOTA can also use on a rotating-interruptible basis
• MacBook / Mac Mini (45:54) - The minimum-viable IOTA training node (~30-min slots)
• Synchronous vs. asynchronous training (54:00) - The technical bet behind IOTA: design for the slow / interruptible node, not the fast one
• DePIN (Decentralized Physical Infrastructure) (53:45) - The category IOTA’s Airbnb-for-compute model lives in
• Trump Accounts (formerly Invest America) (39:32) - $500-1000 per young person; Brad Gerstner / Altimeter Capital championed
• The Palace (0:52) - Mark Jeffrey’s 1990s online virtual community
• AngelList syndicates / SPVs (35:47) - Calacanis’s preferred analogy for picking subnets vs. buying TAO directly
• ActiveTrack (50:22) - Workplace monitoring software whose CPU readout would betray a covert IOTA installation
• Zebra G-750 (5:35) - Calacanis’s favorite ~$14 retractable pen, recommended by the Staples Baddie
• Sirât (73:21) - Spanish desert-rave thriller on Hulu; Lon’s off-duty movie pick

People

• Mark Jeffrey (0:52) - Stillcore Capital partner; 30-year internet OG; Bitcoin early adopter; built The Palace
• Sam Dare (12:55) - Templar / Covenant AI founder; allegedly dumped $10M in TAO
• Constantine “Const” Steeves (15:17) - Bittensor co-founder; built the original Templar codebase Sam inherited
• Jacob Steeves (13:00) - Other Bittensor co-founder named in the conflict
• Ken Miyachi (24:43) - BitMind co-founder/CEO; ex-Amazon recommendations, ex-Near
• Will Squires (24:03) - Macrocosmos CEO/co-founder
• Stefan Cruz (24:03) - Macrocosmos CTO/co-founder; built the very first Bittensor subnet
• Jensen Huang (13:00) - Discussed Covenant’s 70B model with Sam Dare on All-In; Calacanis’s “call up Jensen” punchline reappears here
• Brad Gerstner (39:32) - Altimeter Capital; champion of Invest America / Trump Accounts
• Michael Dell (39:32) - Cited as one of the whales rumored to back the Trump Accounts effort
• Paul Atkins (40:13) - SEC Chair; pushing an accreditation-by-test framework
• Carl Rinsch (17:44) - Netflix director who took the budget, bought crypto + luxury mattresses, now in jail — the perfect non-crypto rug-pull comp
• Dave Chappelle (71:49) - Off-duty story: post-Chase-Center late-night crowd-work set in San Francisco
• Justin Bieber (62:24) - Coachella 2026 weekend-1 headliner who browsed YouTube on stage
• Sabrina Carpenter (62:24) - The other weekend-1 headliner; full pop spectacle, drove a car off-set
• Jimmy Iovine (67:05) - Mark’s recent dinner companion; produced Dire Straits’ Making Movies and Springsteen’s Born in the U.S.A.
• Bruce Springsteen (67:24) - Springsteen on Broadway as the meta-commentary archetype
• Roger Waters (69:53) - Mark and Calacanis caught The Wall live tour together
• Kaden Roland (the “Staples Baddie”) (75:23) - Staples store employee; 582K TikTok followers; sold Calacanis on the Zebra G-750
• David Zaslav (31:46) - Cited as a possible (but denied) candidate for the studio head who told Calacanis he’d read Angel

Surprising Quotes

“It looks like, and this is sort of allegedly what it looks like on chain, it looks like this is what Sam did. And it also looks like that his developers, people on his team, were caught blindsided by this also.” — Mark Jeffrey, 16:30

“If the trail is true, then essentially one might speculate that a large amount of money was at stake and it would be like somebody raising venture capital and saying, ‘Wow, there’s millions of dollars in the bank. I should deposit it in my personal account and shut the company down.’” — Alex Wilhelm, 17:11

“This is the most ferocious form of capitalism ever invented as far as I can tell. So, yeah, I love it.” — Mark Jeffrey on Bittensor’s incentive engine, 41:21

“Specifically in the US, AI is more unpopular than Trump… The reason there is because no one can participate in the economic system that is building. And what’s so exciting about Bittensor is these are all extremely early stage companies that you wouldn’t ideally have the ability to participate in in any other way.” — Ken Miyachi (BitMind), 38:49

“We actually used to run five subnets. Um, but we decided that we would — we would like to live long enough to actually see these things reach the market, so we had to — we had to downsize a little bit. Running a subnet is no joke.” — Stefan Cruz (Macrocosmos), 42:26

“She’s probably done more for the brand of Staples with young people and all the way up to Gen X myself, than Staples could ever do with a hundred million dollar ad campaign… She should be getting paid a million dollars a year. Minimum.” — Jason Calacanis on the Staples Baddie, 76:00

Transcript

Note: speakers below are taken directly from the interview-transcriber JSON. The four labels in the JSON are Mark Jeffrey, Jason Calacanis, Lon Harris, and Alex Wilhelm; later guests Ken Miyachi (BitMind), Will Squires and Stefan Cruz (Macrocosmos) were not seeded as separate speaker labels and so their lines are merged under the host labels. Where context makes the actual speaker clear, it is named in the prose summary, highlights, and quote attributions above.

Mark Jeffrey: 0:00 Do I believe that Sam’s epistle about why he left because of decentralization is correct? No, I do not.

Lon Harris: 0:06 If somebody runs a subnet, they do get a little bit of power. And they can do what’s called a rug pull. If you have leadership and you have responsibility, you can mess with it.

Mark Jeffrey: 0:15 The reason why this all happened, the incentivization alignment engine works spectacularly well, but then it fell out of alignment when there was sudden success. It created this sort of overwhelming temptation. It looks like, and this is sort of allegedly what it looks like on chain, it looks like this is what Sam did.

Alex Wilhelm: 0:32 And this is all allegedly and a lot of ifs here, but if the trail is true, one might speculate that a large amount of money was at stake and it would be like somebody raising venture capital and saying, ‘Wow, there’s millions of dollars in the bank. I should deposit it in my personal account and shut the company down.’

Jason Calacanis: 0:52 All right everyone, welcome back to Twist. It is Monday, April 13th, 2026. E 1977 something. There was a lot of hand-wringing in the Tao BitTensor community over the last week or so, and there’s been a lot of progress made. So to break it all down with us, Lon brought on my friend Mark Jeffrey, known for 30 years, who was an early OG in Bitcoin, created one of the first online virtual communities, The Palace, back in the 90s when many of you were not born, and is one of the OGs of the internet online space and also now crypto and Tao distributed computing. Welcome back to This Week in Startups, Mark Jeffrey.

Mark Jeffrey: 1:40 Thank you for having me, sir. Great to be back.

Jason Calacanis: 1:42 Mark started telling me, Lon, about BitTensor and Tao because I was like, ‘Hey, what’s going on?’ because I think crypto’s turning a corner here. It’s becoming legal, it’s becoming regulated in a very intelligent way. And I’m curious, is there anybody doing anything where a consumer or an enterprise gets value other than store of value and money transfer? Very well established uses of crypto of course, between Bitcoin and you know stablecoins, great to transfer money around and great to, you know, speculate and all that stuff, but I care about the application layer and actually some value being created. He said, ‘Yeah, let me tell you about this thing, BitTensor Tao and the subnets.’ And I was like, ‘That’s that’s legit.’ And then I watched the people getting involved in it and said, ‘These are legit people.’ As opposed to crypto, Mark, which I think you would agree, you had the early cypherpunks, cryptoheads, and then it kind of got hijacked a bit along the way, but…

Mark Jeffrey: 2:42 I would I would say that within the Ethereum community, when that all booted up, when decentralized finance booted up, you had the invention of Uniswap and the decentralized exchange and the invention of decentralized lending. And that was valuable. These are financial products. So they’re not really, you know, products in the way you and I are used to thinking… to seeing belt, but they do have value and they’re not like sort of nonsense. But I’d say 80 to 90% of it is nonsense. When I encountered BitTensor and started digging around the subnets, you know, having been a participant in many early ecosystems including the 90s as you, you know, talked about earlier as well as early crypto, what I saw in BitTensor was quite a lot of early signal. I would say, you know, half the teams were doing something of value and were decent hypotheses. You don’t know whether these hypotheses are going to pan out to become very large companies or phenomena, but they are pretty good. So I’ve never seen an ecosystem where there was more early signal than the BitTensor ecosystem. So that’s one of the reasons why I got very excited about it.

Jason Calacanis: 3:51 Maybe explain Uniswap. You mentioned that. I’m sure people in the audience are going, oh, that is real. Uniswap is kind of an automated market maker of some type where people can swap value, yeah?

Mark Jeffrey: 4:04 Yeah. I mean, it’s really the decentralized Coinbase, right? Yes, everything you said is correct, but a lot of people don’t know what an automated market maker is. It is technically called that, an AMM. But the thing that was really that was happening before Uniswap, if you wanted to get your coin listed on a—you had to first of all get it on a centralized exchange. There was no other option, which meant you had to pay Binance or Coinbase or Kraken or somebody usually like a million bucks to get the listing, right? And then you had to provide your own market making, which meant that you had to provide a bunch of other coins to trade against your coin when it listed on the exchange. The exchange would not provide that liquidity, right? So along came Uniswap and they said, why don’t we do something like a Coinbase or a Kraken or a Binance, but the entire thing is just powered by smart contracts and lives on the chain? Nobody owns it, nobody controls it except for the smart contract, and we attract liquidity by promising some portion of the trading fees to anyone who provides the liquidity to the smart contracts. So anyone could become a fractional owner through providing liquidity of an exchange, of this decentralized exchange. So you had two real big advantages: one, you had permissionless listing. Anybody could list any coin so long as they were able to define the swap pairs on Uniswap. And two, the community provided the liquidity. And this worked extraordinarily well and is basically the backbone of all DeFi today.

