“Whether hot or cold water freezes faster remains unknown.”

Thus begins a wonderful Quanta Magazine article about the “Mpemba effect”, named after a Tanzanian teenager who saw something funny happen to his home-made ice cream. Reality is complicated.

“A lot of people simply won’t read a 15-page whitepaper, but will be impressed by flowcharts. By making the language of Web3 meandering and impenetrable and by building a culture that is very self-referential, investors make criticism harder to come by.”

Today’s Galaxy Brain newsletter is about Web3, but replace “whitepaper” with “manuscript” and “investors” with “researchers” and you get bad science in a nutshell.

“If you are trying to figure out a thinker and his or her defects, see if you can spot that person’s “once-and-for-all” moves. There will be plenty of them.”

Cowen is right, though we can debate whether early closure is a defect or a feature.

“This contains a combination of various life-extension medicines (metformin, ashwagandha, and some vitamins), and covid defense gear: a CO2 meter… masks, antigen tests and fluvoxamine.”

The reddest of crypto world’s red flags is their belief in longer life through chemistry.

“The convergence of genomics of the cancer—be it from the person’s DNA or tumor directly or the blood (known as liquid biopsy)—matched with the appropriate therapy is leading to outcomes that are being described as ‘unheard-of’ by expert oncologists.”

So writes one Eric Topol, who seems to have made a career out of telling high-status people what they want to hear. For reasons why most of what he wrote is wrong, take a few minutes to watch Vinay Prasad’s reviews of the articles in question.

The trend disrupting medicine back when Dr. Topol was writing about its Creative Destruction turned out to be not technology but rather the opioid epidemic. As a non-expert oncologist I would wager something other than liquid biopsies is leaving its hallmark on the field right now.

“The speed of review times and increasing number of FDA-approved cancer medicines has long been used as a metric for successful regulatory processes and improvements in patient outcomes.”

Here are three good reasons why that isn’t so, to which I would add Goodhart’s law.

The NYT dostarlimab article is reverberating through international media with predictable consequences: being hailed as a miracle cure for cancer. I wrote about it in Serbian, and Google’s translation of it is readable, in an AI-generated spam sort of way.

Residency programs’ publication requirements generate the worst kind of scientific BS — useless case reports and rushed chart reviews that clog lit searches without creating any meaningful insight, all while wasting residents’ precious time. Please stop.

“Brilliance represents an upper bound on the quality of your reasoning, but there is no lower bound.”

A frameable quote if there ever was one, from an excellent blog post on why smart people may believe dumb things.

“At a conference a few months before the pandemic, a scholar told me how, in his department, everyone wrote lengthy pre-analysis plans that would, in theory, constrain P-hacking. In practice, he admitted, researchers could give cherry picking free rein, counting on the fact that no one has the time or patience to read a 100-page pre-analysis plan and compare it with the later publication.”

This is from an opinion piece in Nature about reproducibility of science in general, but the sentiment holds for clinical research (arguably, clinical research is even worse, for those hailing results of a clinical trial as game-changing often can’t be bothered with reviewing even a brief ClinicalTrials.gov registration and its history of changes, let alone a 100-page document). [Note: You could argue that open analysis plans at least give authors a reason to fear, as a deviation from the plan could result in negative letters to the Editor, retractions, or worse yet, shaming on social networks. In practice, it is those who call out deviations who are more afraid of name calling, Twitter mobs, and employer @-mentions for… being too aggressive, I guess? ]

Once you incentivize scientists to produce hundreds of pages of text that no human will ever read and no algorithm will parse just so a box or two can be checked, do not be surprised at the outcome.