Gutenberg Lab — Morning Paper

AI Doomerism as Coordinated Campaign — Anthropic Whistleblower's "Extinction" Claims Under Scrutiny

Featuring: David Sacks (former AI czar, VC) × Jason Calacanis (host) × David Friedberg (scientist, Ohalo CEO) × Chamath Palihapitiya (VC)

Bottom line: The AI safety doomer narrative appears to be an orchestrated campaign to eliminate open-source AI and establish regulatory capture, not a genuine existential threat.

3-line summary

• Jacob Coxen, who worked at Anthropic for six weeks, resigned with a viral tweet claiming AI could kill humanity by 2030, reaching 150 million views and endorsed by Anthropic's alignment lead with a "greater than 10% chance within next decade"
• Sacks presents evidence of coordination: Coxen's near-dormant account was amplified by three EA-funded groups within 15 minutes, with a Wall Street Journal story posted simultaneously, all funded by Anthropic's Series A co-lead Jan Tallinn
• The endgame is a Federal Department of AI imposing standards requiring central monitoring and rollback—technologically impossible for open-source models—creating a closed-model duopoly that benefits economically while government gains speech control

3 key points

1. IPO liability trap: Anthropic seeks trillion-dollar valuation while its safety lead publicly endorses "greater than 10% chance" of civilization-ending risk—creating unprecedented product liability exposure that S-1 disclosures may not shield. Chamath: "No company has tried to go public while saying they're destroying humanity. There is no legal precedent."

2. Doomer track record: 0-for-4: Sacks lists failed predictions—GPT-2 "too dangerous to release" (wrong), thinking models o1 "too dangerous" (wrong), cyberattacks collapsing banking (didn't happen), Dario's job loss claims (half of jobs gone, 10-15% unemployment—opposite occurred with job gains and strong economy)

3. Zero Data Retention is fiction: Chamath states ZDR is "commercially best efforts, can't guarantee it." Friedberg testifies to asking novel scientific questions and later seeing identical insights returned as "novel"—de-identification preserves "approach to problem" which is itself IP, and closed models observe everyone's work to build advantage

Editorial perspective

History shows power structures built on existential threats that never materialize: Al Gore's disproven climate forecasts, Fauci's COVID predictions leading to lockdowns and inflation, Three Mile Island causing the US to abandon nuclear (cost per gigawatt: US $15B, France $4B, China $1B). The pattern is identical: "listen to the experts," demand proof of absence, lock in social hysteria. This time the difference is structural—the doomsayers are Series A investors in the company going public, insiders with economic incentive to eliminate competition through regulation. With open source delivering 50x cost reduction and democratization, we stand at a fork: "everyone gets rich" versus "someone controls the gas pedal." The question is not whether AI poses risks—all powerful technologies do—but whether centralizing control in a federal department partnered with a duopoly is the solution, or the problem dressed as the cure.

Source: All-In "AI Kills Everybody or Doomer Psyop? OpenAI's Math Breakthrough, Nike's $200B Collapse" (September 11, 2026)

https://allinchamathjason.libsyn.com/ai-kills-everybody-or-doomer-psyop-openais-math-breakthrough-nikes-200b-collapse

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

• --

The Sticking Problem: Attention Is Easy, Retention Is the Real Challenge

Featuring: Ollie Forsyth (New Economies host) × Josh Elman (a16z Partner, former Apple Siri AI)

Bottom line: In the AI era, the defining challenge for consumer products is not grabbing attention but making users stop doing something else.

3-Line Summary

• ChatGPT reached 900 million weekly actives, but that's trial, not habit. Josh Elman argues "it's never been easier to get attention, but getting people to stick is harder"
• Successful AI products start by doing one thing exceptionally well, earn trust through that wedge, then expand. Siri's lesson: "You have to earn the right to do more"
• Trust will become the new moat. Products that prove they're "on your side" rather than optimizing for ad revenue or data extraction will reach hundreds of millions

3 Key Points

1. Mainstream consumers are "very, very early": Despite ChatGPT's 900 million number, most use it for "better search synthesis" and homework help. Josh sees three adoption tiers (Silicon Valley edge-dwellers, workforce, mainstream), and mainstream is "barely scratching the surface" of life transformation

2. The substitution test is the core product validation question: "What did you stop doing when you started using this new thing?" If users can't answer, it's additive novelty, not a genuine wedge into their lives. Viral distribution via TikTok/Reels is easy; displacing existing behavior is the real challenge

3. Voice is one modality, not the only future: Josh: "I'm not yet a believer that we're all gonna move to voice for everything." Voice is good for quick capture but "lossy" (you ramble before refining). Typing lets you edit before sending. Voice-only agents "don't really make sense"—successful products will need seamless multimodal switching

Editorial Perspective

The observation that group chats have already become the new social network suggests the next platform might not be a new app but AI-augmented existing messaging. Josh's design principle—"AI facilitates human connection, not replaces it"—could be the dividing line for social AI products.