Lon Harris: 5:35 Lon, we’re off to the races here. So much interesting information, hard to keep track of, and I’ve got my notepad here and I’ve got my favorite pen. This is my zebra G-750. But as fast as we’re talking, I’m not going to get it all down, right? I might write down two or three words. It’s not going to be structured. I’m going to have to go back. I can’t keep up and I’ve got to be present here. So what do I want to do? I have my…

Mark Jeffrey: 6:00 Plaud pen, I press the button. Boom, I get a haptic. Now it’s being recorded and it’s going to be summarized, red light on it, so I’m not like covertly recording anybody. I’m just keeping my notes. So when you’re in a meeting, if you don’t have your phone handy and you don’t want to flip through, you know, your different apps to try to record something, the Plaud stores it all. When I put it in the cradle at night to charge the battery, and the battery lasts for hours, it seems like I charge this thing every two weeks and I use it constantly, it syncs everything and then it automatically runs it against an LLM and does a summary. It’s one of the more brilliant products and I’m addicted to it. I hear you got one for your mom, huh?

Jason Calacanis: 6:35 I did. My mom watches our show and she’s like, “You know, they should be advertising these to seniors, not executives, because I’m forgetting stuff constantly.” So she’s hitting it like all during the day, remind me to do this, update my calendar for that, and it’s just keeping her more like in the flow and helping her sort of dodge her forgetfulness. And it’s the smartness of it. It’s not just taking notes for you, it’s filing them, it’s arranging them, it’s coordinating them, it’s keeping you organized, all with just hitting your Plaud pen. So if your work relies on conversations, you too need a Plaud note pen just like my mom. Check it out at plaud.ai/twist and if you use the code twist, they’re going to give you 10% off. Okay, so thanks Plaud. Let’s get to Tau and BitTensor. What attracted you, yeah, what attracted you to this specific project and why, Mark?

Mark Jeffrey: 7:21 I remember talking to you at a conference, one of your conferences where you said, you know, “Pivot from crypto to AI,” right? And I thought about that because you’re a smart guy and, you know, you have advice like that, I’m going to listen. And my conclusion was, well, you can’t really pivot to AI because the only people who are going to own AI are the big players, right? There’s like five companies, right? You have to have giant iron to compete. And anything other than that is you’re just, you know, an API layer on top of one of those five companies and you’re probably, you’re probably just running into a buzzsaw. That was my initial conclusion. And then, you know, with BitTensor, I saw a way to sort of, you know, fuse the crypto stuff with the decentralized version of what the centralized people were doing. And the more I understood what was going on there, the more I realized that this was sort of the Linux approach to, you know, it sort of reminded me of the OS wars when you had OS/2, you had Microsoft NT, you had Sun Solaris, but who won? It was Linux, right? Linux powers 90% of our world today. And what I saw with BitTensor were sort of the early moments of what looked like an AI Linux. OpenAI, but like what OpenAI should have been, truly open AI.

Jason Calacanis: 8:41 I wasn’t saying crypto’s bad in any way, just saying if you’re in crypto, I would pivot to AI because there’s just such a much more, there’s so much more application for AI because intelligence touches everything, whereas with crypto, finance doesn’t touch everything. Finance people might say, “Yeah, finance touches everything.” Finance doesn’t touch me writing a poem. Finance doesn’t touch me going and…

Alex Wilhelm: 9:00 You know, cooking a meal or something. Like it doesn’t touch everything, but intelligence does, right? And and that was very similar to I think what the Tao people were saying was hey AI is so complex and there’s such a need for compute, there’s such a need for transport moving packets around. Hey, maybe there’s something here.

Jason Calacanis: 9:19 You’re running essentially a fund like a venture fund, but for Tao. I’m a partner in it, I’m an LP in it, I’m deep down the rabbit hole on this. But you’re picking subnets in which to buy those tokens and invest in. So what are some of the ones you’ve invested in and why? Debugging sucks, and it takes up time your team could be spending on awesome new features and products. But now there’s a better way. Sentry’s AI-powered debugging agent, Sir, isn’t just guessing about what might have gone wrong with your system. It’s analyzing your actual data. And because it has context, it spots buggy code before that code ruins your entire day. Plus, Sentry works alongside coding agents like Cursor, passing along an idea for a fix that can then be applied to your codebase and sent to a human team member for review. End-to-end automation, all the way from bug detection to pull request. So join the millions of devs and companies like Claude and Disney+ who use Sentry to move faster. Check them out at S-E-N-T-R-Y.io/twist and use the code twist for $240 in Sentry credits.

Mark Jeffrey: 10:29 Yeah, so we’ve invested in about eight so far, and we, you know, sort of the big ones are Ridges, Shuts, Hippias, Vidio, Score, and basically all of these. What we look for, we’re sort of conservative investors so far as one exists in the Bittensor ecosystem in that we look for subnets that have product-market fit and revenue primarily, right? So it’s pretty much the same criteria that you use probably to judge startups, right? We look for mature team, we look for mature products, and there’s a class of product inside of Bittensor that I call the research subnets. And these are people that are trying to do extraordinary things that don’t necessarily have an obvious product-market fit right out of the gate, but if they work, they could be absolutely enormous, right?

Alex Wilhelm: 11:24 Exactly.

Mark Jeffrey: 11:26 Yeah, so you know, so a great example of that, well, Nova is a great example of that, right? So Nova is hunting pharmaceutical molecules, and it’s going to take a while for them to probably find one, but if they do find one, it could be a billion-dollar subnet kind of overnight, right? But it’s very speculative, right? We’re mining for gold, basically. We might not find it.

Jason Calacanis: 11:44 Now Lon, it’s important. Incentives matter, and so I want to sort of tee up this story here. If somebody runs a subnet, they do get a little bit of power. They have the subnet. They recruit miners, they have validators, making sure the work was done properly. But they’ve become like a small business owner and they can do what’s called a rug pull. So they could either be virtuous and they can provide people with this incredible opportunity to give an… give knowledge through a large language model or an algorithm, they could give compute, they could give storage, they could contribute something in order to win the prizes of Tau or the subnet token, which is, you know, essentially one and the same in some ways. So they could earn money, but they can also, if you have leadership and you have responsibility, you can mess with it. So we had last week somebody running one of the subnets decide they wanted to take their marbles and leave the game, and that’s the core of what happened. Yeah, Lon?

Lon Harris: 12:55 We actually had spoken with Templar, which was the subnet in question. Sam Dare is the sort of creator of it. And so, yeah, what’s happened: Sam Dare, he dumped 37,000 Tau on his own subnet holders. It’s about $10 million in money and he sort of walked away from the entire project. It was Covenant AI, the designers of that 70 billion parameter model you’d spoken about it with Jensen Huang on All-In, we’d discussed it with Sam here on the show. So, you know, he sort of walked away from Covenant AI. They’ve been operating three subnets on Bittensor: Templar, Basilica, and Grail. Templar was sort of like the, you know, the big one, the high profile one, was the reason that Tau had sort of been taking off, getting a lot more mainstream attention, the coin had sort of gone up in money. He and Const had some sort of—we’re still getting sort of all of the details—they had had some sort of back and forth. He felt that Const, Jacob Steeves, one of the co-founders of Bittensor, was sort of overstepping his authority. There was some sort of conflict between the two of them. And then, as you said, Sam sort of took his marbles, walked away, sold 1% of the alpha holdings on three Covenant subnets that were near 100%. And so… yeah, so that is sort of the situation. 37,000 Tau, that was kind of the rug pull, that $10 million. And now we’re sort of in the aftermath. The price of Tau dropped about 25% in the immediate aftermath.

Jason Calacanis: 14:25 I don’t know if you were on the episode when this gentleman was on, but I found him a little squirrely.

Alex Wilhelm: 14:31 I was.

Mark Jeffrey: 14:32 Yeah.

Lon Harris: 14:32 Yeah, we discussed this after the show.

Alex Wilhelm: 14:34 I mean, big personality, a funny guy. Like, it was a great guest on a podcast because he had very big outspoken opinions and he was a big personality guy. I really enjoyed talking to him.

Lon Harris: 14:45 But we did discuss this afterwards and there was also an edge to him. It was a little bit hard to put your finger on. There was something a little unpredictable about his demeanor and…

Jason Calacanis: 14:54 Yeah, and how it… what was coming next, I kind of pushed him on, ‘Hey, what are you doing next?’ and ‘How does this work?’ and the…

Alex Wilhelm: 15:00 The answers Mark weren’t as crisp or as tight as I might have expected from somebody running something significant, let’s just say. Yeah. So what is the back channel here? And maybe you could explain Const, the founder of Bitensor and his role in all this.

Mark Jeffrey: 15:17 So Const is Constantine just fyi.

Alex Wilhelm: 15:20 Oh I did not know that. I did not know that.

Mark Jeffrey: 15:23 Yes, I know, a lot of people don’t know that. So Const actually built the initial subnet version of Templar 3 and Sam used to work directly for Const at the Open Tensor Foundation. And, and so Const basically bought the subnet for him, wrote the initial version of the decentralized training environment Templar and handed it off to Sam, right? So he gave Sam his start. And, and then basically what, you know, what we discovered here is, you know, let’s just stop for a second because what happened here, the reason why this all happened is because something in Bitensor worked extraordinarily well. There’s several things that are working pretty well, but it’s because something worked, the incentivization alignment engine worked spectacularly well, but then it fell out of alignment when there was sudden success. And it created this sort of overwhelming temptation for, you know, a subnet owner who, when you’re a subnet owner, you acquire a large number of your subnet tokens. They’re emitted into your wallet, right? Because you get paid effectively by the chain for being a subnet owner. And if you have enough of them and there’s the liquidity pool in the on-chain Uniswap that I described earlier, you can at any time just sort of flood that Uniswap with your tokens and exchange it back out for TAO. And if you have the majority of the tokens, you can get the majority of the liquidity pool. And now you’ve got TAO, now you can run off to Binance, sell the TAO for dollars, and now you’re gone. Which, it looks like, and this is sort of allegedly, this is what it looks like on chain, it looks like this is what Sam did. And it also looks like that his developers, people on his team, were caught blindsided by this also.

