AIpocalypse.Now
Today'sdoom3.9
Doom 6/10 · 8 stories

Meta Eats Your Photos While Amazon Borrows Another $25B

A safety veteran walks, philosophers get hired, and San Francisco bidding wars price homes a million over ask.

Published · By · Story-level doom average 4.4/10

Meta Turns Your Feed Into a Training Set

Meta shipped two things this week that rhyme uncomfortably. First, the default. Public Instagram photos are now fuel for generative models unless users find the opt-out toggle, which is buried where opt-outs always live. Second, the payoff. Muse, the new image generator from Meta's Superintelligence Labs, can synthesize other users, meaning the platform that harvested your face can now render new faces that resemble it. The Verge caught the synthesis capability. The New York Times framed the launch as Meta catching up in the global race. The Financial Times noted Muse is being wired into Instagram, WhatsApp, and Meta AI as a full ecosystem play.

Read the three stories together and the strategy is not subtle. Scrape the users, train on the users, then sell the users a tool that generates users. It is the most vertically integrated content pipeline ever built, and the raw material is unpaid.

The opt-out framing is doing enormous work here. Wired's write-up is blunt about the mechanics. If you have not gone looking for the setting, you have consented by inertia. That is the industry standard now, and no regulator in either the US or EU has moved fast enough to make it otherwise.

Amazon Borrows Because Cash Flow Is Not Enough

Amazon issued another $25 billion in bonds to fund AI infrastructure. Read that sentence twice. This is the most profitable cloud business on earth choosing debt over operating cash to keep up with GPU demand. Either the capex curve is steeper than the market thinks, or Amazon wants to lock in rates before something breaks. Both readings are grim.

The hyperscaler debt trend is now the most reliable leading indicator in the sector. When companies that print money start borrowing to build, the model is no longer pay-as-you-grow. It is bet-the-balance-sheet-and-hope-demand-holds. If demand holds, everyone wins. If it wobbles, the writedowns will be historic.

The Safety Bench Keeps Thinning

Joshua Achiam left OpenAI after nine years leading safety research. Nine years is a geological epoch in this industry, and Achiam's departure adds to a pattern too consistent to ignore. Every few months, another senior safety figure walks. Wired reported the exit without much editorializing, which is appropriate because the pattern speaks for itself. Whatever internal thesis kept these researchers at the frontier lab is no longer holding.

Meanwhile, NPR reports labs are hiring philosophers to work on ethics and alignment. This is either a serious intellectual investment or a marketing hire dressed in tweed. Philosophy departments have spent decades thinking rigorously about moral status, agency, and value alignment, so the talent is real. Whether the labs will actually let philosophers slow anything down is the open question. Historically, ethicists inside frontier labs have found that their advice is welcome until it is inconvenient.

The optimistic read is that the industry finally understands alignment is a humanities problem as much as an engineering one. The cynical read is that philosophers make excellent shields when the next scandal hits.

The Wealth Surge Has a ZIP Code

The Guardian reports San Francisco homes are now closing more than a million dollars over asking price, driven by AI-adjacent wealth. This is what a concentrated boom looks like at ground level. A small number of people at a small number of companies are cashing equity into the tightest housing market in North America, and the overbidding is setting records that would have been unimaginable two years ago.

This is the trickle-up story of the AI economy. The compute spend flows to Nvidia. The equity flows to a few thousand engineers and executives. The housing bids flow to sellers who happen to own scarce dirt in the right neighborhoods. Everyone else watches from the sidelines and gets an image generator as a consolation prize.

The throughline across all eight stories is the same. Capital, data, and talent are consolidating faster than governance can track. Meta owns the feed. Amazon owns the racks. OpenAI is shedding the people who used to argue for caution. And San Francisco is pricing out anyone who is not already inside the circle. The philosophers arrive just in time to describe what happened.

Sources cited in this digest
  1. San Francisco Homes Now Cost Million More Than Asking · The Guardian · 4/10
  2. Meta Trains AI on Your Instagram Photos by Default · Wired · 6/10
  3. OpenAI's Safety Researcher Departs After Nine Years · Wired · 5/10
  4. AI Companies Begin Hiring Philosophers for Model Development · NPR · 4/10
  5. Meta's Muse Image Generator Can Synthesize Other Users · The Verge · 5/10
  6. Meta Releases Muse Image Amid Global AI Competition · New York Times · 3/10
  7. Meta Integrates Muse Spark Image Into AI Chatbot Ecosystem · Financial Times · 3/10
  8. Amazon Borrows $25 Billion More for AI Infrastructure · · 5/10
Today's doomWeekly column