Today's 10

Oregon Senator Ron Wyden and Massachusetts Senator Edward Markey have sent Meta a letter asking a blunt question: does its smart glasses have any safeguard at all against someone recording intimate moments without consent?

The sharpest line is about AI training. The people filmed never agreed, yet the footage may be retained, shown to human reviewers and fed into models. Earlier reporting found men in Ray-Ban glasses walking into massage parlours, filming, and posting the videos to accounts with hundreds of thousands of followers; the people in them often did not know they were on camera.

The marketing runs on A-list celebrities. The reply now comes as a Senate letter. That is quite a journey for one product line.

Only a few years ago, very smart researchers were publishing papers asking whether ideas were getting harder to find. Now Anthropic says its AI has made progress toward a food allergy treatment, and Google's Gemini is credited with helping on nuclear fusion reactor design.

One number stuck with me: research expansion has been estimated to lower average researcher ability by 37% to 53%. AI simply removes the constraint on the number of researchers. Short on people? Build more chips.

These results are still very preliminary, and the author calls them just minor, early efforts.

It took centuries to train a human scientist; an AI reads the whole literature in hours. That gap is hard to digest.

SpaceX has told NASA that Crew Dragon will stop flying by or before 2030, and lunar lander customers have been told Falcon 9 rides are no longer for sale.

The sharpest line is Gwynne Shotwell's: let Boeing have some business, the government paid a lot of money for that capsule. Musk once used 'ILoveNASA' as a password; now government contracts are a rounding error to SpaceX.

Technical brilliance is not the same as commitment. When the focus moves elsewhere, a contract is just paper.

A vulture's stomach acid digests rotting meat and takes rabies, anthrax and botulism down with it. Now H5N1 avian flu is finding them through those same carcasses, and 16 of 23 vulture species are already listed as threatened.

The twist is a feedback loop: the more vultures die, the less carrion gets cleared, the more pathogens spread, and the more vultures die.

I did not expect that protecting a bird that eats corpses would end up protecting us.

Anthropic filed a 300-plus page prospectus but went into detail on one quarter: the second, when it booked a $974 million profit because two of its three months came with SpaceX compute discounts.

The third quarter surfaces only in fragments: revenue up 45% quarter-over-quarter, compute spend up 76%, and more than $430 billion in commitments, most of which have not started to come due.

For a company seeking a $1.4 trillion valuation, leaving blank the line on what it owes over the next 12 months does not look like an oversight to me.

Xianyu writes Funeral AI, a satirical newsletter about China's AI industry. He built a site called MemeBench where ordinary people throw meme prompts at video models and vote on the results.

The open-source, free MiniMax H3 beat ByteDance's Seedance 2.0 and 2.5 — the exact reverse of the technical rankings. He considered tweaking the weights to flip the result back and didn't.

What surprised me most was how he recruited raters: a ten-yuan survey posted in QQ student groups, with some respondents asking whether it was a scam. He calls it China's demographic dividend.

London's churchyards were so crowded that gravediggers sometimes broke apart cadavers with choppers and pickaxes to fit more into one grave, and stripped old coffins for the handles and nails to resell.

When the 1850 Burials Act stopped churchyard burials, a private company trained the dead out to Brookwood Cemetery: coffins travelled first, second or third class, mourners rode separately, and the cemetery had its own stations for Anglicans and Non-Conformists.

One of the Bishop of London's objections was that the train mixed the upright and the immoral in death. What gets me is that even the dead were sorted by ticket class, in a city spooked by cholera. That is scarier than any ghost story.

Spider silk is tougher than Kevlar and stronger than steel, yet labs keep failing to replicate it. A team at San Diego State University went after the black widow's silk gland.

They fed the spiders isotopes and used NMR to see two amino acid side groups briefly stick to each other inside the disordered proteins, pushing water out so the proteins condense into semiliquid droplets and finally get drawn into silk.

What surprised me most is that those same two amino acids line up at the bends of the beta sheets in the finished fibre. The spider's protein soup holds a construction plan humans haven't learned to read yet.

A philosopher applied to graduate programmes in the US and was rejected more than 20 times over three years. He decided to give up for a while and took over an office job from a friend.

He lasted a week. The office job was worse than the sting of the rejection letters, and that is how he confirmed what he really could not give up was philosophy, while the thing to abandon was the job.

He applied for two more years and was accepted at the University of Arizona. What stays with me is the question he ends on: would I know myself better if I gave up?