Lon Harris: 17:08 And this is all allegedly and a lot of ifs here.

Alex Wilhelm: 17:11 But if the trail is true, then essentially one might speculate that a large amount of money was at stake and it would be like somebody raising venture capital and saying, ‘Wow, there’s millions of dollars in the bank. I should deposit it in my personal account and shut the company down.’

Jason Calacanis: 17:29 Which is actually one of the big fears in venture capital. We run on a trust-based system and it has happened. You can go look at the history of venture capital. Every year or two, somebody takes the money and goes to Vegas or absconds to another country.

Lon Harris: 17:43 Sure.

Alex Wilhelm: 17:44 There was just that story about the director, the Netflix director Carl Rinsch, do you remember this? Netflix paid him to make like an eight-episode sci-fi series for them and he just took the money, bought crypto, bought a bunch of luxury mattresses, bought some cars.

Jason Calacanis: 17:57 He’s doing a little jail time right now for that, but—

Lon Harris: 18:00 Yeah, it happens in every industry where people get paid up front and then they walk away with the bag.

Alex Wilhelm: 18:04 Yeah. And important to say we don’t know…

Lon Harris: 18:06 Right.

Alex Wilhelm: 18:07 if that’s the case here, but that is part of any capitalist system. So there’s no way to stop this unless you have a board of directors, if you have a controller, a CFO, and then even if you do, people, you know, the board meets four times a year and then there’s 361 days in between those board meetings where shenanigans can happen.

Lon Harris: 18:32 I guess I had sort of a two-part question I’d love to throw both of you about this. One is do you guys believe that Sam had legit grievances against Const or the system? He’s been saying this is activism, like Bittensor is overreaching, it’s not truly decentralized and I had no choice but to do this. So one, do you think is there anything behind that at all or do you suspect that this was just that’s how he’s explaining himself after the fact that he just wanted the money? And then two, how would you have preferred a subnet owner react if they do feel like I want out of the Bittensor community, I’m not happy with the way it’s being run, what should Sam have done to avoid rug pulling all of the people who believed in his project? Allegedly, but still separating himself from Bittensor more generally.

Jason Calacanis: 19:23 Founders scale faster on Deel. That’s the deal. You can grow your company without borders and you can set up payroll for any country in minutes. Hire anyone anywhere like a modern startup or large company does and Deel is gonna get all the visas handled fast so you can get back to building. There is a great talent war that’s going on right now and you need people with superpowers for your startup to be competitive to beat your competitors to get your products to market. But anytime you try to grow your team with overseas hires oh my lord you’ve got to reinvent the wheel and you got to navigate a tangled web of international laws, regulations and you can’t get these things wrong folks. You want to onboard new staffers in other countries, you want to get them set up on your network nice and secure IT access all that good stuff? You want to manage their benefits? Trust me, this is all a nightmare unless you partner with Deel. They are the people staff for startups. They’re going to take care of all the onboarding, payroll, HR, IT, benefits, everything you need quickly in one place done perfectly. So visit deel.com/twist. That’s d-e-e-l.com/twist.

Mark Jeffrey: 20:24 Do I believe that Sam’s epistle about why he left because of decentralization is correct? No, I do not. And you can see this evidence, like nobody in the ecosystem believes that. And you can see this from all the posts from all the subnets, everybody saying, you know, my subnet is Bittensor and Bittensor is my subnet. There’s a ton of these posts.

Lon Harris: 20:44 Yeah, I’ve seen them.

Mark Jeffrey: 20:45 It’s basically the Bittensor version of I am Spartacus, right?

Lon Harris: 20:48 Right.

Mark Jeffrey: 20:49 And everyone is standing up and saying no, no, none of this stuff is true. And Const gave a very, I thought, thoughtful reply to the letter that Sam posted as you know his rage

Alex Wilhelm: 21:00 …letter basically, right? Const replied to it and it was a very legitimate response. So, um, yeah, so I think that I don’t believe, you know, I don’t think there’s anything to his claim of decentralization. Nobody else feels that way. So.

Jason Calacanis: 21:15 Purely hypothetically, Sam had a legitimate grievance, you’re running a subnet, you’re like, I don’t like the way BitTensor is being run. How should he have dealt with it?

Alex Wilhelm: 21:22 How should he have dealt with it?

Mark Jeffrey: 21:23 He should have gone to Const and said, you know, look, I’m not happy. Um, you know, let’s find a way to amicably part ways.

Jason Calacanis: 21:29 Right. Okay.

Mark Jeffrey: 21:30 Let me hand off - let’s - let’s figure out how - you know, how many - how much tokens - how many subnet 3 tokens I should be allowed to keep.

Jason Calacanis: 21:41 Right. Okay.

Mark Jeffrey: 21:42 And then I should hand over the subnet private key to the foundation to be redistributed to somebody else to run, and Const could figure out who that was, or the foundation could figure out who that was. Basically negotiate an exit, which, you know, he gets something for his contribution to date, but then the subnet and the token holders are allowed to carry on without him.

Jason Calacanis: 21:56 Right.

Mark Jeffrey: 21:57 That’s what he should have done.

Jason Calacanis: 21:59 Mark before we get to our guest,

Alex Wilhelm: 22:00 Yeah.

Jason Calacanis: 22:01 how can we avoid this in the future and who owns a subnet? And who should own a subnet? Should it be the miners themselves? Or should there be governance here? Should there be a 51% rule where the people contributing to it own it? What’s the best practice here? Because, hey, as an investor, if I wanted to back Lon to do, you know, an algorithm on - and do a subnet, I would be like, okay, I want to own it. I want Lon to own it. I want the community to own it. How does that get codified outside of the Delaware C-Corp, the Texas, Nevada corporation where we set up a board of directors, where we set up shares in a company, we have preferred shares and that whole architecture which is old but trusted.

Mark Jeffrey: 22:52 The architecture in crypto is smart contracts. So the answer to this is to implement a new smart contract or change at the chain level in BitTensor’s case. The problem is that Sam had access to all of his tokens and could spend them in one gulp before anyone knew what was happening, right?

Jason Calacanis: 23:13 Right.

Mark Jeffrey: 23:14 He had information asymmetry, which - and he controlled the subnet, so he could just dump this giant amount of tokens on the market.

Jason Calacanis: 23:22 Right.

Mark Jeffrey: 23:23 So the answer is to lock those tokens. To basically make it so he can’t spend them. And, you know, if he’s operating the subnet, you can’t both operate the subnet and have unlocked tokens. Like there has to be, you know, if you’re operating the subnet, then your tokens are locked. And there are - there’s a couple different proposals underway as to exactly how to do this. Const released a new one this morning where basically the ownership of the subnet is always up for grabs based on who locks the most tokens for the longest period of time. That’s effectively the mechanism. So you might lose your subnet to someone who has more conviction. So the idea is that the chain will reward capital conviction in a subnet. So it is possible you’ll lose your subnet to somebody that believes in it more than you.

Jason Calacanis: 23:57 We have some guests Lon. Let’s bring some of them in.

Lon Harris: 23:59 Alright.

Jason Calacanis: 24:00 Our guest’s on and we’ll have a talk about some of the other subnets out there I understand.

Alex Wilhelm: 24:03 Yes, speaking of great subnets as we often are, we have three more subnet founders joining us on the show today. First off, from Macrocosmos, they also run three individual subnets. We’re going to welcome Will Squires, the CEO and co-founder, and Stefan Cruz, the CTO and co-founder. Gentlemen, thanks for being here. Also joining, he is the co-founder of subnet 34 BitMind. They’re a system for detecting deepfakes and other AI-generated content. Give it up for Ken Miyachi. Ken, thanks for being here.

Jason Calacanis: 24:32 Okay. Alright, so maybe Ken you can explain a bit about your subnet, who you are, and why you created it.

Mark Jeffrey: 24:43 Yeah, thanks for having me on. Hey everybody, I’m Ken. I’m the co-founder and CEO of BitMind. So we are an AI security company, and we’re very focused on deepfake detection. We have a variety of consumer applications where if you’re scrolling X or you’re on the internet, you can put in a video, image, piece of audio, etc. It will tell you if this is real or fake. And then we also have a bunch of enterprise services where we create custom models for essentially businesses trying to secure their communications, their data, etc. And what we’re really focused on right now is going towards a proof of human. There’s a lot of grifters, there’s a lot of North Koreans, fake people trying to infiltrate your systems, and just AI slop in general throughout the internet, bots trying to sign up and use your services. And we are creating services to essentially detect and stop that type of activity from happening on the internet.

Alex Wilhelm: 25:37 And how is this done? Is there an algorithm that analyzes a piece of data and says, ‘Hey, this is human, this is likely an LLM, this is a deepfake, this is a real photo’? And what are the miners contribute to this? They have their own algorithms or their computer or both?

Mark Jeffrey: 25:56 Correct, yeah. So essentially the miners are competing on developing the models themselves. These detection models that essentially in different modalities, whether it be image, video, or audio, that essentially they win the competition, they earn our subnet token if they produce the highest accuracy, most high-fidelity models. And then on top of that, we also have another set of miners that are actually incentivized to generate data and try to fool them. So it’s a continuously learning system trying to improve over time.

