This episode of Bell Curve features hosts Mike, Miles, and Zave discussing the intersection of AI and crypto at what they describe as a critical inflection point. The conversation is prompted by Anthropic's controversial launch of the Fable model, its geographic restrictions, rapid jailbreak, and undisclosed response nerfing — events the hosts argue expose the structural risks of centralized AI.

The discussion covers decentralized training projects including Prime Intellect, Pluralis, and Noose Research (Hermes), the privacy-first consumer AI app Venice by Eric Voorhees, and the broader platform risk argument for open source and decentralized AI models. The hosts also explore analogies between the current AI investment frenzy and crypto's own alt-lab boom, the fat protocol thesis, and Jevons paradox as applied to frontier model pricing, drawing on a blog post by Alok from Standard and a piece by Sam Lessin of Slow Ventures on SpaceX's valuation and minority belief systems creating trillion-dollar markets.

Anthropic's Fable Launch Exposes Centralized AI Risks

Anthropic launched its highly anticipated Mythos model as Fable with significant guardrails, restricting biology, cybersecurity, and model training use cases — and blocking access entirely in countries like Canada.

A team of Amazon researchers jailbroke Fable quickly after launch; the US government reportedly hoped Anthropic would patch the vulnerabilities, but they did not.

Anthropic was caught nerfing responses for certain use cases — particularly anything using Fable to train other models — without disclosing the downgrade to users.

"Maybe this is Anthropic, one, trying to get regulated and put in the seat of incumbency. And then two, maybe they're thinking about ways to monetize outside of just the $200 a month" - Mike

Anthropic's data collection policy change created HIPAA compliance violations for healthcare providers using Claude, turning a philosophical privacy concern into an immediate legal business risk.

Mike notes Dario Andreessen has only one direct report (his chief of staff) and is not heavily involved in day-to-day operations, drawing parallels to governance failures seen in early crypto companies.

The Real Case for Decentralized AI: Cost, Not Ideology

Miles argues the primary driver for decentralized training is economic, not philosophical: frontier labs spend roughly $100B to train a model and expect $150-200B in return, but smart routing to cheaper models is eroding that revenue stream.

"The real demand is for cheaper but effective models. And how do you basically break that bottleneck of needing $100 billion to train them? Well, you spread that out over many, many people" - Miles

Jensen Huang stated approximately 1.5 years ago that the industry needs to crack decentralized training, lending early institutional credibility to the thesis.

Miles is bearish on decentralized inference — "I think that will always perform worse than something that is centralized" — but bullish on decentralized training as a supply-side cost hack.

Miles compares decentralized training favorably to GeoNet (GPS sensors distributed globally), arguing both represent products that couldn't exist without distributing massive CapEx across many contributors.

Zave distinguishes between open source AI (e.g., Alibaba's Qwen), decentralized AI (Prime Intellect, Pluralis, Noose Research), and closed source — noting open source models can still censor users or revoke licenses.

Noose Research's Hermes: A Crypto Team Beating Closed-Source AI

Zave argues Noose Research's Hermes harness is now better than OpenClaw — the harness ChatGPT acquired — making it arguably the best product in its category despite being built by a crypto-native, open source team.

The harness-versus-model distinction is key: "The best way to think about this is like it's the car and then the model is the engine" - Zave, with Prime Intellect potentially becoming the engine and Hermes owning the customer relationship.

Noose's strategy is to use the harness as a user-facing wedge, smart routing to various models today, while simultaneously training their own distributed model — with the endgame being vertical integration by routing to their own model.

Index Network, a portfolio company, operates as an agent skill within Hermes, proactively connecting investors with relevant startups at events like Edge City — demonstrating real PMF for crypto AI applications within the harness layer.

Miles notes Noose is valued at approximately $1 billion, and if they launch a token post-PMF, it will not be a speculative small-cap but a fundamentally-driven asset.

Venice AI and the Consumer Privacy Wedge

Venice, founded by Eric Voorhees, runs AI models client-side with no server-side data collection, making it the hosts' pick for the best-performing consumer crypto AI application currently live.

"What Venice is doing, which is cool, and it was also early in terms of the insight, is allowing people to use AI in a way that is fully private" - Zave

Zave argues Venice demonstrates what users actually want: AI that doesn't accumulate intimate personal data, contrasting with Claude and ChatGPT which know "literally everything" about a user.

Mike pushes back that censorship resistance and privacy tend to resonate more with businesses (platform risk, data leakage) than individuals, and that the business use case is what typically gets these products across the finish line.

Zave also highlights Near's Ironclaw as a privacy-focused harness for personal use, and notes that cryptography applied to organizational AI workflows touching sensitive data remains a significant underexplored opportunity.

Token Maxing Is Over: The Shift to Smart Routing and Open Source

The era of using the most expensive frontier model for every task is ending: "The era of just token maxing the state-of-the-art models is kind of over" - Miles, as harnesses now smart route to the minimum viable cost model per task.

