I’m familiar with Adam Mastroianni’s thesis that science is a strong-link problem — here is at least one mention on this blog — and I am certainly familiar with the recently uncovered shenanigans in Alzheimer’s disease research, but I never thought to connect the dots in the latter refuting the former. Well, paleontologist Matt Wedel did just that:

I’m going to wane philosophical for a minute. In general I’m very sympathetic to Adam Mastroianni’s line “don’t worry about the flood of crap that will result if we let everyone publish, publishing is already a flood of crap, but science is a strong-link problem so the good stuff rises to the top”. I certainly don’t think we need stronger pre-publication review or any more barrier guardians (although I have reluctantly concluded that having some is useful). But when fraudulent stuff like this does in fact rise to the top in what seems to be a strong-link network — lots of NIH-funded labs, papers in top journals (or, apparently, “top” journals) — then I despair a bit. Science has gotten so specialized that almost anyone could invent facts or data within their subfield that might pass muster even with close colleagues (even if those colleagues aren’t on the take, he said cynically — there is a mind-boggling amount of money floating around in the drug-development world).

There is indeed. Lots more at the link, mostly about paleobiology, ending with these wise words:

So if you want to do good work — in this metaphor, to be at the top where the good science floats (eventually, alongside a seasoning of not-yet-unmasked bad science) — then I think you have to be aware that other cells exist, and occasionally peer into them, if for no other reason than to make sure you don’t accept an idea that’s already been debunked over there. And you need to read broadly and deeply in your own cell — there’s almost certainly valuable stuff you don’t know because the relevant works are stuck to the bottom of the pot. Go knock ’em loose.

I was peripherally aware that large language models have crossed a chasm in the last year or so, but I haven’t realized how large of a jump it was until I compared ChatGPT’s answer to my standard question: “How many lymphocytes are there in the human body?”.

Back in February of last year it took some effort to produce an over-inflated estimate. Today, I was served a well-reasoned and beautifuly formatted response after a single prompt. Sure, I have gotten better at writing prompts but the difference there is marginal. Not so marginal is the leap in usefulness and trustworthiness of the model, which went from being an overeager high school freshman to an all-star college senior.

And that is just the reasoning. Creating quick Word documents with tables and columns just the way I want them has become routine, even when/especially if I want to recreate a document from a badly scanned printout. My office document formatting skills are getting rusty and I couldn’t be happier for it.

In his Kefahuchi Tractt trilogy, M. John Harrison conjures up alien algorithms floating around the human environment, mostly helpful, sometimes not, motives unknown. Back in the early 2000s when the first novel came out I was wondering what on earth he was talking about but for better or worse we are now headed towards that world. Whether we are inching or hurling, that depends entirely on your point of view.

(↬Tyler Cowen)

For some Sunday pre-holiday week reading, here is a detailed analysis of what went wrong in Valencia from the Financial Times that shows both the human and technical side of the flooding there earlier this year. It is excellent throughout, and really got my blood boiling near the end with this series of paragraphs:

Cutting the risk of flash floods is not impossible. After the 1957 disaster, generalísimo Francisco Franco oversaw a vast engineering project to reroute the Turia river away from Valencia’s city centre. It is the reason why the capital was largely unscathed on October 29. But dictators do not have to consult stakeholders and such poured-concrete solutions are out of fashion today.

Still, Spain has not lacked modern proposals to stop the Poyo ravine flooding. But its slow-moving state has failed to implement them. The Júcar river basin authority put forward a risk reduction plan in 1994. Three of its four parts were blocked on environmental grounds, so it only stabilised the walls of the ravine from Paiporta to the coast — a job finished in 2005.

By then the basin authority had commissioned work on an alternative plan, which was authorised by the central government in 2009. It involved restoring forests to improve soil water absorption and building a “safety” channel to siphon water from the ravine to Franco’s rerouted river.

By the time it won environmental approval in 2011, Spain was heading into austerity. A new conservative government then shelved the plan. When the socialists returned to power in 2018, the environmental approval had expired. Pedro Sánchez’s government concluded a new plan was needed, but cost-benefit studies and new environmental demands at regional level threw up fresh obstacles. On the ground, nothing was done.

Valencia is a beautiful city as I saw for myself not long ago, and big part of it was the dry river bed-turned-park going straight down the center, orange groves and all. To think that what enabled it was a fascist dictator’s big project, when he probably didn’t care an iota about the park. And the people who care about the parks are clearly not capable of doing these large-scale projects. It’s the yin and yang of humanity.

A one-two punch on clinical trials from Ruxandra Teslo and Willy Chertman today: first their on-point agenda for clinical trial abundance as a guest post in Slow Boring, then Ruxandra’s longer essay which has been so thoroughly research that even yours truly gets a name-check. As I noted elsewhere, every US institution has made one bade tradeoff after another in how it conducts clinical trials to the point that it’s impossible to conduct a RECOVERY trial equivalent over here. That needs to change.

On my way back from #ASH24 I’ll go back through the abstract book and check out how many cell therapy oral presentations were given by investigators from China. This is the first meeting I’ve attended since 2019 and the difference is striking. Kudos!

PCA maps are the new PET scan, only with zero clinical relevancy instead of at least some. Much more subjective, too! #ASH24

100% of patients developed grade ≥3 neutropenia. “The safety profile was mild” #ASH24

Seeing those PET scans after CAR-T 5–10 years ago was transformative but it has now become superfluous. Yes, yes, that was a nice anecdote, can I now please see some data? #ASH24

Quote of the week is from Adam Mastroianni:

For example, the National Institutes of Health don’t like funding anything risky, so a good way to get money from them is to show them some “promising” and “preliminary” results from a project that, secretly, you’ve already completed. When they give you a grant, you can publish the rest of the results and go “wow look it all turned out so well!” when actually you’ve been using the money to try other stuff, hopefully generating “promising” and “preliminary” results for the next grant application. Which is to say, a big part of our scientific progress depends on defrauding the government.

The article is mostly about Paul Feyerabend, author of Against Method and self-proclaimed scientific anarchist. Recommended.

Some weekend links, old and new:

• A Case Against the Placebo Effect
• Frank Auerbach and the unexamined life
• The Worst Generation (from the year 2000!)
• Merlin Mann on The Great Discontent (2013!)