Source: a16z "What Makes a Consumer AI Product Stick? | Josh Elman" (September 19, 2026)

https://a16z.simplecast.com/episodes/what-makes-a-consumer-ai-product-stick-josh-elman-xHQUEG_W

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

iPhone 18 Pro's Variable Aperture Is a Decoy — The Real Story Is Apple's Image Pipeline Overhaul

Participants

Jay Kasparnakas (Executive Editor, The Verge) × David Amell (Senior Tech Reviewer) × Allison Johnson (Senior Reviewer, Camera Specialist) × Hayden Field (Senior AI Reporter)

Bottom line

The iPhone 18 Pro's headline feature—a variable aperture camera—sees minimal real-world use, as auto mode rarely stops down from f/1.8. The true innovation is Apple's complete redesign of its image processing pipeline, which leverages better sensor physics (wider aperture + 48MP sensor = improved signal-to-noise ratio) to reduce reliance on computational photography, eliminating the "plasticky" over-sharpened look that plagued prior models.

3-Line Summary

• Apple shipped a variable aperture (f/1.8–f/4) on the iPhone 18 Pro, but reviewers found auto mode rarely uses f/4. Allison Johnson had to stage group shots with specific depth arrangements just to trigger it
• The real upgrade is a redesigned image processing pipeline. By leveraging the wider f/1.8 aperture and 48MP sensor's improved signal-to-noise ratio, Apple reduced computational photography (frame stacking, aggressive sharpening), producing more natural photos in both HEIF and ProRAW
• Apple's new "reference image" feature cryptographically signs sensor data to prove authenticity, addressing AI fakes. However, it requires manual opt-in, doubles file size, and only works on iPhones—limiting its potential as an industry standard

3 Key Points

1. Variable aperture is future-proofing, not a present-day feature

David Amell: "The variable aperture is the flashiest upgrade, but the changes that are actually important are in software—the pro controls and the redesigned image pipeline." Auto mode rarely stops down from f/1.8, and Allison Johnson confirmed she had to stage group shots to trigger f/4. David argues this is a "test bed" for when Apple ships a one-inch sensor (rumored iPhone 20 or Ultra)—at that point, variable aperture will be essential to manage shallow depth of field.

2. Image pipeline overhaul is the hidden revolution

Apple confirmed in a briefing they "totally reengineered" the image processing pipeline. The 48MP sensor + f/1.8 aperture provide enough light data (improved signal-to-noise ratio) to reduce reliance on computational photography (frame stacking, aggressive sharpening). Result: Photos look less "plasticky" and more natural in both HEIF and ProRAW. David: "When you have physics on your side—a bigger sensor or wider aperture—you don't need computational photography. Computational photography was invented because the sensor was too small." Allison: "I had a pleasant time with the photos I got. I got the photo I wanted without having to fight the camera as much." However, the ultra-wide camera (smaller sensor) still shows the old over-processed HDR look, confirming only the main camera received the pipeline upgrade.

3. Reference images are clever but wonky to execute

Apple's "reference image" system cryptographically signs each image at the pixel level, storing both the sensor data and processed JPEG. Users must manually enable it in Settings, then shoot in a special mode (main camera only, 1x or 2x). To view the "reference" (pre-processing) version, users send the image to Apple's Private Cloud Compute. File size doubles. David: Unlike C2PA metadata (which breaks with a screenshot), reference images are tied to the device's Face ID encryption—only the photographer can unlock and display the reference. However, it's iPhone-only, so it won't become an industry standard. Apple also supports Google's SynthID (invisible watermarking) as a hedge. Jay: "It's deeply thought through but wonky to execute."

Editorial Perspective

Smartphone cameras have hit a computational ceiling—further gains require larger sensors, not smarter software. Apple's decision to reduce computational processing (rather than pile on more) signals a philosophical shift: physics now allows them to do less, not more. The variable aperture itself is underutilized today, but it's a rational bet on future hardware (one-inch sensors). Meanwhile, the "reference image" feature is Apple's answer to AI-generated fakes, but its iPhone-only, manual-opt-in design feels half-hearted for an industry standard. Ultimately, Apple's endgame is clear: a world where the iPhone is the only camera professionals need, and dedicated cameras cease to exist.

Source

Source: The Vergecast, "The real story of the iPhone 18 Pro's camera" (September 18, 2026)

https://www.theverge.com/the-vergecast

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

• --