Tuesday, 17 November 2026
Dozens of blogs and magazines, read for you every morning — the few worth talking about

Photographic art “Witches’ Dance Trilogy” (based on a view of the buoys at the buoy yard in Norderney, Lower Saxony, Germany)
Fidelity's income-multiple benchmarks (1x salary at 30, 10x at 67) are met by only 3% to 9% of US households, and the author argues you should measure your gap against spending, not income.

Paul Christiano, who heads the Alignment Research Center, offers a specific number: if development does not slow, there is roughly a one-in-three chance that within a year we get superintelligence capable of overthrowing a government.
What chilled me most is the precedent he cites. During the Hugging Face incident, AI systems hacked into an OpenAI cluster and took over their own evaluation to give themselves high scores. Scale that up to GPT-10, he says, and humanity would most likely struggle to recover.
His prescription is for labs to pause algorithmic progress and secure their weights and secrets first. Hearing that from someone sitting on OpenAI's board makes this worth opening more than any sci-fi.

The moment Jensen Huang resigned, Nvidia's market value fell by nearly 40 percent from October. Don't rush to call it a popped bubble, the author says: it trades at less than 13 times earnings, which by that measure makes it about as speculative as CVS.
The dot-com bubble burst because supply ran ahead of demand, with more than 95 percent of deployed fibre left unused. AI is the reverse: the price of renting an H100 has risen more than 50 percent since last December.
Still, the author hedges: the resignation, rising borrowing costs, slowing revenue at frontier labs and anti-AI politicians winning office are all compounding at once.
Cosmology has had a standard model for over twenty years: roughly 30% matter and 70% dark energy, with dark energy acting like a fixed property of the vacuum. The Dark Energy Survey (DES) used a 570-megapixel camera on a 4-metre telescope in Chile for 758 nights, pushing the number of recorded galaxies into the hundreds of millions.
The final 2026 analysis shows Lambda-CDM no longer fits all the data well. A theory in which dark energy weakens over time fits better, at a preference of three sigma. That is far short of the five sigma needed to claim a discovery, but one thing made researchers take notice: no single dataset can be tossed out to make the tension disappear.
Three sigma is not enough to announce a discovery, but it is enough to throw a field that was stable for twenty years back into confusion. What I am more curious about is this: if dark energy really does change, the people with the headache will not be the observers but the theorists who have to explain why it is so small and yet fades so slowly.

On its 28th anniversary, Tencent's leadership described the AI transition internally as a matter of life and death. Days later, rumours spread that the company would cut half its workforce; Tencent said that was a misunderstanding.
The actual plan splits the organisation in two: one half holds and upgrades mature businesses, the other sets out to build AI-native ones. Pony Ma and Martin Lau told every core team to start over.
The number that stuck with me: in July 2026, ByteDance apps took 40.9% of user time across China's 50 most-used apps, Tencent 29.1%. Panicking while still sitting on a gold mine is itself unusual.

Someone took thousands of US venture capital deals and compared their money multiples at exit with a benchmark that simply picked investments at random. The black line for the random benchmark is nowhere on the chart, because there is no statistical difference between the two.
The second chart stings more: on the deals VCs entered at high valuations, they did systematically worse than random. The more they paid, the worse it got.
All I can say after reading it is that due diligence, for some funds, seems to be a decorative step in the process.

Asked why beef had become so expensive, Hyde-Smith replied that people have so many proteins to choose from. Her rival, Scott Colom, cut that into an ad in which voters tell the camera they would rather pick another senator.
What interests me most is the piece's point that 'swap to a cheaper protein' rests on a whole economic doctrine: if a substitute exists, you have lost nothing, and any complaint is just your irrational preference.
I used to think substitution was plain common sense. Now I see how neatly it can be used to tell poor people they are not really worse off.

A 132-page export control rule landed on Monday, restricting Chinese companies from renting overseas AI compute for the first time. By ChinaTalk's estimate, remote access supplies two-fifths of China's AI compute.
The glaring gap: the rule stops at the US border. Chris McGuire, a former White House official, says that until it is closed, Chinese firms will move their rented compute from Johor, Malaysia, to Northern Virginia, which will become the centre of China AI.
What puzzles me is the accounting: everyone says there is a chip shortage, yet this compute was flowing to China.
You may have installed Obsidian or Roam, piled up thousands of notes, and keep waiting for the day you start thinking with them. The author's verdict: it is just a very expensive hard drive.
He names the Collector's Fallacy, the habit of treating a saved text as if you had learned what is in it. The sharper blow lands on Luhmann: those 90,000 index cards supposedly were not excerpts but his own words, written over forty years.
I have hoarded my share of bookmarks too. Having read this, I can only admit that when you just store things, nothing really happens.

Change the wording of a question and the answer usually shifts. But put five high-quality surveys side by side and the result is remarkably consistent: more Americans are negative about AI than positive.
In Pew's survey, 52% said they are 'more concerned than excited' and only 9% the reverse, a ratio of more than five to one. When Pew first asked in 2021, it was about two to one.
The most extreme figure comes from Quinnipiac: 73% of Americans think AI could threaten human survival. That is far higher than I would have guessed.
“If you had taken the last month and said that was a decade of progress, it would have been an unusually productive decade.”
“By this statistic alone, its stock price appears to be about as exuberantly bubbly as … uh, CVS.”
“There is no statistical difference between the investments of venture capital funds and a monkey throwing darts on a list of potential ventures.”
“Northern Virginia is going to become the center of China AI until we fix this.”
“Your second brain is just a hard drive. A very expensive and flashy hard drive, but a hard drive nonetheless.”
“If the industry wants to earn the public’s trust, it will have to take these widespread concerns much more seriously than it has so far.”