Layout preview — Direction A: Front Page
One lead + three supporting + briefs tail. Single tone, no colour bands.
PHOTO: EFREM EFRE · PEXELS
Signal & Noise · Episode 7

Anthropic at the Vatican.

2026-05-27 6-day window 2026-05-22 → 2026-05-27
This week
19 voices · 6 days
№ 01 · Lead

Anthropic's institutional moment lands at the Vatican.

Pope Leo XIV's Magnifica Humanitas encyclical draws on Anthropic-style interpretability framing. Chris Olah was at the rollout. The Anthropic Institute was formally announced as a parallel research arm.

The encyclical's name was chosen in dialogue with Leo XIII's 1891 Rerum novarum on capital and labor — casting AI as the labor question of this century. This is the deepest institutional moment for an AI lab so far, full stop.

Olah's reported lines at the Vatican became the most-cited quote in the post-encyclical conversation. Boris Cherny amplified inside Anthropic. Jack Clark now leads the Anthropic Institute as Head of Public Benefit.

We keep finding things that are mysterious, even unsettling. We find structures that mirror results from human cognition. Chris Olah · Vatican

The Builder reads it as mainstream legitimacy unlocking regulatory partnership. The Skeptic reads it as institutional capture working both ways — the Vatican is now in dialogue with companies whose product it doesn't fully understand.

Either way: the conversation about AI ethics now has a Catholic-doctrinal anchor that didn't exist a week ago.

Pope Leo XIV Chris Olah Boris Cherny Jack Clark
Also this week
№ 02

Productivity gaps become publicly audible.

Uber COO Andrew Macdonald admits no proportional productivity gains from AI spend. Marcus calls it the bubble-pop indicator: "if enough other companies report the same."

Narayanan deflates Google's $916-OS claim with independent-eval critique. Daniel Stenberg (curl) reports 4–5× more security reports than 2024, drowning in AI-assisted bug noise.

Marcus · Narayanan · Stenberg · Willison
№ 03

AI-written prose damages relationships.

Paul Graham: AI-written founder emails "feel like being lied to." Armin Ronacher: issue reports laundered through LLMs lose the human voice.

Mollick ships Choosing to Stay Human — explicit framing of what to keep human. Chollet: "Whenever AI tells me I'm absolutely right, my trust drops."

Graham · Mollick · Chollet · Ronacher
№ 04

Software 3.0 becomes teachable.

Cherny ships /usage and endorses "use auto" as the #1 Claude Code tip. Willison ships Datasette Agent + 3 plugins.

Schluntz's "ask without derailing" feature lands. Agrawal et al. publish FST — 3× sample efficient over RL, 70% less catastrophic forgetting. The discipline is teachable now.

Karpathy · Cherny · Willison · Agrawal · Schluntz