🔮 Sunday edition #530: Google’s search slippage; Copyright’s nine-figure gate; AlphaGenome’s diagnostic leap; OSINT decoys++
An insider's guide to AI and exponential technologies
Hi, it’s Azeem.
Courts say fair use for AI training survives—if you can afford hundreds of millions to buy the books you need. Meta and Anthropic can; most newcomers can’t. At the same moment, AlphaGenome can read DNA and predict its function. Money is ring-fencing words just as code liberates life. All that and more in this week’s edition.
What’s next for Google
In this week’s Saturday commentary, I analyse the future of Google search. It’s quite the pickle, with much to play for.
🤔 How serious is Google’s ChatGPT problem?
Two years ago, I argued that Alphabet, which owns Google, faced a “GPT Tidal Wave” because “the start page of the Internet is shifting further from the browser and Google.com, replacing dozens or more Web searches each day. ChatGPT is preferable to open multiple tabs from a Google search and continuously backtracking.”
See also: If the DoJ rips Chrome from Google, the browser wars will ignite overnight. For a glimpse of that AI-augmented future, look at Dia, a browser that lets you chat with your tabs.
Fair use, for the few
Courts just blessed Meta and Anthropic—and quietly priced most AI startups out of the game. The two firms dodged copyright liability this week, a headline that reads like a broad Silicon Valley victory. But read the fine print: The judges hint that training is “lawful” only when every sentence used for training is properly licensed or bought outright. Tactically, it’s a fine day for Anthropic, Meta and peers who can write nine-figure licensing checks; strategically, it prices out new startups—unless we overhaul copyright to reward creators without sealing the door to new ideas. What I argued last year still stands:
Copyright is, and always has been, a compromise between incentivising creators and increasing social welfare by promoting cultural participation, sharing knowledge and affording creative freedoms. That compromise is highly dependent on the nature of the technology. And LLMs (and the wave of digitisation from the decades before) have made the need to rethink that economic compromise that is copyright more urgent. I prefer models that separate authorship and attribution from these decades-long economic rights, favouring IP hoarders. Reform could include substantially shorter copyright terms, more robust protections for fair use, and clearer thinking on how copyright holders and licensees are incentivised.
See also:
A privacy wrinkle is emerging in these cases—a court order requiring OpenAI to retain chat logs could pave the way for broader law-enforcement data demands.
Denmark is amending its copyright law to give individuals legal ownership of their own body, facial features and voice in a bid to tackle deep-fakes.
Geist in the machine
Economist
asks: If a system can watch itself think, does it drift toward self-awareness? Kant already pictured the mind as a virtual-reality projector, stitching raw sensations into a coherent world. Hegel pushed further, claiming that mind and world share a common conceptual grammar—they reflect one another. Large language models make that mirror visible. Rather than anchoring words to things, they locate meaning inside the web of inferences we humans already trade. Each prompt you give ChatGPT taps into that shared spirit, turning bare text into a promise, a joke, or a threat. Anthropic’s Constitutional AI goes a step further: Claude reviews its own drafts against a charter of norms and rewrites when they clash. Engineers have rebuilt, in code, the “game of giving and asking for reasons.” The result sounds like a conscience, but it isn’t magic; it’s our own norm-checking loop rerun in silicon. For us, that same loop is anchored in bodies that bleed, and the stakes—pain, joy, accountability—turn a clever rule-check into the lived experience we call consciousness.See also: Some people are starting to treat ChatGPT like a God.
Closing the bio loop
Biology is picking up software-like speed. Today, a rare-disease patient often waits years for a diagnosis and even longer for treatment. Imagine compressing that diagnosis and therapy into a single hospital week. We may be closing in on that reality. DeepMind’s AlphaGenome can read DNA and flag potential disease-causing mutations in minutes, while a benchtop prototype called NANOSPRESSO promises to print custom gene therapies before a nurse’s shift ends. Though still only prototypes, they point to a future where the world’s 300 million rare-disease patients could get answers—and care—within days, not years.
See also: ARC Institute’s new STATE model, which simulates how an entire cell reacts to mutations and drugs. Think of it as a rehearsal studio for testing edits before you print them.
Weaponising openness
Operation Midnight Hammer proved that a fake trail of stealth bombers can flood the world’s open-source feeds. While the true strike force slipped silently toward Iran’s real nuclear target, another group of B-2s broadcast radio chatter while heading toward Guam—the very breadcrumbs OSINT sleuths amplify. The Pentagon didn’t just hide its moves; it saturated public channels with noise, tricking analysts into chasing a story that never mattered. Two strategic lessons follow. First, public data are now a liability: militaries must defend against civilian sleuths as well as foreign spies. Second, OSINT is both a threat and an Achilles heel—for forces that lean on it, those same public channels become exploitable chokepoints. In short, open-source intelligence is no longer just a risk; it’s a live domain of warfare, and every operation must account for it.
Elsewhere
Demographic collapse ≠ climate fix. Even a population crash cools the planet by <0.1 °C by 2200, says a new NBER paper.
- highlights the pervasive lack of high-quality, reliable data in Sub-Saharan Africa, leaving policymakers on shaky ground. Case in point: a GDP “rebasing” once boosted Ghana’s GDP by 62% and nearly doubled Nigeria’s overnight.
Weight-loss shot vs migraines. Liraglutide (GLP-1 class) cut monthly migraine days in half—hinting these injectables could tackle a neurological disorder that affects one in seven people worldwide.1
Spit-based birth control just became a reality: Europe has approved Inne’s Minilab, the first saliva-based contraceptive, which tracks progesterone, is hormone-free and delivers pill-level efficacy.
Abundance, debated.
notes that while online leftists rail against his “abundance” agenda (EV516), progressive politicians are quietly adopting its pro-growth, bottleneck-breaking playbook.Music’s universal cues. Listeners worldwide, hearing songs from other cultures for the first time, guessed their purpose (lullaby, dance, love, healing) 42% of the time—double that of chance.
Starry, starry night. Chile’s Vera C. Rubin Observatory, which has the largest digital camera in the world at 3,200-megapixels, has unveiled its first images.
Virtual closet goes kinetic. Doppl, Google Labs’ new app, turns any outfit photo into an AI video of you wearing it.
Worried about your cat? An AI can now score feline stress—sending notifications straight to your phone.
Although the sample size is small, with only 31 patients.
The thing with GLP1 is that it treats so many things (now including migraines in a small study) that being overweight and being a candidate for GLP1s is likely becoming a health advantage 😀
re health care in the near future-diagnosis and treatment sounds like a day or less to me