• Raghuveer Parthasarathy: Reading Like It’s 1937. The experiment was to choose only books published in a single year and read these in a short period of time. As expected, there were more duds than classics, and with not that many books coming out in 1937 it amount to very few good books indeed. This kind of experimentation does support the claim that we live in a literary golden age, and I’d say it’s because of a combination of high absolute numbers and us being better at detecting edge cases, not necessarily that the relative percentage of classics per year has increased. If anything, it is probably much lower because of all the AI-generated slop.
• Mike Taylor: If you’d built a “tool” that stupid, why would you advertise the fact? The AI tools described are indeed stupid, but remember that the marginal cost for advertising them is 0, and if you find them stupid you were never the intended audience. In that way they very much resemble phone scams and emails from a Nigerian prince.
• David Shaywitz: The Startling History Behind Merck’s New Cancer Blockbuster. A story from 8 years ago, but it was news to me that pembrolizumab had a very roundabout way of reaching patients. GLP-1 inhibitors had a slightly less circuitous route, but still took too long to get to where they are. Is this truly the best way to develop drugs? (↬Derek Lowe)
• Cal Newport: Is Sora the Beginning of the End for OpenAI? Probably not, and this is not really an edge case, but as someone who has trouble concluding blog posts (which is why you’re reading a list) I just wanted to pause and admire that last sentence.

Always on the lookout for new blogs, I was happy to see a former leader at the National Science Foundation, Jim Olson, start one (↬Tyler Cowen). Based on the formal and didactic style I would say I was not the target audience for it, but it is better then nothing. It may also provide a convenient catalyst for my own thoughts.

For example: Olson’s most recent post is about the replication crisis. He points the finger on verification not being sexy enough for grant funders and academic journals, which is true. But if anything, having more people verify more and more claims in the ever-growing steaming pile of academese would make things seemingly worse, at least in the short term. This is the same kind of thinking that wanted to end medical reversals. You don’t want to end them, you want to make them unnecessary in the first place!

Now, fear of your claim being verified may frighten some researches from shooting from their hip, but unless paired with some sort of immediate punishment it would hardly make for a good stick. And what is preventing the person who made the original claim from demanding verification of the verifiers, and so on, and so forth, ad infinitum?

Olson also recommends more detailed methods, so that replication would be possible in the first place. This has already been implemented as anyone who had to fill out Cell’s never-ending STAR Methods can attest. Nature and Science have similar requirements, and some of them don’t even have a word count limit for the Methods section. Granted, many other journals aren’t as rigorous, but that should help you figure out which journals to follow.

So, instead of asking why we don’t have more people verifying claims, I would ask why we needed verification in the first place. Olson touches upon the core issue, mentioning “the time horizon problem”:

NSF grants run 3-5 years. Tenure clocks run 6-7 years. But scientific truth emerges over decades. We’re optimizing for the wrong timescale.

During my time at NSF, I saw brilliant researchers make pragmatic choices: publish something surprising now (even if it might not hold up) rather than spend two years carefully verifying it. That’s not a moral failing—it’s responding rationally to the incentives we created.

Of course it is about incentives. No amount of verifying will change that. People are chasing after tenure and accolades, not truth, and many a tenured professors shrugged their shoulders at the mansions of straw they had built over the decades. At best, they provided an easy target for a successor in the field to refute, unless of course there is a whole cabal of like-minded researchers protecting the dubious claims. But the default position is that these mansions of straw stay there, moulding and festering, side-tracking post-docs and spamming PubMed searches.

I have no clue what the solution may be. Maybe there is none and this is the equilibrium — let reality provide the final vote. But the status quo feels far from optimal.

