Ben Hall

AI tools are immensely powerful. The use of AI as a general purpose chat bot that is used like an encyclopaedia is the worst possible use of AI. AI is now, given good instructions from a knowledgeable user, capable of doing absolutely vast amounts of work extremely fast.

Miller & Wankowicz (2026) Binding Entropy Can Be Predicted by Crystallographic Ensembles. #bioRxiv #preprint #CompChem #PhysChem #biophys #cheminformatics #chemsky 🧪
doi.org/10.64898/202...

I think this is relatively common, although there’s some assessment on application

UKRI pauses several funding calls amid priorities shake-up - Research Professional News “Period of transition” at funding agency sees some Medical Research Council funding opportunities on hiatus

This is all I could find

www.researchprofessionalnews.com/rr-news-uk-r....

👀

There’s been quite a lot of nothing good recently tbf

What’s happening?

Airport Book Brain How faddish ideas keep seducing.

oh, this is really good on what @joxley.jmoxley.co.uk calls "airport book brain", where glib "solutions" are jumped on by politicians who want smooth narratives, discarding nuance and complexity. Complete with a nice pop at Jonathan Haidt, and at the "UPF" brigade, too. (ht @rorycj.bsky.social)

Oh! Is that just for one room, or multiple?

Been considering it for one or two rooms, but I believed you needed permission if it served more than one room.

What was the rule change?

Finally, the evidence I needed to show my kids that, "Yes, you did this to me!"

Simon van gelder played by Morgan Woodward. He looks older than 43, I tell myself

43 in Star Trek episode dagger of the mind

Honestly sometimes it’s like meeting a dog that talks total nonsense but in perfect English. It’s super impressive that it can do it at all, but it’s really not useful or helpful

Second preprint of the year in which @sarahgersing.bsky.social from @rhp-lab.bsky.social mapped the effects of >7500 variants in glucokinase (GCK) on the interaction with the glucokinase regulatory protein so that we now have a 3D GCK scan (abundance, interaction, activity)

doi.org/10.64898/202...

“Tell me you were trained on a bunch of stack overflow RTFM comments without telling me you were trained on stack overflow”

I did have some entertaining failures though. A particular high point was an increasingly aggressive set of repeated responses from chatgpt when its solution to a bug was incorrect

My deep suspicion is that what it’s doing is taking existing known solutions (and code) and fitting them to my spec- that is, it’s not inventive, it’s adapting in the way I would have to do from examples, stack overflow etc

The caveat is that I know exactly what I want and can describe and test the high level specifications from whatever is generated. I think it would have failed if I didn’t have a clear view, or if I couldn’t fix the issues manually myself.

My experience in the last 6 months has been
- the code is much more easy to read and works well enough for me to debug functionality to get it there
- for ugly to use functions (matplotlib) it’s really helpful in finding the combination of options I want
- its good at helping with cryptic errors

a cartoon character is pointing at a chart that says disco record sales Alt: a cartoon character is pointing at a chart that says disco record sales, with unrealistic exponential growth

I’ve had some eye opening experiences with coding assisted by LLMs recently, and I wonder if at least some of the froth is down to the genuine success for programmers as consumers, and then incorrectly extrapolating to the rest of the world

I don’t think that’s fair. He wasn’t deeply technical but he had a unique understanding of both worlds.

If it was just marketing that determined his success at Apple, that level of success would be commonplace

Funny, but also one of the two best career pieces of advice I got -- get exercise, especially when you are too stressed or don't have time for it.

It’s not lost on me that it’s spending tokens like a sailor when it does this

I’ve found Claude has a propensity to be verbose, including going back to additionally and unnecessarily integrate new requests into older things that don’t need it

“Breakfast. Cups.”

Artificial intelligence tools expand scientists’ impact but contract science’s focus - Nature Artificial intelligence boosts individual scientists’ output, citations and career progression, but collectively narrows research diversity and reduces collaboration, concentrating work in data-rich a...

One of my concerns with AI-assisted research in biology is that we're already looking only in the lamplight and AI incentivizes us to narrow the beam. Seems like it's happening...
www.nature.com/articles/s41...

I prefer this explanation to it being nostalgic for my favourite trackballs!

I don’t know/understand Logitechs business reasons, but I bought some switches this Xmas to revive my favourite of their trackballs, no longer for sale. Amazed how attached I (and others) to what are essentially cheap peripherals

'Another Warrington has hit the North-West'.

I’ve reposted this twice now, it’s great, and it’s getting referenced in my updated lecture covering the subject (alongside the 1979 IBM slide)