Links from February 2025: plant memory, AI designed plants, AI scribes, procurement, Rust traits
Plant memory, AI designed plants, AI assisted medicine and scribes, medtech procurement, Rust standard traits
The Robot Doctor Will See You Now
NY Times
“[…] half the scans were assessed by two radiologists, as is usual. The other half were evaluated by A.I.-supported screening first, followed by additional review by one radiologist […] The A.I.-assisted approach led to the identification of 20 percent more breast cancers while reducing the overall radiologist workload almost in half […] This might be the clearest path to dealing with the shortage of health care workers hurting medicine.”
Are we adopting AI scribes based on… vibes?
Stat
A commentary on a JAMA paper: AI scribes save around 15 minutes of “after hours” time a day, but with wildly different clinician experience. Some trust it, some need to review and edit a lot. “The only ways to reconcile these two different experiences are: 1) The clinicians have different definitions of what’s acceptably accurate, 2) The AI tool is performing differently for different people, or 3) Some clinicians are checking the output carefully and some are not. To me, either of those latter two (very likely) explanations are alarming and show the need for studies looking at the accuracy of the output of AI scribes, not just whether physicians and patients are happier using the tool.”
How plants are able to remember stress without a brain
The Conversation
Just fascinating: epigenetic modifications keep plants primed so they can quickly respond to stressors; and using the soil as an external memory. Part of The Conversation’s “Plant Curious” series.
Heritable Agriculture, a Google spinout, is bringing AI to crop breeding
The Economist
Computational plant breeding: “Once the desired genetics for a given environment have been determined, a different model determines the quickest breeding path to take to reach those genetics, based on the plants available to a given breeder. […] Heritable’s co-founder, says the firm’s system can breed a crop with the right genetics to achieve a desired trait in just one year.”
How evolution builds genes from scratch
Nature
“Conventional wisdom was that new genes tended to arise when existing ones are accidentally duplicated, blended with others or broken up, but some researchers now think that de novo genes [from non-coding regions] could be quite common: some studies suggest at least one-tenth of genes could be made in this way; others estimate that more genes could emerge de novo than from gene duplication.”
The Scientific Fight Over Whether Aging Is a Disease
WSJ
“He’d prefer to think of aging as a risk factor for diseases. That’s similar to how we think of high cholesterol, he notes, which isn’t typically considered a disease in and of itself. Instead, it’s viewed as a risk factor for heart disease and can be treated with statins.”
Scientists flock to DeepSeek: how they’re using the blockbuster AI model
Nature
In summary: Low cost and freely available weights has lead to more LLM use and fine tuning for research projects.
Run LLMs on macOS using llm-mlx and Apple’s MLX framework
Simon Willison
Works for me.
Tour of Rust's Standard Library Traits
“pretzelhammer” (if that is their real name)
Much of this may be covered elsewhere, but it’s great to see a single long page giving a comprehensive guide to the Rust traits system.
Guidance to simplify medtech procurement to launch in 2025
Digital Health
The UK may move away from a pure cost saving assessment to one that takes other factors into account: “Value-based procurement (VBP) is a procurement approach that delivers a reduction in the whole life costs of healthcare where value can be created from financial, efficiency, patient, and environmental benefits.”
Accelerating scientific breakthroughs with an AI co-scientist
Google
Google feels like a good starting point for this: big internet search index, including Google Scholar.