Matt Alldian

Far too many diseases are still untreatable. It’s painful to imagine gradually forgetting your children’s names, forgetting how to hold a conversation, how to recognize your own home, and eventually even how to swallow food, while your family watches someone they love disappear. But that’s the reality for over thirty million people with Alzheimer’s globally, and we have almost nothing to offer them. Part of the reason may be because the brain is so complex, or because it’s lined by the blood-brain barrier, a tightly sealed layer of cells which keeps out most pathogens and toxins, as well as most drugs. That makes it challenging to design a drug that can cross the barrier, reach the right cells, and do what it’s meant to do without any unexpected side effects.

A gorgeous WIP article on medical innovation. A good reminder that intelligence is rarely the bottleneck to progress.

Every dog license New York City has issued since 2014 — 819,323 license records, 34,663 distinct names. Search a name to see where it sits in the long tail, filter by ZIP code, and browse breed and name heatmaps. Counts are license records (dogs renewing across years appear more than once).

New project! NYC Dog Name Explorer shows licensed dog name popularity across the city. Inspiration for this from John Cook

We're breaking five stories today based on verifiable financial data which everyone missed for years.

Effort.News automates investigative journalism with five verified financial stories via marginalrevolution.com

Why do I think I will be able to handle all scams in a few years, when I am already struggling to detect simple AI slop, increasingly ignore cold emails and have to write off whole swathes of social media as a source of information, and I can already see how eager all my peers are to offload all their thinking and writing to chatbot assistants unworthy of that trust, and how many projects or mailing lists have had to clamp down on unvetted contributions (eg. today as I write this, ⁠Project Ladybird)? In a few years, won’t I be the equivalent of a rich old person with declining faculties getting a call from the IRS about how I owe them fines, conveniently payable via gift cards…? And if not, why and how not—concretely?

Gwern announces Guardian Angels.

The BBC Sound Effects Archive is available for personal, educational or research purposes. There are over 33,000 clips from across the world from the past 100 years. These include clips made by the BBC Radiophonic workshop, recordings from the Blitz in London, special effects made for BBC TV and Radio productions, as well as 15,000 recordings from the Natural History Unit archive. You can explore sounds from every continent - from the college bells ringing in Oxford to a Patagonian waterfall - or listen to a submarine klaxon or the sound of a 1969 Ford Cortina door slamming shut.

...

The earliest recording in the collection is probably the world’s earliest natural history sound recording – made onto wax cylinder in 1889 by Ludwig Koch, and later bequeathed to the BBC. Some of the early BBC field recordings were made by David Attenborough, and others by recordist’s such as David Toombs and Nigel Tucker in the 1980s, and Chris Watson in more recent years.

BBC Sound Effects provides thousands of free, downloadable audio clips via DG

If a market researcher wants to know how people will respond to a new product, or a political pollster wants to know how people will vote in an election, the models will give plausible answers. That’s what they do, but the answers will only be plausible, no more. To the extent that the situation were are asking about is novel, exactly when we might gain from accurate predictions, they will be useless.

There’s more in the talk I haven’t covered, it really is an excellent introduction to both the limitations of LLMs and some deeper issues about why psychological science is so hard. Recommended.

Dokyun Lee's talk on why LLMs fail at predicting human behavior with a great write-up from tomstafford.substack.com

In a few short years, Oink grew into a massive community of like-minded music enthusiasts, offering high-quality downloads of virtually every album in existence. “It was like opening a secret door to this incredible world,” Sheridan reflects. “File sharing was everywhere, but you'd never seen this level of care and detail.

What.CD, Oink, and lost communities - what an era I still think about often - via kottke.org

To that end, here’s an even more interesting question around distillation: why exactly is it bad? After all, what are large language models but the distillation of all of the knowledge on the open Internet, scraped by the frontier labs and distilled into the models that are themselves being distilled? Who is exactly being wronged here?

In fact, this paradox is the solution. I believe that open weight models are good for innovation (and, per the above, I think that labs on the frontier will be fine), but it’s a problem to be dependent on China. The U.S. should pass a law that (1) makes explicit that collecting data for training models is fair use, and (2) bars terms of service that forbid distillation, for U.S. companies at a minimum. Stopping distillation — which is literally just querying the API — is nearly impossible; the U.S. should go the other way and lean into a new copyright policy that both indemnifies the labs and also guarantees that what they learned fuels further innovation for everyone else.

Stratechery on Chinese AI models. Good points all around on R&D costs versus COGs, distillation and open source in general.

NYC has been using aerial photography for over 100 years to catalog and measure the city. Thanks to some superb data preservation, you can explore some amazing aerial photography data going back to 1924.

BPD covers NYC's building footprints dataset updated weekly and covering 1M+ structures. Garages make up about a fifth of NYC buildings.

The cost of DNA sequencing has fallen faster than Moore's Law. Since 2001, the National Human Genome Research Institute (NHGRI) has tracked costs at its funded sequencing centers — from $95 million per genome in 2001 to around $500 today.

Genome sequencing costs fell 190,000-fold since 2001. Data here stops at 2023 and I can't seem to find reliable sources since then.

As a new parent, my mind immediately went to movie-viewing volume. With a four-month-old, I am not exactly plugged into movie culture—or really any culture unrelated to said four-month-old. So I wanted to know whether this “parent tax” is universal: Do we watch fewer films as our responsibilities grow?

The answer, unsurprisingly, is yes. In the MovieLens dataset, average rating volume peaks among users ages 25 to 34, then declines with each passing decade.

MovieLens data shows movie taste shifts with age, driven by nostalgia bias via Don't Worry About the Vase