Alex Wilhelm: 26:27 So you have a white hat and a black hat. You got the black hats coming in and saying, ‘Hey,’ or or red team, I guess, might be a way to say it even better. ‘Here is some content, and we’re going to make a generative AI picture of Lon, but we’re going to make him, you know, 10% more robust, or we’re going to GLP him and make him 10% thinner,’ and then they try to trick your the other miners and that makes their algorithm sharper. So steel making steel, you know, steel on steel making it stronger.

Mark Jeffrey: 27:00 Exactly, trying to give the chisel jaw, I don’t know, whatever, whatever you want, right? Um, sure. But why this is so important is because generative AI and AI in general, right, is the velocity in which these new models coming out is so fast. And so if you did this in a traditional way, what you’d have to do, wait for C-dance to come out, wait for, you know, nanobananapro2 to come out, generate a bunch of that data, retrain your detection model. And what we have is an open decentralized permissionless system where people are generating data for us all the time and people are also coming up with new detection algorithms, retraining quickly, and so we can actually update our system on a daily, weekly, at a very fast cadence to always stay ahead of the newest generation models.

Jason Calacanis: 27:44 So, okay, this seems like an incredible game you’ve constructed. One team is truth-maxing, one is fake-maxing, they battle it out, their algorithms get sharper, fantastic. But is there a commercial product that is created for enterprises or consumers? Can the New York Times or could X, Twitter say, ‘Hey, we want to put these trending posts in here and then maybe in the community notes say, ‘Hey, it has this score, you know, from your subnet, you know, and it’s a certified 98.7% truthy or, you know, truth-maxed’?

Mark Jeffrey: 28:24 Yeah, definitely. So I think the two, the two enterprise use cases you touched upon were essentially social media and news outlets. So with news outlets, right, that’s a really important use case for us. If you are in traditional media and media at all, you want to make sure what you’re reporting on is actually real. And you lose trust and credibility, which they’re already struggling with. And so, right, as a reporter, you’re going to want to verify whether this image, whether this data that you’re receiving is real or not. And so there’s definitely use cases like that where you can either drag and drop a image or video, or you can use our API, whether, you know, however way you want to use it, to determine that. And then on social media, it’s a really big use case. We have something called BitMind Bot, which is very similar to Grok, but it has a specific focus on, you know, ‘is this real or not?’ But the other use cases I think on the enterprise side are really about securing essentially business communication. There’s been a lot of recent drama around interviews, whether you’re…

Jason Calacanis: 29:21 Yes.

Mark Jeffrey: 29:22 …like, I think there’s two main use cases. There’s one where it’s like just this person is completely fake. Yeah, and like, right, the person can’t put their hand in front of their face or something like that, and they’re trying to infiltrate your organization. But then there’s another use case where it’s just like actually grifting people trying to get multiple jobs and, and putting on a facade of themselves. And so we’re trying to stop both of that. And then the other actually really big use case that is pretty shocking, but not a lot of people are thinking about is actually in government. So more and more there is evidence being, uh, being shown in cases in a…

Alex Wilhelm: 30:00 A wide variety of contexts that is AI generated and you need to be able to prove this. You know, the uh, the this is AI generated is the new my dog ate my homework, right? Like they say something damaging about you, you just say it’s AI generated.

Jason Calacanis: 30:13 We say it on every episode these days, the AI revolution is here, and sitting on the sidelines is no longer an option. No more waiting outside with NetSuite by Oracle, you can start putting AI to work for your business today. NetSuite is the number one AI cloud ERP and it’s trusted by over 43,000 businesses. It’s not just an AI add-on or a chatbot that sits in your browser. No, this is a unified source of truth that brings together all the data you need to run your business. From software and IT services to healthcare, equipment manufacturing, financial services, and many other great American industries, NetSuite delivers a customized solution for your business. If your revenues are at least in the seven figures, get the Daily Business Guide demystifying AI at netsuite.com/twist. The guide is free to you at netsuite.com/twist. What’s interesting to me, Mark, is just putting on my angel investor hat, go read the Angel of the Book, and in 13 or 14 languages. It was really nice, I was at a major Hollywood party last night. Just very famous studio head lon came up to me and said, ‘You’re Jason Calacanis’ and I said, ‘Yes, I am.’ and this is in like a room full of celebrities or whatever. He said, ‘I read your book. Um, it’s just like movies. You would have been a great studio head.’ Um, anyway, enough promotion of my book. It was one of the great moments of my life that in a room where I assumed nobody knew who I was, a studio head, a literal studio head had read the book.

Alex Wilhelm: 31:46 David Zaslav, this is what he’s doing with his time now.

Jason Calacanis: 31:50 It wasn’t David, but it was on that, even bigger than that. But anyway, it’s as crazy as it seems. I won’t say who. Uh, but I’m looking at this through my angel lens, Mark and saying, wow, um, there is a business here. Uh, I don’t know exactly what it is, but you can see Ken is doing product market fit, he’s doing a go to market function where hey, we’re trying to find customers and uh, he’s also raised $3 million for this project as well, correct Ken? And in who put $3 million into the company? And then Mark as an investor, um after Ken tells us who put the $3 million in, I want you to explain to me how I should think about this as a partner in Stillmark and an LP with, you know, hundreds of thousands of dollars of my own personal money at stake versus hey, should I invest in Ken’s corporation? So Ken, maybe you can tell us who invested the 3 million? Is it crypto people giving you tokens or is it like a venture capital firm?

Lon Harris: 32:46 So my background, I come from both AI and crypto. Um, I worked at Amazon for a couple of years doing recommendation systems and then prior to that, I was at a blockchain protocol called Near. And so I had relationships with a variety of VCs in both of those spaces.

Mark Jeffrey: 33:00 Arch capital, canonical crypto, mechanism capital, blockchain builders fund. There’s a wide variety when we raised our seed fund about a year ago now.

Jason Calacanis: 33:11 And how did they think about… Well, let me have Mark answer the question of where is value going to be created and how should I think about this because now I’m getting pulled in both directions. I’ve got my venture firm here and then I’m a partner in Stillcore. Okay, I’ve placed my bet here but should I also be like, hmm, maybe launch or my syndicate should talk to Ken about investing in his corporation.

Mark Jeffrey: 33:32 Yeah, I think it’s very different in Bittensor than it was in the old Ethereum and Solana days where you had this token or equity kind of bifurcation and they both appeared to be ownership of the company. The ownership of the token in Ken’s subnet — the subnet that he operates and runs is actually the product department of a much larger organization, the company, right? So the business and the intellectual property so much as it exists, um, the revenue stream of his company, that the equity is how you own a piece of that. The product department and the engine that makes his product possible, being a, you’re basically a part owner in a Bitcoin-like fashion of that piece of it and you’re always liquid, right? With his corporate owning equity in this corporation, you’re not liquid, right? You’re basically tied up for whatever period of time. Um, but you can sell anytime you want as we’ve seen in Bittensor, right, recently. You can in fact sell those tokens, right? So, um, it just depends on, on which one you believe in more and, and what optionality you want to have.

Jason Calacanis: 34:39 So it’s interesting Lon, in the way one way to think about this is if you owned equity in Airbnb the corporation Lon, you could also be an Airbnb host and make money in the marketplace but you could also own Airbnb and I do know people who were Airbnb investors who were like, I want to start an Airbnb or there are people who were Tesla investors who want to, you know, have a fleet of robo-taxis and manage those. So, uh, or it could be you are running a corporation at scale and the corporation has debt and you can buy corporate paper, uh, that is the loan that people pay 12 or 15% on and so you could own Tesla and you could own Tesla’s debt or you could own Uber and Uber’s debt or Amazon and Amazon’s debt. So there’s multiple ways to, to get exposure I think. What do you guys think just in general, if I’m a regular person watching this who’s like, I like the cut of this project’s jib, I want to get in on it, would it just make sense to go to a place like Kraken or Coinbase and buy the Tau token? Is that a valid way to get exposure here or should you really be thinking about specific subnets and sort of digging in deeper?

Alex Wilhelm: 35:47 This is an important question I think. I, I think the easiest thing to do is buy a little bit of Tao, you get the mutual fund effect, which would be like if you asked me should I be an angel investor or should I put money into a VC fund.

Jason Calacanis: 36:00 If you put money into a VC firm, you’re saying I trust the partners there to make, you know, 20 bets. And uh, I don’t make those 20 bets, but in 10 years, I hopefully get back more than double my money and I beat the public markets. Uh, but if you want to be an angel investor and you read my book and you join AngelList and the syndicate and you join other SPVs, you get to pick and choose which ones you bet on and you get to make your own decision and some people like that style. In other words, you can back somebody to go play a series of poker tournaments—there are people who do that—and there are people who want to play in the poker tournaments. Depends on your—I would say it’s basically a time commitment thing. I don’t have the time to evaluate every subnet, so I’m a partner and investor in Stillcore Capital, which Mark is doing. All of these things are incredibly speculative; 90% of the bets I make go to zero. So you need to have a certain constitution, I think is the way to say, yeah, Lon?

Lon Harris: 36:53 Yes, I could not afford to have 90% of my bets go to zero. That’s definitely not on the table.

Jason Calacanis: 36:56 Well, but you could afford to have 90% of 10% of your bets go to zero, which would be 9% of your overall net worth. So that’s another way to think about it, right? If you’re going to put 5 or 10% of your net worth into speculative things. Mark, any thoughts here of how an individual—without giving financial advice, but just how you think about this—of where the value could be created? Stillcore has probably, I don’t know if it’s half or two-thirds of its money in Tau and half or a third of its money into subnets at this point. What’s the blend you think you should have?

