Gutenberg Lab — Morning Paper

Zcash executes three-layer overhaul to deliver quantum-resistant privacy at 50,000 TPS

Speakers

David Hoffman (Bankless host) × Sean Bowe (Zcash cofounder, Tachyon project lead) × Dev Ojha (ZK researcher, Osmosis cofounder)

Bottom line

Now that AI has turned blockchain surveillance into a mass capability anyone with an LLM can wield, Zcash is building the only path to true financial privacy at scale—combining recursive SNARKs, 100 MB/s proof-of-work throughput, and private information retrieval for wallet discovery.

3-line summary

• Zcash is executing parallel upgrades across three layers: cryptography (Tachyon), consensus (100 MB/s PoW scaling), and wallet UX (private information retrieval for payment discovery)
• Quantum recoverability went live in August 2026—the only blockchain where users can recover funds even if quantum computers arrive tomorrow
• Formal verification caught all ~1,000 injected bug classes in testing, establishing the first "audit the auditors" regime in the industry

Three key points

1. AI democratized surveillance

Sean Bowe: "Now with AI, a teenager with access to an LLM can identify and track everybody" on transparent ledgers. What only chain analytics firms could do is now trivial for anyone. The prediction: free public databases will emerge with complete tracking of all transparent ledger users. Privacy shifted from theoretical concern to urgent practical threat.

2. Recursive SNARKs achieve "stateless blockchain"

Dev Ojha: Full node work changes from verifying 1 million zero-knowledge proofs to verifying one proof. The perpetually growing nullifier set (double-spend prevention data) gets compressed, removing cryptography as the bottleneck. Remaining challenges are "just" engineering.

3. Private wallet discovery is harder than private transactions

In a fully private blockchain, "how do you figure out you got paid?" is the hardest problem. The old solution (download entire chain and test every transaction) doesn't scale to 100 MB/s. The new solution—Private Information Retrieval (PIR)—uses homomorphic encryption and AI-optimized GPUs (H100) to query terabyte-scale data without revealing anything to the server. Already deployed in the 2026 coin holder vote.

Editorial perspective

The pattern of the industry copying Zcash's technical choices (Ethereum adopting the BLS12-381 curve) reveals its role as the R&D lab for privacy tech. The foresight to design quantum recoverability in early 2025—before the industry panic—and ship it live, combined with formal verification and AI model self-attacks, raises the bar for the field. Yet the philosophical rigor of avoiding the "99% private" trap (where two transactions can eliminate all other users from the anonymity set via set intersection attacks) creates constant tension with scale. Dev Ojha's warning that governance capture via "slow committees" is the primary threat reflects awareness that technical excellence can be nullified by bureaucratic stasis—a lesson from the history of standards bodies.

Source: Bankless, "The Complete Zcash Roadmap | Sean Bowe and Dev Ojha" (November 2, 2026)

http://podcast.banklesshq.com/

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.

AI Solves Math's Hardest Problems in Weeks, Forcing Mathematicians to Redefine Their Purpose

Featuring

Lex Fridman × Alex Kontorovich (number theorist, Rutgers) × Ankur Moitra (complexity theorist, MIT) × Steven Brunton (fluid dynamics, University of Washington)

Bottom line

In an era where AI solves math's hardest problems, a mathematician's value lies not in proving theorems but in "thinking deeply"—building theories, connecting concepts, and discovering what questions to ask.

3-Line Summary

OpenAI and Anthropic solved 700 problems in recent weeks, including Navier-Stokes blow-up, quasi-Riemann hypothesis, unique games conjecture, and Fermat's Last Theorem formalization (13 million lines in Lean, two days). Mathematicians experience "awe and terror" as theorem-proving—the "most visible part of the job"—becomes automated, forcing redefinition of math's essence: theory-building, conceptual connection, and question discovery. Terence Tao laments "years of research casually destroyed overnight"; at the 2024 Fields Medal ceremony, participants debated "is this the last time a Fields Medal is awarded?"

3 Key Points

1. Auto-formalization is solved: Anthropic formalized Fermat's Last Theorem in 13 million lines of Lean code in two days with 2-3 humans (no domain expertise required—just "keep going" prompts). Kevin Buzzard had a million-dollar grant to spend five years formalizing just the final step; AI blasted through prerequisites and all. Alex declares "the problem of getting a sorry-free proof from a paper is done" (though achieving "MathLib quality" elegance still requires thousands of hours).

2. Quasi-Riemann Hypothesis shock: AI proved no zeros of the zeta function exist with real part between 7/8 and 1 (second proof reached 11/12). This region "hasn't moved in seventy years." Full Riemann Hypothesis requires reaching 1/2, and "the methods used fundamentally can't get there—you need a new idea," says Alex. Still, "if a human did this, it's an instant Fields Medal."