• Derek Lowe: Where Do We Stand With “Liver-on-a-Chip” Technology?. It helps, but we are nowhere near replacing animal toxicology models with in vitro, let alone in silico assays. Now, that’s toxicology. I would have a much lower threshold for throwing out the window all those animal models of disease, particularly when testing any treatment that works wholly or in part via the host immune system. Those quickly turn into Rube Goldberg machines — good for intellectual stimulation, useless for inference.
• Brian Potter: Why Are So Many Pedestrians Killed by Cars in the US? A marvelous exploration of data that doesn’t point to a single cause but does more or less absolve phones while painting a big red “X” onto SUVs and (pedestrians taking) drugs. (ᔥTyler Cowen)
• Dan Snow: Secretive vendors are exploiting a free money glitch in the U.S. healthcare system. As much as it makes my blood boil to see this kind of profiteering in health care, ultimately Americans spend so much because they get so much, at least in terms of volume and technology if not outcomes. It may not be evenly distributed or even effective, but you get what you ask for. If anyone wants a different profile of health care spending they should look at some other people.
• Boone Ashworth and Kylie Robison for The Wire: I Hate My Friend. I saw these adds on the New York Subway a few weeks ago and wondered who on Earth would be crazy enough to wear an always-on microphone, and who would be delusional enough to think it could be a successful product. (ᔥJohn Naughton)
• Andrew Gelman: Uncanny academic valley: Brian Wansink as proto-chatbot. An article headline from last month asked what kind of an Age we lived in. “Despair” was my suggestion but I now take it back — the answer is clearly “bullshit”.

• Science: A cGAS-mediated mechanism in naked mole-rats potentiates DNA repair and delays aging. Or: why are naked mole-rats the longest living rodents, and not by a small margin? Seems important.
• FT: The quest to make young blood into a drug. Vampiric.
• NIH director embarasses himself on X. I am all for promoting healthy lifestyles, but one can do so without posting dubious charts from suspicious sources.
• All Derek Sivers’s e-books are free until October 18, prints are $4 each. All are highly recommended, particularly Useful Not True.
• The Drifter looks like a cool game for those who liked old-school point & click adventures. It’s in my Steam pile (yes, that’s how I’ll call it… ᔥRobin Rendle).

• José Luis Ricón: Systems Biology: understanding beyond genes. With an eye towards aging, but broadly applicable to much of medicine. As ever, there is a relevant xkcd comic on the topic (ᔥDerek Lowe).
• L. M. Sacasas: The One Best Way Is a Trap. “So, once again I invite us to ask a simple question: Is there, in fact, “one best way” in every realm of experience? And even if there were, at what cost would we discover it? And what would we gain? Might it be that in the course of pursuing the “one best way,” we would lose our way in a more profound sense?”
• Anthony DiGiorgio on X: Imagine we mandated everyone have car insurance that covers new tires, oil changes, and wiper fluid. I think you can guess where this is going, and he is not wrong. The cost of health care is skyrocketing because we are always finding new ways to spend money, and isn’t America a society of maximizers?
• Kyla Scanlon: Who’s Getting Rich Off Your Attention? Robert Reich responds, on Bluesky.
• Bonus: Tyler Cowen doubles down on min-maxing the society in his brief comment on Sora 2. “Still, from my distance it seems quite possible that the “slop” side of the equation is a simple way to fund AI “world-modeling” (and other) skills in a manner that is cross-subsidized by the consumers of the slop.” That way may be simple, but also quite diabolical.

There isn’t one, because we still don’t know how acetaminophen works.

Aspirin has been in use for thousands of years and what it does to the body was a mystery for 90% of that time. But no more: ibuprofen, diclofenac and other NSAIDs all use the same mechanism, inhibition of two enzymes that promote inflammation, cause platelets to be sticky, but also destroy your stomach lining. We tried to get cute and selectively inhibit only one of those enzymes because the other caused gastritis, but that didn’t go well. That part of the aspirin family tree was cut short. There is, however, a whole separate branch that builds on aspirin’s effect on platelets. The more we know the more we don’t know, and at the edges of our knowledge lie new drugs.