Hardcore agent engineers are already using open source models for most execution tasks, reserving closed-source frontier models only for evaluation and spec planning at the start and end of workflows.

Mike observes that most SaaS products (Notion, HubSpot) have naively jammed AI chatbots into their interfaces without solving a real form factor problem: "We haven't moved past the naive chatbot UX, and I don't think that's a compute constraint. It is a creativity, a product constraint."

The game theory for founders has been to publicly embrace token maxing while quietly figuring out what actually works — and the hosts argue we are now entering the phase where that quiet rationalization becomes public.

Crypto as a Leading Indicator: Fat Protocols, Alt Labs, and Jevons Paradox

Alok from Standard published a piece arguing the current AI lab investment frenzy mirrors crypto's alt-chain boom — where Bitcoin and Ethereum were immaculate conceptions that triggered a massive n-horse race, with only Solana ultimately claiming a third seat.

Miles draws a parallel between the fat protocol thesis in crypto and the current moment in AI: frontier labs are in the 'fat protocol' era, but block space (compute) is getting commoditized, and the value will shift to the application layer — except OpenAI and Anthropic, unlike Ethereum, are also very good at building apps.

Sam Lessin of Slow Ventures argued in a blog post on the SpaceX IPO that "DCF is no longer the only globally scalable story of value" and that "the internet lets minority belief systems create trillion-dollar markets" — a thesis Mike notes crypto investors have lived for 8 years.

Mike warns that Jevons paradox — the idea that cheaper compute will generate proportionally more demand — has a dangerous timing gap: Ethereum successfully collapsed block space fees by two orders of magnitude (from ~$200 to cents), but demand didn't materialize fast enough, and now Ethereum's ability to generate sustainable fees is out-of-consensus.

"Crypto does lead the way. Crypto tends to forecast social and financial trends. You keep thinking crypto is this crazy Wild West thing, but it's directionally right, even on the stuff that I don't want it to be directionally right on" - Mike

What Comes Next: Tokens, Wearables, and the Crypto AI Stack

Zave predicts all crypto AI teams with any relationship to non-enterprise customers will launch tokens: "Pretty clear. Unless you really pivot hard into the pure AI space and every customer you serve is a Fortune 500 company."

Miles flags that token launch standards for crypto AI companies remain undefined — revenue thresholds, profitability benchmarks, and whether recurring revenue models differ from transaction-based models like Hyperliquid are all open questions.

Wearables are flagged as the next major catalyst, with Miles noting they will likely run local open source models for day-to-day tasks rather than expensive frontier models, further driving demand for decentralized AI infrastructure.

Zave references a recently published paper claiming crypto assets can now be fundamentally valued for the first time, suggesting the next cycle of crypto AI tokens may see fundamentals-driven valuations rather than pure speculation.

Bittensor's Templar subnet is cited alongside Prime Intellect and Pluralis as teams that have been working on decentralized training for years and are now approaching production viability.

Resources Mentioned

Cryptoassets The Guide to Bitcoin, Blockchain, and Cryptocurrency for Investment Professionals

, further driving demand for decentralized AI infrastructure. Zave references a recently published paper claiming crypto assets can now be fundamentally valued for the first time, suggesting the next

66-Book Bible Study Guide, Bible Study Supplies & Journal, Christian Scripture Summaries, Daily Devotionals Gift Idea Religious Faith-Based Notebook for Women Men Teens Beginners Prayer Reflections

, actually, if you look at Noose Research, Pluralis Research, etc., they all actually have the word research like in the title. And I think what you're seeing now is research over many years actually

over many years actually now coming to production

, they all actually have the word research like in the title. And I think what you're seeing now is research over many years actually now coming to production. Somewhat in tandem with other things ha

is pretty much better than OpenClaw and maybe even the best in this kind of like harness space

bably even surprised other people in the space, like I would say with some confidence now that Nuis Research is pretty much better than OpenClaw and maybe even the best in this kind of like harness sp

Team Topologies, 2nd Edition Organizing Business and Technology for Fast Flow of Value

d of like harness space, which is super cool because it's built by an open source and decentralized research team, uh, which is doing it with crypto ideals and values. And this is a product that's bet

was from 2 years ago

$100 million, you just put the word applied in front of like whatever your thing is, right? I mean, research was from 2 years ago. Now we're applied. I think another topic worth talking about, which

who just left Stellar Crypto and was talking

0% happening. I mean, you know, okay. I had this long, you know, shout out to our, our old head of research who just left Stellar Crypto and was talking. There was a public, we were going back and fo

Out of the Easy

s and networks, which is going to be very cool to see. I don't know if you guys saw, there was that paper that came out, you know, a day or two ago saying that you can now fundamentally value crypto a

The One Hour Content Plan The Solopreneur’s Guide to a Year’s Worth of Blog Post Ideas in 60 Minutes and Creating Content That Hooks and Sells

, I am curious to get your guys' take on this. So did you see Sam Lessin of Slow Ventures put out a paper or like a blog post on, you know, we are now in a, he was talking about SpaceX. Hold on, let m