Ben Hall

30% price rise, and could not buy more than one at a time (Amazon) or only preorder (other retailers)

Gosh, hard drives too. Been struggling to replace my nas because of availability

Really interesting post on AI for data collation. I’ve had a different experience with patient data from a public cancer database- better & easier than manual.

The difference is instructive- this is a real research task, whereas mine was technical. This is Claude’s “intelligence”, not thinking

New post, on whether I could get Claude Code to complete a data task that had taken me AGES a decade ago…

kucharski.substack.com/p/how-much-t...

Britain's AI bear case A handful of troublingly plausible ways the next decade could not go Britain's way

First, AI kills the lawyers.

Thoughtful, analytical and depressing piece from @archiehall.bsky.social.

notes.archie-hall.com/p/britains-a...

Looks fantastic. Never heard of the Spanish black pudding before- what’s it like?

Frontiers | AI needs physics more than physics needs AI Artificial intelligence (AI) is commonly depicted as transformative. Yet, after more than a decade of hype, its measurable impact remains modest outside a fe...

Oops, sorry, link is here www.frontiersin.org/journals/phy...

Screenshot of the "Does that use a lot of energy?" online app

Hannah Ritchie has built a fun little tool where you can compare energy usage of various products and activities.

This is super helpful imho, because it's so hard to develop intuitions even just about the scales involved here.

hannahritchie.substack.com/p/does-that-...

The frontiers link is broken

Being fascinated by AI's shortcomings (see, eg www.frontiersin.org/journals/phy....

..interesting to see that deep learning systems trained for cancer pathology may be relying on hidden shortcuts rather than genuine biological signals www.eurekalert.org/news-release...

Just back from a conference where a radiologist said that some of the pronouncements of “time saved” came from artefacts of testing in optimal conditions rather than realistic ones. They had seen no reductions in load but were aiming for 20% in 5 years

Economists: you know I love you, but I think it's time you gave it a rest with all these "which jobs are most exposed to AI?" analyses. www.ft.com/content/8d3b...

Poor show from Lego. Scotland deserves its own themed sets, not to just have London themed ones for sale

I swear to god, these people think that universities are experiencing an academic essay shortage, rather than academic essays being a way to evaluate student skills and learning.

Opinion | The Rise of the Bratty Machines

'In this case, a single A.I. agent endeavored to ruin the reputation of a volunteer code librarian and could have done considerably more harm.
“It was like an angry toddler throwing a tantrum,” Shambaugh told me, “except the angry toddler has full command of the English language.”

Science over medicine because I thought scientists had the potential to help more people, though a higher potential to help few. I was happier with the chance of greater impact

Younger than 16, wanted to study biochemistry for a life in science. Prompted by need to choose post 16 qualifications, which sounds quite normal, but had to make choices about what not to take (my natural abilities are mathematical, and I was limited in choices after biology was taken).

Nature Medicine to investigate study that found cancer treatment is better in morning Nature Medicine starts an investigation into inconsistencies in a study that found it was better to have immunotherapy in the morning.

www.statnews.com/2026/02/20/c...
i hope they can get the data from the authors. I was told by the editor that data weren’t available because of patient confidentiality- so many authors use that excuse , it makes data sharing policies worthless

i co-teach a research methods in ecology & evolution class and my colleague was lecturing and said "chlorophyll" and immediately my brain went "chlorophyll??? more like borophyll!!!"

1/5) Re-reading this excellent paper on using mutation-selection balance models to quantify positive selection. The discussed lack of congruence between MK approaches and classic dN/dS test is entirely unsurprising, see thread below.
www.pnas.org/doi/10.1073/...

I’m doing some of my coding on tight/non existent budgets so it’s enabled work that would not have been done. But all my outputs are ultimately some degree of prototype and that’s not a good enough output for many scenarios. I expect that rather than eliminating work we will see new work enabled

I think a big missing factor is *what work* might be replaced. @stephenkb.bsky.social had an excellent article about the uncomfortable fact that responsibility isn’t going away, and checking and managing automated coding is still going to be necessary (and a harder skill).

Metabolic control of innate immune activation in TET2-mutant clonal hematopoiesis Kim and Hergott et al. uncover a metabolic foundation for the pathogenic inflammation arising from TET2-mutant clonal hematopoiesis (CH). In myeloid cells, TET2 loss disinhibits the nutrient sensor/glycosyltransferase OGT, driving aberrant and inflammatory lipid accumulation via the metabolic enzyme ACLY. Inhibiting ACLY offers a therapeutic strategy for controlling CH-related hyperinflammation.

Metabolic control of innate immune activation in TET2-mutant clonal hematopoiesis

Is AI making work more intense? Using agents appears to increase the number of hours worked and the exhausting nature of tasks

Really interesting back and forth by @jburnmurdoch.ft.com and @sarahoconnorft.ft.com in their newsletter today:

Our preprint is now published in MSB. link.springer.com/article/10.1...
We decompose multi-omics into distinct phenotypic axes (drug response vs ARID1A-driven cell state), improving interpretation and revealing how baseline cell state rewires signaling and shapes MAPK inhibitor resistance.

Honestly, I didn’t have much more than that for turning python modules into a pip package

🌻 claude code psychosis are your problems software-shaped?

Great stuff. I find I'm on a slight post-hype downswing at the moment, that realisation that most of my problems aren't "software-shaped" and that my actual productivity has gone down, as I frantically switch between agents, terminal, mcps, etc. This captured a lot of it jasmi.news/p/claude-code

This captures something I’ve found too recently- a threshold has been passed and LLM are now remarkably capable. 2 long delayed projects completed with LLM support, one using fully automated approaches.

Scientifically they’re still poor at *concepts* but there’s a lot of opportunities elsewhere.

How I've used AI to improve my work My adventures in vibe coding

How I've used AI to improve my work <-- I've written about my adventures in vibe coding!

takes.jamesomalley.co.uk/p/how-ive-us...