Mark Jeffrey: 37:15 Yeah, I mean it’s about 80% subnets right now, but I want to back that off a little bit to, you know, maybe 30-70. Um, but I think because I think we’re in a unique period right now where, you know, if you make the right early bets, it’s going to be extraordinarily rewarding. And you have to know, you know, it’s basically evaluating the subnets and the companies around them and the products like you would a regular startup. Knowing when to get in, also knowing when to get out, right? And this is a highly, highly, highly dynamic environment as we’ve seen recently. So sort of scrambling around and getting to know everyone and everything and knowing as much information as possible is more than a full-time job. And that and that’s, you know, you can do that if you want. There are people who are doing it. Um, but we we take over that function for accredited investors if they’d like to park their money with us and not worry about all that stuff, right? So… However, if you’re a normal person in the normal world and you’d like to play in this arena, it is open to you. You don’t have to be accredited, right? So that’s one advantage to betting on subnets as opposed to going trying to get into equity. Most people can’t get in equity, right? So… So this is a way for you to invest in the AI future as an unaccredited investor that you normally don’t get.

Alex Wilhelm: 38:49 I kind of want to touch on that too. I feel like that’s one of the most important things. I think something that’s so interesting that specifically in the US, AI is more unpopular than Trump. And I think a lot of…

Mark Jeffrey: 39:00 The reason there is because no one can participate in the economic system that is building. And what’s so exciting about Bittensor is these are all extremely early stage companies that you wouldn’t ideally have the ability to participate in in any other way. And this is- it’s kind of a double-edged sword, right? Your early-stage startup is getting marketed, you know, is getting valued by the public market, but the public can participate in it. And I think that’s what you need to do to keep AI beneficial to all humans, have everyone that’s participating um have economic upside here.

Jason Calacanis: 39:32 Which, you know, I think everybody in America- this is- like really leveling up the discussion. But it’s a great point, Ken. Um, the- there is Invest America accounts, now called Trump accounts, where they’re giving every young person five hundred or a thousand dollars and companies are going to start to contribute. There’s some other whales coming into this system, I understand from Brad who told me there’s going to be some other Michael Dell type people who will probably contribute. I’ve heard Brad Gerstner of Altimeter Capital, who, you know, championed this idea, give him a lot of credit for that. If people could- more people could participate. And then I interviewed the CEO, the- the new head of the SEC, and he’s- this is his second stint as SEC chair, and he is all in on-

Lon Harris: 40:13 Paul Atkins.

Jason Calacanis: 40:14 Atkins, yeah, he’s all in on creating a test at the SEC where you could become accredited and participate in private companies. So participating in private companies, where the value is, could be a major unlock for people. Also, incredibly risky, which is why rich people when they go to dinner talk about whatever private deals they’re doing and what they have access to because they understand if it’s a one in ten chance of going a hundred X, you should probably take that bet every day.

Alex Wilhelm: 40:46 So incentives matter. I think that’s the key thing we’ve learned here. Mark, there are incentives all over the place, um, and the game of Tao, the brilliance of it, dare I say, is that it’s very focused on incentives. Validators, uh, miners competing, just it’s- um, an elegant and beautiful system. When you look at the first subnet we looked at, what comes to mind in terms of where that could wind up in two or three years?

Mark Jeffrey: 41:21 I mean, I think you’re right. The- the brilliance of it, the incentive alignment engine of Bittensor is its core invention, and everything within it is contests within contests within contests. Everybody’s competing with everybody and only the most excellent get anything. So, you know, in companies, even- even the most- the best-run startup, there’s a lot of inefficiencies. You have to hire people, right? Some of them hide in the company successfully without doing very much, um, and they’re sort of mediocre. There is no middle management when there’s contests within contests within contests. You know, this is the most ferocious form of capitalism ever invented as far as I can tell. So, yeah, I love it.

Alex Wilhelm: 41:56 Let’s meet our next subnet. And I think, you know, at the pace where we- going with two or three subnets, uh, you know, every couple of episodes, we’re gonna get through all 120.

Jason Calacanis: 42:05 We are. That’s—that’s become our goal.

Alex Wilhelm: 42:07 And actually, Will and Stefan from Macrocosmos, they run three different subnets, Jason. So we’re knocking out three. We’ve got Apex subnet one, Iota subnet nine, and then Daddy Universe slash Gravity, that’s subnet 13. So, thanks for Will and Stefan.

Lon Harris: 42:20 So greedy, so greedy. Taking up all these subnets. He owns 2.7% of the subnets.

Jason Calacanis: 42:24 Okay, let’s hear about it.

Mark Jeffrey: 42:26 We actually used to run five subnets. Um, but we decided that we would—we would like to live long enough to actually see these things reach the market, so we had to—we had to downsize a little bit. Running a subnet is no joke, as—as Mark can attest and certainly can. But, um, I think Will and I would—would love to speak a little bit about our most ambitious work on Bittensor, which is Iota. So Iota is subnet nine and, um, perhaps I’m being hyperbolic but I firmly believe it’s the most ambitious thing that is being built within Bittensor right now. And, I think from a very high level, what we want to be able to do is orchestrate the world’s compute in the way that Bitcoin did, but to train models that rival ChatGPT, frontier-scale models. And there is—there is actually a growing consensus in the industry that training the next generation of very large models is—is becoming prohibitively expensive, as I’m sure you guys are aware. We’re talking about data centers the size of downtown Manhattan, and the next generation’s gonna dwarf even this. And I think what becomes inevitable at some point is we need to think about cost efficiency. We need to think about CapEx, and something… In fact, one of the core ideas that inspired Bittensor from Const himself was the idea of doing what’s called distributed training. So distributed training is taking compute, uh, from pockets all around the world and coordinating that compute and making that equivalent to a large data center. Yet it’s just one that is dispersed all around the world. Now, there’s enormous amount of work you have to do, as you can imagine—a lot of networking, a lot of algorithm, uh, and a lot of research that needs to be done. But, um, it’s a very high ceiling for the project. So we’ve been working on Iota for about nine months. Will, my co-founder and CEO, myself, the CTO. Uh, we’re really, really excited and motivated about it. And, um, yeah, the—the dream of training in Bittensor is very much alive and kicking.

Jason Calacanis: 44:08 And—and to put this in context, Lon and Mark, to create a frontier model now, and you do a run basically to create the next ChatGPT, the next Claude, the next Grok, whatever. The—the estimates are $200 million in compute costs. This isn’t the staff, this isn’t the company, you know, testing or anything. Just the raw compute cost is 200 million. Now, that’s a big number. But if you had, well, if you had, uh, let’s say 10 million people participating and contributing some compute, well, you know, then you—it would be a couple of dollars, it’d be $20 per person. So you start, or if it’s a million, it’d be $200 worth of compute each. If it was 100,000, it’d be 2,000. If it was 10,000, it’d be 20,000 in compute. So maybe you get down to 10,000 people per… viding 20,000 in compute and hey, it’s conceivable in the same way SETI@home was. Yeah, Mark?

Mark Jeffrey: 45:07 That’s absolutely correct. And the thing I love the most about what these guys have done is, you know, as someone who has watched mining occur within Bittensor but been unable to participate until I got my claw and I started vibe mining with that, but minus the claw and all that, really these guys were the first people to provide a ready-packaged mining application for their subnet. Nobody else had done that, right? So you at home could now be a Bittensor miner for the very first time and not have to know anything about anything, just download this thing, set it up, and now you’re mining Bittensor. It was fantastic. I loved it.

Jason Calacanis: 45:43 So, talk to me about how that IOTA install works. Is it as simple as like downloading Slack or downloading a Chrome browser and then it just runs in the background?

Alex Wilhelm: 45:54 I think a lot of the principle of this and you know, I’ll talk about the differences in our training methods later, but Train at Home, as we call it, the application, or IOTA, is like two clicks to install. And again, Mark is one of our miners, so I don’t actually have to pitch for myself. You install the application, it’s a simple binary. Stefan has apparently gone to the wrong place, but you basically just install the app, put in a key to be paid out with, and we use your computer when you’re not using it. And one of the reasons training models is so expensive is not just that you need immense amounts of compute. It’s also that you train with like the prime rib or like the ribeye steak, you know, you take the best compute for the longest period of time. You know, it’s the filet mignon cut. Um, and that makes it even more expensive because it’s the filet mignon cut, but the cow has to be rested for 50 days and you’ve got to leave it there and you’ve got to look at it. And what we want to do is fundamentally make AI models out of ground beef. So we take bits of compute from here or there, we perform some magic in the background, and we make meatloaf. And you know, everybody loves ribeye steak, but actually meatloaf’s what you want when you’re comfortable. And you know, there’s some sort of joking in the metaphor there, but materially, um, you have two major markets for AI compute and GPUs. One is inference. Everybody here knows that. Mark’s running Claude, you guys are using ChatGPT or your preferred model, and it’s actually inference that creates the money. It’s where revenue is associated with. You know, when you track sort of OpenAI or Anthropic budgets, it’s all from inference tokens. But the reality is you have these incredible training workloads, like huge data center-sized training workloads that the main labs are spending immense amounts of money to satisfy. And these are taking up city blocks of compute, and the sort of lost opportunity there is you can’t use that compute to make inference. So the way IOTA works is the sort of minimum compute unit we can use is, um, at the moment the size of a MacBook and for about 30 minutes, and we can actually extract value from that.

Mark Jeffrey: 48:00 Our ultimate goal of this is we actually want to be able to make, you know, we call it indistinguishable training compute from this mixture of compute nodes. And this is sort of one of the philosophies Steph and I have had when we first started Macrokosmos is a lot of people when they approach any project in Web3 or crypto is they say, ‘Oh, how can I do what people do in a centralized way but in a crypto way?’ So, um, you know, we were talking about Templar. Templar does training. Templar effectively allows you to train in a centralized way using crypto rails. I won’t go into the technical details, but the nodes they used were Blackwell. So B200 nodes, it’s a quarter of a million bucks for each compute unit. So, you know, it’s prime rib. Maybe it’s prime rib and, you know, you run it around your houses doing that, but it’s so incredibly expensive. What we think the interesting thing about, you know, this sort of merger between crypto and AI is, is how can you use like fundamentally fractal compute, distributed compute, 10 minutes here, 20 minutes there, you know, Mark’s sort of Mac Mini when it’s sat idle in the evenings and sum that together to something that creates economic value in a very, very constrained market. Um, so we actually released a research paper today that validates all of our work, but we, we’ve been working on this problem for a year and it’s been, um, it’s hard but compelling work.