3. The "doorman fallacy" and redefining math: Ankur argues reducing a mathematician's job to theorem-proving is like reducing a doorman's job to "opening doors"—it misses the complexity. A mathematician's essence is "thinking deeply"; theorem-proving, theory-building, and conceptual connection are byproducts. Even as AI solves theorems, understanding "what constitutes a good question" and "why the proof works" remains human territory.

Editorial Perspective

Mathematics has walked through the forest with "a candle" for over two millennia, but now holds "a flashlight"—Alex's metaphor suggests this is not a crisis but the entrance to a golden age. Yet the flashlight illuminates extensions of known paths; the ability to "ask questions in the dark" and discover new coordinate systems remains distinctly human. As Andrew Wiles wept saying "I will never again have a moment as powerful" when solving Fermat's Last Theorem, math's meaning lies in the journey, not the answer. In an age where AI delivers answers, can we preserve the value of "climbing the mountain on our own feet" rather than being "dropped at the summit by helicopter"? That is the question.

Source: Lex Fridman Podcast "#503 – AI Crisis in Math: Navier-Stokes, Riemann Hypothesis & Complexity Theory" (October 26, 2026)

https://lexfridman.com/math-ai-crisis/?utm_source=rss&utm_medium=rss&utm_campaign=math-ai-crisis

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.

• --

Anthropic's S-1 Reveals "Fastest Value Creation in History" as AI Capital Demand Threatens US Treasury Market

Speakers

Jason Calacanis (host) × David Sacks (investor, political commentator) × Chamath Palihapitiya (investor, enterprise sales executive at 8090 corporation) × David Friedberg (investor, founder of OHalo, former Google employee)

Bottom line

AI companies now require capital equivalent to 30% of US Treasury issuance through 2027, making them "too big to fail" and turning compute access into a political battleground where democracies must choose between treating AI as a regulated utility or leaving it in private hands.

Three-line summary

• Anthropic targets a $1.5-2T IPO after six years, achieving 85% gross margins and $16.3B H1 revenue—"the fastest value creation of any venture-backed company in history"—but its capital needs now compete directly with sovereign debt markets
• Google's search revenue decelerated for two straight quarters, wiping out $500B in market cap, as AI direct answers structurally disrupt the cost-per-click model that funded the internet era
• Seven to eight AI companies owe $344B in debt and need to raise 30% of all US Treasury issuance through 2027, creating systemic risk where either AI firms or the broader economy faces higher borrowing costs

Three key points

1. The "$80B loss" was accounting fiction — Media reports of $80B in losses were based on convertible note accounting (non-cash); actual burn over six years was only $9-10B. Anthropic holds $112B cash with 85% gross margins, making it "extraordinarily capital efficient" (Sacks). Cost per task improved 10x YoY due to "compute multipliers" (algorithmic improvements, data curation, MOE architectures)

2. Recursive Self-Improvement (RSI) is both moat and existential risk — Anthropic uses its own models for 26% of development tasks (as of August), driving 10x cost reductions YoY. But RSI is the exact "breakaway superintelligence" scenario safety advocates fear most. To stay ahead of open models, Anthropic must keep pushing RSI, creating an "emerging divorce" with doomers (Sacks). Dario's "pacing letter" calling for global coordination was an "alibi"—he knew coordination was impossible

3. Enterprise penetration is only 1% — Customers spending $100K+ grew from 200 (2024) to 11,900 (Sept 2026), but this represents only 1% of ~1M addressable businesses globally where $100K equals <1% of spend (Friedberg). Only 1,900 companies spend $1M+, mostly AI-native startups; "a couple hundred" are Global 2000 enterprises. Million-dollar deals require cross-functional buy-in and ROI proof, but "the ROI case is still a little speculative" (Chamath)

Editorial perspective

AI companies reaching a scale where they compete with sovereign debt markets is not a replay of 1990s internet firms exhausting phone line capacity—it is a collision with the state's own capital-raising function. Chamath's observation that he has "never seen this in [his] adult working life" signals a regime shift where compute may be allocated politically, like oil. The revelation that Anthropic's "pacing letter" was an alibi rather than a genuine coordination proposal shows that competitive logic overpowers safety considerations, raising the question of whether private control remains viable.

Source

All-In Podcast "Anthropic S-1 Shocker, Greatest Business Ever? Google's $500B Wipeout, Midterm Predictions" (October 30, 2026)

https://allinchamathjason.libsyn.com/anthropic-s-1-shocker-greatest-business-ever-googles-500b-wipeout-midterm-predictions

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.

• --