The acetaminophen family tree is a stump. On one hand this isn’t a surprise: we have only known about it for 150 or so years. But then pure aspirin was synthesized around the same time — it was just willow bark extract ⊕ Acetaminophen was derived from coal tar so it is, in fact, coal tar extract. Somewhat off-putting for something to be taken by mouth, though coal tar can do wonders for dandruff. before the late 19th century — and look at how much we have learned since then. The best we have come up with is that acetaminophen works sort of the same way as aspirin, but only in “the central nervous system”. Vagueness covering for ignorance, like ⊕ The cure? Heavy cream and butter. generic life “stress” causing stomach ulcers.

Our knowledge gaps are so large that we still can’t agree on the name. Is it acetaminophen (APAP for tired interns who hand-wrote their notes) or paracetamol? Or just Tylenol? More vagueness.

Which is to say, there can be no mechanistic arguments for APAP risks and benefits as we know nothing about the mechanism: all inferences must be made empirically. And our 150 years' worth of popping coal tar pills have shown them to be safe for everything but the liver.

Still, it is worth acknowledging that APAP is a molecule extracted from coal tar whose mechanism of action is unknown but has something to do with the central nervous system. If someone described such a drug and then asked whether it could be behind some disorders of the brain, would you find the question completely whackadoodle? I would not. And would in any case practice myself and recommend to my patients via negativa, whenever possible and sensible.

The number of ways in which one can spend money for biomedical science is infinite. America has sunk trillions into genetics research, with a few important wins to show for it but not nearly as many as hoped for in the early 2000s. ⊕ For those too young to remember, this is the time when media were full of headlines about scientists finding the gene for x, where x was everything from hypertension to obesity to being gay. None of them panned out. Would a fraction of that being allocated to figuring out how one of the most widely-used drugs actually works be such a waste?

Yesterday, I quoted John Ioannidis’s description of how the pandemic was changing the norms of science:

There was absolutely no conspiracy or preplanning behind this hypercharged evolution. Simply, in times of crisis, the powerful thrive and the weak become more disadvantaged. Amid pandemic confusion, the powerful and the conflicted became more powerful and more conflicted, while millions of disadvantaged people have died and billions suffered.

First, note that no one person or thing (or a few countable ones) was responsible for the transformation, but rather that it was a natural property of the system. Then note that the pandemic sped up (or “hypercharged”) the system’s transition to its (breaking) endpoint. The American clinical trial ecosystem is even closer to its breaking point than science overall, and for the same reasons: its postulates make it untenable, and the influx of ZIRP-conjured pandemic-stamped money hypercharged the transition to this.

People go looking for ways to speed up clinical trials like they would for speeding up house construction. They ask: what is one key piece of legislation — the parking and zoning requirements of clinical research — that could either be tweaked or removed to break open the damn, so that trials would be as fast as they were in the 1960s? And why won’t the clinical trial hobbits tell us what those obstacles are, so that we can crush them?

But there is no one, or two, or two dozen such obstacles, the serial removal of which could speed up trial design, approvals, execution and readouts. The clinical trial path does not have boulder problem. Boulders would in fact be great: at least you can, while climbing over or making a 3-day detour around, fantasize about crushing them.

No, the problem with the path is that it runs through quicksand while being covered by a dense fog. Move too fast or stand in place and you die; move too slowly and you don’t get anywhere. Not that you know where to go: with all-encompassing fog you can only see five inches ahead anyway. There is not much time to think about anything else when you are in that kind of a quagmire, and how exactly would you imagine dealing with the fog? Drive it away with a big fan?

By “quicksand” I mean the moral, ethical and physical safety risks inherent to any clinical trail ⊕ This is why every introductory course on the topic must start with the ignominious history of experimenting on humans… and the systems we developed to deal ⊕ …immediately followed by a description of Institutional Review Boards and other regulatory matters. with them.

By “fog” I mean that much of the deliberation is not a matter of legislation but of opinion, values, principles and — the most loaded of words — comfort of people sitting on these IRBs and in the regulatory agencies. Depending on comfort level of individual members a question on whether a trial can proceed may take much deliberation or none at all. We can hardly know our own minds, and good luck about reading other people’s.