Jason Calacanis: 49:19 Are people calling open clock Claws?

Mark Jeffrey: 49:21 Yeah, I am anyway.

Jason Calacanis: 49:22 You are, okay. Fine, just wanted to make sure there’s not a new piece of software that I am not aware of. So explain how I get the package on my Mac to join Iota.

Mark Jeffrey: 49:34 You can download it through terminal, through the command line interface if you’re more coding proficient, or you can go to the website iota.macrokosmos.ai, scroll down and you’ll see a download button. As you mentioned before, Jason, it’s just like downloading Slack or any other application, exactly. You download this, it will generate a wallet for you, and then what will happen is when you open the application, you click start, and your machine will connect to our globally distributed training cluster. You will be, you will earn passive income while you sleep between 10:00 PM and 6:00 AM. Your machine can be quietly being used as a training node. So the benefit to you is you get to accumulate Iota, you get to be part of the big mission, and the benefit for us is we basically have a, we have an always-on supercluster of compute, and that’s, that’s really the value.

Jason Calacanis: 50:21 Got it.

Lon Harris: 50:22 I have a question. If I start doing this overnight on my work laptop, is ActiveTrack gonna look like I was up all night working? Cause this could really work out.

Mark Jeffrey: 50:31 For sure. Yeah, it’ll look like you’re working. We will literally do your work for you.

Lon Harris: 50:35 Yeah, this… this is going to look amazing. I’m going to look like a hero.

Alex Wilhelm: 50:39 Well no, it would show massive CPU usage. So we have for background, there’s a piece of software called ActiveTrack, just to lock down software and there’s 20 of these. So we have confidential data on our network, but it shows like what the computers at our offices are doing. If somebody were to download everything, you would see that in ActiveTrack and you’d be like, ‘Whoa’…

Jason Calacanis: 51:00 exporting stuff. Look how much GPU Lon was using all night! Lon was using so much compute he must have been crunching numbers all night.

Alex Wilhelm: 51:07 What a genius.

Lon Harris: 51:08 All night.

Jason Calacanis: 51:09 All night. So, give him a raise. There’s… there’s multiple ways to participate in IOTA. One is as a user and you would download, you would have a wallet, and then you would earn Tau by contributing your compute. But there’s also becoming a node. So maybe you could explain what being a full node is versus that.

Mark Jeffrey: 51:31 I suppose that there’s a couple of expressions of IOTA. One is this very like participatory everybody shares, everybody co-trains a model version of which, you know, we sort of nicknamed ‘train at home’ because because that’s kind of what it is. The second version of this is like the benefit of this tech is it has applications at sort of any scale within the AI stack. Now, one of the ways we expect this to materialize in some of the earlier runs is it will use frontier compute. So H100s, B200s, those quarter of a million dollar machines, but what we will have availability for is anybody can register their nodes on a sort of rotating basis. So we’re working on this in the background where you can post compute for a fixed interval, 20 minutes, half an hour. There’s some fine tuning on the machine learning side for us to work out the sort of sweet spot there where we get value from it. But you know, we’ve spent a lot of time talking to GPU providers and others, and the sort of thing we’ve found is that particularly if you can make an instance interruptible is what they call it. So, you know, Jason, you’ve got ten computers, you sell them to Mark most of the time for five bucks an hour, but you know, Mark sort of on a Thursday is like, ‘eh, I don’t need the compute’ last minute, you know that he might sort of pick it up Friday and you don’t want to go sell it to somebody else. People in that instance are just looking to recover capital returns and operational costs. So you can get compute at 10 cents on the dollar. So we want to work with major providers to basically harness all that compute at 10 cents on the dollar. Because if we can get our training efficiency within, you know, a factor of 10 at that point, we’re making money.

Jason Calacanis: 53:01 Okay, this is fascinating. So somebody who has extra compute, like let’s say it was even CoreWeave, I interviewed the founder of that company, let’s say they had old H100s and they hadn’t sold them yet and they’re five years old and they’re still in a rack and they could say, ‘you know what, just we haven’t sold them yet, we don’t have a new customer for them, we might as well put them to use.’ It would almost like being if you had a house and you hadn’t sold it yet, you’d put it into the Airbnb inventory just to cover your cost basis or something at a discount just to make sure you pay your taxes, your real estate taxes and your, you know, operating costs for the home.

Mark Jeffrey: 53:45 It’s like that. But it’s… this is sort of one of the foundations of DePIN and decentralized AI is people are like, ‘oh, you know, I can use my compute for a day here.’ What’s unique about IOTA is it’s like an Airbnb, but you can subnet for 20 minutes with no notice and get paid for it. That’s a better analogy there. Because it’s not the whole home for a whole night.

Alex Wilhelm: 54:00 Yeah, I know. And that’s where we think the real value is, because if you’re renting out your house for a day or a week, then you want a reasonable bit of money for it, you want to account for all these wear and tear. If you’re trying to get 20 minutes or you’re just trying to make a quick buck of something that’s already in production, the economics change really dramatically. And we’ve validated this with seven or eight different data center clients who all say basically the key thing is they want to be able to take the compute off you, and there’s no training method in the world that allows you to have the level of interruptibility we’re designing for. Because typically training, you know, in a data center is something called synchronous training. So all of the nodes move in lockstep and it’s actually the slowest node that sort of drives the speed of the whole system. We’re trying to design for this world where we assume that Mark will slam his laptop shut, we assume that CoreWeave will take their rack back and sell it to a premium customer. We actually assume that in, sort of, even the case of large labs, you might just have an inference spike and you want to reallocate training compute to revenue workloads and then pull it back when you come down. Like, you know, my father was an energy engineer and you always design for the sort of peak load in the sinusoidal, and we’re trying to make training compute liquid so that we can sort of achieve ultimate economic value really.

Jason Calacanis: 55:19 Any questions there Mark, and your thoughts on the project, just generally speaking, and how disruptive it is?

Mark Jeffrey: 55:25 Yeah, I mean I think it could be one of the bigger projects in BitTensor. I think these guys have been around for a while and they know the system better than most of the other subnet owners. And so I, look, I’m very bullish on them and I love what both of them are doing.

Jason Calacanis: 55:42 And is this a corporation, well, and you have venture capital backing it or is it just a straight anachronistic crypto weirdo project? How is this running?

Lon Harris: 55:55 We are actually a company, but we are fully bootstrapped. We haven’t raised venture capital to date. We’ve been fortunate to build, actually Stefan built the first subnet on BitTensor for his sins. So we’ve been around for a while and, yeah, we have no outside franchise except for our relationship with BitTensor. So we’re very fortunate to that, very fortunate to the people who’ve sort of invested us over time, but we’ve created this ourselves.

Jason Calacanis: 56:22 And what do you think of the, Stefan, what do you think of the kerfuffle with one of the subnets taking their tokens and marbles and leaving all of a sudden, and the impact that has on the ecosystem as one of the early adopters of subnets?

Lon Harris: 56:37 I think it’s very disappointing. I come from an AI background, not a crypto background. So being honorable and being trustworthy is a really important thing to me and something that I am extremely sensitive to when I operate in a crypto world. But it’s disappointing. It’s disappointing that someone did that to a lot of people. I won’t dive too deep into it, but I think it’s a shared emotion that we have right now.

Alex Wilhelm: 57:00 I think the response of Jake is on point, which is to say this is a great opportunity for us to reflect on a part of BitTensor that was too immature and needed to be iterated on and refined. And I think that’s the right pragmatic lens to see the experience. So yes, this will be, I’m sure, just a bump in the road in the long term.

Jason Calacanis: 57:22 So you’re frustrated, disappointed. And what do you think the solution is to avoiding it in the first place?

Mark Jeffrey: 57:27 I think what Mark described earlier, which is this idea of owner lock-up. So you should have more of a fiduciary duty to the people that believe in you, to put it simply. And I think there’s a way that you can translate that into things like smart contracts or similar, which shows the commitment of founders and certain operators to their community. And that can actually be used for price discovery for the project itself, right? It’s a measure of intention, it’s a measure of conviction. I think that’s quite an elegant solution and I’m sure we’ll see a few different prototypes of that. We might not get it right the first time, but it feels like roughly the right direction of travel to not have this experience again.

Jason Calacanis: 58:07 Athena, I’m the first investor. They are the greatest assistants in the world for a great price. $3,000 a month, you get an awesome full-time assistant to work with you on your scheduling, on research projects. And you say, well, oh, we’ve got to talk about all this AI. These assistants are AI-powered, and so they will do all kinds of jobs for you and help you have a human in the loop as you’re working. I was the first investor in the company. They’re now a partner of All-In, they’re a partner of ours here, and it just makes me super, super productive. So you can basically turn your EA into your Chief Health Officer.

Lon Harris: 58:51 That’s this month’s productivity hack. Thanks to Athena, for we do a productivity hack once a month. This month’s productivity hack: you should turn your EA into your Chief Health Officer. You know, for founders, their health can often take a back seat, Jason. So one of the ideas we came up with were all the different ways that you can sort of delegate some of these tasks to your EA: keeping up with your regular checkups, dental cleanings, handling your referrals and follow-ups, taking care of your insurance paperwork and your reimbursements, helping you plan healthy meals and what groceries to buy, worrying about your nutrition. All of these things can be things that you pass off to your EA to help save time. If you want to learn more, athena.com/jcal, J-C-A-L.

Jason Calacanis: 59:29 There you go. That’s where to go. And as an example, you know, I lost weight, shout out to my guys at ro.co with GLPs, but now I want to add more muscle. And I’ve been adding some, but I need to get a trainer. So I said, hey to my Athena assistant, within a 20-minute drive of my house, of the ranch, I want a trainer who has their own facility to do one-on-one training with me, you know, one to three times a week. Let them know I travel a ton, my schedule is very variable, and here’s how I’d like to proceed. Because a lot of times they’re like you have to pick the same slot every week or whatever, doesn’t work for me. So I said, hey, I want to buy 20 sessions up front and then I want to be able to just drop in, you know, and it could change dynamically. I might need to cancel within, you know, the same day, etc. So they explained the rules of the road and some people are like, yeah, we don’t work that way, you have to do Tuesday at 4:00 PM every week for the year, whatever. And it’s like, well, that’s not going to work. So they pre-vetted and did all those discussions because I don’t want to have that awkward discussion with the person and then I’m negotiating with a trainer. Then I said, and then find me three people to come to the house and I would do the same thing. I’ll pay them for 20 sessions up front so they get the value of that money up front, but they have to be willing to let me cancel same day, you know, within two or three hours if like I have to get on a plane. And it it worked. They went through 20 different people and found me three of each. And I just give them a call like that. Just keep calling people until you get to three.

Alex Wilhelm: 60:29 Yeah, and they can take a day to do that. You can’t do that. You have too much other stuff.

Jason Calacanis: 60:33 I don’t have the time for that. The other thing that I had them do was I had them go onto different people’s LinkedIn, Mark, and I said I’m not happy with the flow of people coming in for these three positions. So here’s what I want you to do. I want you to go on Lon’s LinkedIn, my LinkedIn, or Heidi’s LinkedIn and just send InMail through LinkedIn to anybody with this title in these three cities: Austin, Houston, San Antonio. Hey, we’re looking for this position in Austin in an office. If you know anybody, let us know, or if you might be interested, please let us know. We get, you know, a 2% response rate to that, but we filled two positions doing that. But that means you have to get to 500 outbound.

Alex Wilhelm: 61:49 It’s a lot of InMail.

Jason Calacanis: 61:50 So I just said, well, I can’t do 500 outbound. Lon can’t do 500, it’s a full-time job. I said, okay, for four hours a day, this is all I want you to do. 10 people an hour, four hours a day, 40 people a day and for 10 days, go. It worked, but I would never do that. That’s the job of a full-time recruiter. So shout out to our friends Athena.com/jcal. You get a couple weeks off.

Alex Wilhelm: 62:14 There you go. Check it out.

Lon Harris: 62:15 All right. Uh, all right what do we got next here on the docket? We’re ready to wrap. We have a little bit of off duty. I have some viral stuff we could react to or we could just do some recommendations. It’s up to you.

Jason Calacanis: 62:24 Okay, I think I’d love to go off-duty. So let’s go off-duty with Lon, J-Cal and Mark.

Mark Jeffrey: 62:27 And Mark.

Lon Harris: 62:28 Mark’s going to join us. You gotta give us the idea. I don’t know if they told you what off-duty is, but it’s basically when not doing crypto.

Jason Calacanis: 62:33 Yeah.

Lon Harris: 62:34 Whatever we want. We could talk about whatever we want. Products you bought, a gadget you bought, a TV show you were watching, music, an album. Well I did, I want to get your thoughts on one viral story before we get into recommendations. So Coachella, the Coachella Valley Music and Arts Festival, first weekend was this past weekend. There’s been… there were two headliners, very different performances. People are really debating and I think it speaks a lot to… Bieber was one.

Jason Calacanis: 63:04 Bieber was one.

Lon Harris: 63:05 Bieber was one, Sabrina Carpenter was the other, and I think it really speaks to what, what are we looking for in a performance? What is the value of authenticity versus showmanship? I think it gets to some really interesting questions. So first up, Sabrina Carpenter, she headlined, it was massive, cars on stage, huge sets, dancers. We could take a look at a quick clip here that I pulled up.

Alex Wilhelm: 63:29 Yeah, she’s extremely, yeah, Sabrina Carpenter, extremely entertaining. I saw her at Austin City Limits, ACL.

Lon Harris: 63:34 Yeah, so, so here’s Sabrina’s performance. You can see they built this massive set, she’s got tons of dancers, it’s a huge thing that’s going on.

Alex Wilhelm: 63:39 Wow.

Lon Harris: 63:40 This was a… we’re not playing the song, this is Espresso, her huge hit there, of course.

Alex Wilhelm: 63:44 It’s a production.

Lon Harris: 63:46 Yeah, thinking about me Espresso, her huge hit there, of course.

Alex Wilhelm: 63:51 Thinkin’ ‘bout me, Espresso.

Lon Harris: 63:53 Yeah, thinkin’ ‘bout that’s that me, Espresso, exactly. Oh, now how do I do this? Okay, there we go. And then I want to show this this one other part too. She actually drove a car off the set, like she drove through the crowd in Coachella, through the crowd, like this was how she closed it out. If you can see this bigger shot.

Alex Wilhelm: 64:13 Wow, she’s literally driving a car. That car’s not her driving, it must be on a track.

Lon Harris: 64:17 No, look, it’s… she’s just driving. They cleared out this aisle here, and you can see she’s just driving out of Coachella in this car. So that’s cool. It’s very big. It’s spectacle. It was like a party.

Jason Calacanis: 64:31 A lot of showmanship.

Lon Harris: 64:32 It was like showmanship. And then there was Justin Bieber. He was the headliner on Sunday night, and he went on stage alone, just him and a laptop and a screen behind him that was screen-sharing and he played YouTube clips, a lot of them of himself, and then he sang over his old music videos, sort of treating them like a backup singer. I’ll show you a little clip.

Mark Jeffrey: 64:53 Yeah, this was weird.

Lon Harris: 64:55 So he… you can see here, he’s on stage, he’s pulling up YouTube. You can see his screen, he’s typing in Baby, which of course that was his old-school song with him as a kid and then he stands up and he’s watching the video. He’s not even facing the audience, he’s watching the video and he’s like singing along with himself from when he was a kid. And then here’s… he wasn’t even just singing the whole time, he also just watches YouTube for some of it and he’s… here he is watching the ‘Got ‘em!’ meme and then you’ll see he flips over to Double Rainbow and he’s just… he’s hanging out playing on YouTube. So a lot of people were praising this, it’s so authentic, it’s so real, and he started his career as a YouTuber, Justin Bieber, he was arguably the first YouTuber to get super, super famous.

Alex Wilhelm: 65:37 Arguably the first YouTuber to get super, super famous.

Lon Harris: 65:40 Yeah, here he’s watching Double Rainbow.

Jason Calacanis: 65:44 All right, I can give my judge… I can give my judgment. I know you want my judgment here.

Lon Harris: 65:48 So I’m… and there’s a lot of back and forth of course between the Sabrina and the Bieber stands, but I think everyone, I think it’s an interesting debate. Which performance would you be more interested in seeing? And if you pay all this money to go to Coachella, it’s like a thousand dollars plus to get to Coachella, do the… Do the artists owe you a Sabrina Carpenter level spectacle?

Alex Wilhelm: 66:03 Okay, so it really is in the execution of these things. It on its face when you describe what Justin Bieber did, you would say that was phoning it in.

Jason Calacanis: 66:16 A little.

Alex Wilhelm: 66:16 Just played his video and he talked over it or whatever. Okay, sure enough, it’s that would be a surface-level interpretation of it. However, he kind of did some storytelling in there. It was interactive with the audience and at the end of the day, was the audience engaged in this? They were. And I would say they were equally engaged in Bieber’s surfing YouTube as a performance art. So it’s very meta-commentary and it was done so well that I think he hit the right note, which is this is a meta-commentary on my career and I’m going to relive how you experienced me.

Mark Jeffrey: 66:51 Exactly.

Alex Wilhelm: 66:51 And there was some brilliance to that. Now, if you were, you know, any other artist who was not a native YouTube artist, they might not have pulled off that meta-commentary right. There is another analogy to this.

Mark Jeffrey: 67:05 I had dinner the other night and I sat next to Jimmy Iovine.

Lon Harris: 67:10 Mmm. Wow, you had a big weekend.

Mark Jeffrey: 67:12 So anyway, I started talking to Jimmy, he’s also a kid from Brooklyn, he’s from Red Hook, I’m from Bay Ridge, it’s, you know, just right on the Gowanus Canal there. We had a talk and I told him, hey, you know, I’m a big Dire Straits fan, I know you did the album Making Movies, and we talked for an hour about that.

Lon Harris: 67:23 Oh, okay.

Mark Jeffrey: 67:24 And he also had Bruce Springsteen was, you know, the big artist that he had produced back in the day. He did Born in the U.S.A., one of the great seminal works in American sound of all time, 1984.

Lon Harris: 67:34 Yeah, sure. Born to Run, Darkness on the Edge of Town, he did. Yeah.

Mark Jeffrey: 67:40 Yeah, so this is like a, you know, a legend. I was real privileged to talk to him for all this time. And Bruce Springsteen did Springsteen on Broadway. And when he did Springsteen on Broadway, he did a meta-commentary of him. He did sing some of his songs like Bieber did, but he also told the story behind the songs. It was so engaging, it was so rich that as a fan, you can get away with that. Now, I’ve also seen Bruce Springsteen no less than 20 times probably over 30 years in concert, you know, like I’ve seen him two or three nights in a row, my mom’s a big Bruce fan, and I actually took my mom to Bruce on Broadway. You have to execute at a very high level, you have to be a true star. There has to be a lot behind it in order to pull that off. So I give Justin Bieber credit for actually pulling it off, which is not easy. Sabrina Carpenter couldn’t pull that off right now.