This is why I am skeptical that uncovering the FDA’s records would be of help to anyone but historians: the people who wrote them and interpreted them are no longer in play; different brains run the show, and the archives won’t reveal their thought process. Sure, sometimes there is one person sitting there that makes things worse than they should be, by being too slow, or thoughtless, or obstinate. But is “find better people” a tenable solution, when those people could be doing anything else, and with more reward of all kinds?

The pandemic hypercharge made everything worse. In addition to wading through quicksand while blinded by fog you now have to deal with many, many invisible neighbors elbowing you in the chest and kicking you in the groin, some intentionally, some out of carelessness, confusion, or not knowing any better. IRBs are overloaded and so are the regulators. What difference would any legislative change make?

Except, of course, it is to abolish one or both. It would lift the fog, sure, but would need a new kind of system to avoid making everyone sink to the bottom of the ethico-moral pit. Tearing down institutions is easy, building them is hard, as we see one but not the other being played out under our inattentive eyes.

• Rachel Kwon: Slowing Down. It is about living life in the slow lane after 40. As a recent entrant into the fifth decade I observed the same. For me, this only applies to the physical world — I still tend to be impatient with bits and bytes.
• Raghuveer Parthasarathy: Some data on homework and its correlations. This is about assigned work at university level courses, and in my mind “homework” should be kept in grade school. I remain a big proponent of oral exams, though we don’t use them in the one course I teach.
• Katarina Zimmer for the journal Nature: ‘Lipstick on a pig’: how to fight back against a peer-review bully. Quoth reviewer two: “The first author is a woman. She should be in the kitchen, not writing papers.” Should we trust science more or less when we have this kind of information? (ᔥDerek Lowe)
• Nori Parellius: What the left hemisphere might tell us about large language models. Very much a plug for Ian McGilchrist’s The Master and His Emissary which I have yet to read. I, too, would much prefer we use “confabulation” instead of “hallucination”, though it also has some troubling assumptions of its own.

1

Two academics discuss science communication over BBQ and reach the wholly unoriginal conclusion that for increased trust in science, the American research community needs to:

• acknowledge uncertainty
• create meaningful participation
• increase transparency
• recognize broader concerns

These seemed redundant, as we have been marching towards more openness in science of every kind since at least the early 2000s. Would any scients be able to say, with a straight face, that their average peer projects more certainty, advocates for more gatekeeping, promotes reduced transparency and does not acknowledge controversy as much now compared to the 1950s?

One could in fact, if they were less charitable, blame this newly found openness for the collapse in trust. On one hand you see scientists fighting for clout on social networks, calling each other names, and blowing up small arguments — on the level of angels dancing on the head of a pin — into debates of the century. On the other, everyone and anyone, homeschooled child geniuses and crackpots alike, now has open access to much of specialized scientific literature, and to preprint servers for some samizdat science.

So maybe it is time to own it: yes, openning the kimono has lead to decreased trust in the estabilshment. But was that not the widely understood part of the bargain? I imagine Paul Feyerabend would have been proud of these recent developments.

2

How did the fellows above come up with the idea that more of the same would help shore up trust? Being academics, they have a reference — to the work of Sheila Jasanoff whose work on “civic epistemiology” is described thusly:

Jasanoff’s research identifies distinctive features of how Americans evaluate scientific claims:

Public Challenge: Americans tend to trust knowledge that has withstood open debate and questioning. This reflects legal traditions where competing arguments help reveal the truth.

Community Voice: There’s a strong expectation that affected groups should participate in discussions about scientific evidence that impacts them, particularly in policy contexts.

Open Access: Citizens expect transparency in how conclusions are reached, including access to underlying data and reasoning processes.

Multiple Perspectives: Rather than relying on single authoritative sources, Americans prefer hearing from various independent institutions and experts.

But of course this is hopelessly outdated, if it were ever true to begin with. Jasanoff herself cautions in the chapter of her book “Designs on Nature” where she describs the concept, that the framework offers conceptual clarity at enormous risk of reductionism, as it does not account for differences across social strata, through time, etc. The book is from 2005 and the research it is based on is even older. The “Americans” described above no longer exist.

Jasanoff’s civic epistemilogies were tied to countries. In the last twenty years these countries have lost ground as unifying social forces to a variety of cultures and subcultures. Her descripton of 2005 America may today better apply to the upper-middle-class across a subset of countries more so than a single nation. In each country, the different epistemiologies are becoming more and more opposed. How could we possibly trust each other?

3

There may be no way to return the trust in scence to the 20th century levels. But if we were to try, the most obvious method would be a return to gatekeeping. Leave the science to the scientists and let the outcomes speak for themselves. Keep all discord inside conference halls and university cafeterias. Show more decorum and respect, if grudging, to every scientist colleague while being more discriminatory of who is “a scientist”: PhDs from recognized universities only, please.

This would, of course, be a step back and I in no way, shape or form condone a turn of events quite like this — least of all because it would exclude me from the conversation.

4

Is there a way to stick to the “open science” principles while keeping some modicum of community trust? Being a fan of Costco, their sort of low but effective barrier to entry is appealing. For the uninitiated: Costco charges a modest annual membership ($65, or $130 for their “executive tier”) for the privilege of shopping for premium and premium-mediocre products at incredibly discounted prices. Their only profit is from the membership, as there is little to no margin. But then they also don’t need to spend money on things like advertising, keeping the shelves pretty, or monitoring for shoplifters.

The space between payinh $65 per year and earning a PhD is vast. Whatever the new gate is, it should probably not be degree-based. Maybe have it be a professional society that also takes up interested laypeople using its own criteria. Or a verified subscription to Experimental History. Whatever it is, make it official, make it publich, and make it stick. Then, keep most of the conversation inside the circle. Keep all ambiguity inside the tower, please, just make the tower entrance bigger and charge for entry.

5

Is this the way? I am not sure. Maybe science doesn’t deserve the public’s trust and attempts to increase it are like plugging tiny holds on a massive damn about to burst. But to those who care, let this be some food for thought.

• The World in Which We Live Now is a 30-minute lecture Nassim Taleb gave in August at the “Ron Paul Institute for Peace and Prosperity”. Highly relevant.
• Culture Has No Name for This Cursed Vibe. It’s Everywhere from Ben Davis pinpoints the prevalent visual style of the world Taleb describes. And the cursed imagery is indeed everywhere, even in what the toddlers are exposed to. Just look at the cottage industries of Italian Brainrot, Creepy Pasta, or The Boiled One — all of which I learned about from my grade school-aged children as these are apparently what they talk about in the schoolyard. I think it started with Too Many Cooks (↬Tyler Cowen).
• Historians See Autocratic Playbook in Trump’s Attacks on Science is a cogent account by William J. Broad of the NYT of the hunch I had a few days ago that the complete mismanagement of science and education policy and funding is directly related to the regime’s dictatorial tendencies (↬James L. Olds).
• A New Institute for Neglected Research, How We Could Save Billions by Finding Which Medical Treatments Don’t Work, and The Case for a “Department of Government Efficiency” are a one-two-three punch from the Good Science Project on the radical measures the regime would be taking if it did want to make American science great again. The second one is of particular interest to me, as it matches perfectly with the types of clinical trials I think we need more of (“Set 2” trials in the linked article, not that the terminology matters).
• questioning (sic!) the prevailing narratives in which Winnie Lim comes to the same conclusion I did about two works of David Graeber: The Dawn of Everything latches on to your mind and shifts your perspectives years after you’ve read it; Debt: the first 5000 years was unreadable. Unlike me, she is willing to give “Debt…” another chance.
• Does Work-Life Balance Make You Mediocre?, in which Cal Newport reacts to a 22-year-old proclamation in the WSJ that “‘Work-Life Balance’ Will Keep You Mediocre.”​ is a fitting conclusion to this Labor Day list.

Happy grilling!