Jason Calacanis: 68:29 Right.

Mark Jeffrey: 68:30 Sabrina Carpenter, if she has 10 more Espressos, if she lasts, I don’t know how long Bieber’s been in pop culture now. I think it’s 15 years.

Lon Harris: 68:44 Yeah, 20 years or so, close.

Mark Jeffrey: 68:48 20 years. When Sabrina Carpenter’s 20 years into her career, she is…

Jason Calacanis: 69:00 She will be able to pull it off. But not in year three of her career.

Alex Wilhelm: 69:03 Right.

Jason Calacanis: 69:04 So, I think both of these were fantastic. Sabrina Carpenter has to build the foundation brick by brick in order to do the meta commentary which only comes after you have 50 hits, 25 hits, you know, in a certain size audience. So, I give them both credit. Mark, any thoughts here? I know you’re a big rock and roll fan.

Mark Jeffrey: 69:19 Yeah, you guys should know better than to ask me about anything other than classic rock. But okay, I’ll dive in a little bit here. So, I mean, I’m sort of split because I do feel like Bieber… I’ve seen like little clips of it. It looked fun, right? Like it didn’t look bad, right? It didn’t look like it was crashing out on stage. It looked fun, right? So, yeah, I think probably the audience enjoyed it. That said, I love a spectacle. I love a big rock show. You know, JKL and I went to go see The Wall when Roger Waters was touring that around.

Jason Calacanis: 69:53 Oh yeah.

Alex Wilhelm: 69:54 Incredible.

Lon Harris: 69:55 Life-changing.

Mark Jeffrey: 69:56 Yeah, totally agree. Yeah, that was like a very big production of tunes that you’ve listened to your entire life and just love, right? So that’s what I want when I go to a show. That’s probably the apotheosis of that. So, you know, I guess I’m sort of split.

Alex Wilhelm: 70:06 I think I probably would have preferred the Bieber—I’m not a huge fan of either, you know, I’m not like a die-hard Sabrina or Bieber fan, but I think just I would have preferred the Bieber set. It’s something real, you’re seeing him really think about and process this whole career, this trip down memory lane, this sort of nostalgic look back with him. I think that would be a lot more engaging as a viewer than, you know, big dance routines and the big pop. I kind of feel in general, and this is old man of me, but I’m going to say it anyway. When I used to go to Coachella in like the early and mid-aughts, it was just bands playing. There wasn’t this like… it wasn’t like you’re going to see a pop star like fly out of something and parachute into the arena. It was just like bands playing songs you liked, or rappers performing songs you liked, or whatever, DJs. And I feel like I like that more authentic sort of feel than all the spectacle.

Lon Harris: 70:46 Yeah. I mean, listen, the kids love the vibes. There’s meta commentary going on and it’s vibe culture in that, you know, if the vibes are either immaculate or the vibes are inauthentic and you’re taking a little bit of risk there. I have been one time to Coachella and it was quite enjoyable to go from set to set to set. I think I saw Lorde…

Jason Calacanis: 71:18 Oh nice.

Lon Harris: 71:19 And MGMT and like some other incredible artists like back to back to back to back. And it was like, wow, this is pretty impressive.

Alex Wilhelm: 71:29 I went in ‘04. It’s like a classic one. Radiohead, Pixies, Flaming Lips, MF DOOM, Air… really good year that. Beck was there that year.

Lon Harris: 71:40 Anyway, thanks unc.

Alex Wilhelm: 71:41 Yeah, it was like a three-unc special.

Mark Jeffrey: 71:43 Oh man, yeah, we’re old to be talking about this.

Alex Wilhelm: 71:45 But I did, I thought it was an interesting conversation that’s going on between this sort of spectacle and authenticity.

Jason Calacanis: 71:49 I’ll tell you there’s an analogy here. I went one night to see Chappelle. He happened to be in town. A friend of mine’s friends with him. I texted my friend and I was like, ‘Hey, your friend Chappelle…’

Alex Wilhelm: 72:00 …was playing. He’s like, oh, let me text him and maybe we’ll go see the show. Last minute we go see the show. After the show, Chappelle says, hey, you guys want to roll with me? We’re going to go do this after show. And he goes to this, you know, small comedy place in San Francisco that hosts maybe 100 people, 150 people. We go there, and we got a table set up for us. And then he does just him interacting with the audience.

Jason Calacanis: 72:30 Sure.

Alex Wilhelm: 72:31 He’s had a couple of drinks or whatever and he’s smoking a cigarette.

Jason Calacanis: 72:34 Crowd work. They call it crowd work.

Alex Wilhelm: 72:37 What are they- he’s doing crowd work.

Jason Calacanis: 72:38 Crowd work is what they call it, yeah.

Alex Wilhelm: 72:40 But he’s also telling stories and it’s not a set. It’s not a funny set like he had just done at the Chase Center, it is just him bantering.

Jason Calacanis: 72:49 Yeah.

Alex Wilhelm: 72:50 And, you know, some jokes in it. And he was not working on new material exactly, but it was just a different type of performance. To see both back-to-back and then compare them, again, back to this meta-commentary, there was him talking about the show he just did, then him talking about his life, and it was really like, to see those two bookends together it was actually really nice. Just like going and seeing a Bruce Springsteen show and then seeing Springsteen on Broadway talk about Bruce Springsteen and his life.

Jason Calacanis: 73:14 Right. It’s like some of those intimate, intimate evenings, yeah. And you got anything else here on-

Lon Harris: 73:21 I did. There was one thing I want to recommend. There’s a movie I saw it at the Austin Film Society here in Austin, one of my favorite local theaters. Shout out to them. But it is now on Hulu. It’s a Spanish film called Sorat. It’s about a father and a son. The father’s older daughter has gone missing. She was a big raver who attended a lot of these raves in the desert in North Africa. So the father and son are following these moving raves around the desert looking for this missing young woman, and then it leads into this whole separate adventure story. I don’t want to give anything else away about this movie because it takes some real turns, but wow.

Jason Calacanis: 73:53 You don’t give it up. I’ve heard a lot about this film.

Lon Harris: 73:55 It is incredible. It is a real experience. You don’t want to learn too much about it, you just want to watch it and take the ride with it. Yeah, I loved it. Real intense.

Jason Calacanis: 74:02 All right. I’ll give one recommendation here. People have heard me reference my favorite pen.

Lon Harris: 74:08 Yes.

Jason Calacanis: 74:09 I’m not like a pen snob. I don’t like the idea of buying a $100 or $500 pen. I don’t collect pens, but I like to write and I like a great pen. But I know I’m going to lose a pen, so somewhere between, you know, a commodity pen for, you know, a dollar or two dollars and these $100 or pens that cost hundreds of dollars, I found one of the nicest writing pens ever, the G-750 retractable pen. And I think it goes for 10, 15 bucks.

Lon Harris: 74:37 Yeah, there it is. 14 bucks. So it’s like 13.94 here, yeah.

Jason Calacanis: 74:40 Yes. You can get it on Amazon and it’s just an incredible pen. When you write with this pen, you will learn why this is such an amazing pen. Now, the reason I know about this pen is because Unk is basically Unk, that’s me.

Lon Harris: 75:22 Yeah.

Jason Calacanis: 75:23 Was on TikTok and I stumbled upon the Staples Baddie. Kaden Roland works at Staples.

Lon Harris: 75:27 Sure.

Jason Calacanis: 75:28 And if you type in her name and you type in Zebra S750, G750, she’s the one who recommended it to me. And she is this incredible young person who works at Staples. She might be trans or, you know, I’m not sure exactly…

Lon Harris: 75:59 Are you sure?

Jason Calacanis: 76:00 Oh yeah, maybe. But she has an incredible performance where she goes and talks with extreme passion about Staples and photocopies and pens and journals. And she literally clanks her nails and snaps up her favorite office products. And she’s probably done more for the brand of Staples with young people and all the way up to Gen X myself, than Staples could ever do with a hundred million dollar ad campaign. I don’t know what they’re paying her.

Lon Harris: 76:34 Are they paying her?

Alex Wilhelm: 76:35 I don’t… is Staples embracing her?

Jason Calacanis: 76:37 She’s a store employee, Lon.

Lon Harris: 76:39 Oh, but yeah, but they’re not paying her to make these videos, she’s… they’re just paying her to work the register.

Jason Calacanis: 76:44 I think now they are paying her. I think they must be paying her now. If not, there’s a hundred people who would hire her to be the Walmart baddie or the Target baddie. I mean, she should be getting paid a million dollars a year. Minimum. Minimum of a million dollars a year. If one of my startups came to me and said “We want to hire this person to run our social media for a hundred thousand dollars a year plus nine hundred thousand dollars a year in stock options”, I would say “Do it, do it immediately”. Because the ability to authentically engage and then get me to buy the pen and know who the Staples Baddie is, is just infinitely valuable.

Mark Jeffrey: 77:22 Yeah, really.

Lon Harris: 77:32 Yeah, I think I found her TikTok. She has 582.3 thousand followers on TikTok. That’s crazy.

Jason Calacanis: 77:37 I mean, somebody’s gotta post her.

Alex Wilhelm: 77:41 Just for talking about Staples. There’s a… there’s a Papa John’s guy too, right? I’ve seen him a bunch of times on TikTok, where he’s just a guy who works at Papa John’s, but he talks about how they make great pizza, which I don’t…

Jason Calacanis: 77:51 I think now this is gonna become a thing. All right. Another amazing episode of TWIST is in the can for Monday, April 13th. Thanks, Mark Jeffrey.