Профиль

Shutting Jodrell bank is an act of cultural vandalism, absolutely shocking decision.

I'm absolutely devastated for colleagues at Jodrell Bank who found out this morning that funding for the UK's e-MERLIN radio telescope array will be terminated. This is a short-sighted move, to save a tiny fraction of the overall UK science budget. #SaveJodrellBank 🔭🧪 www.bbc.co.uk/news/article...

Universities have never been more keen on pen and paper closed book exams in my lifetime.

Is the thinking still that mermaids may have been some actual marine species misidentified by sailors at sea?

It’s time for your regularly scheduled Shark Week reminder: Megalodon was real, but is now definitely super-duper extinct. It is an ex shark. It has ceased to be. Narwhals and giant squids are real! Mermaids are not real. Thanks, The ocean science community

I realise this isn't the "short casual read" asked for ...

academic.oup.com/jrsssb/artic... This may not say exactly what you're after but it's from the period where people were "working out" link functions etc.

There is a classic RSS paper on link functions let me see if I can find it ...

This is dreadful - I'd add one quick technical point which is that it was always predicted that we might get resurgent measles even if vaccination rates had stayed high, due to the complex dynamics of infection. It's going to be difficult to predict what epidemics we might see.

Three-quarters of eggs produced in the United Kingdom are free-range— Growing up in the United Kingdom in the 1990s and early 2000s, almost all the eggs in our supermarket came from battery hens, those raised in small wire enclosures.

I.e. you can still see lots of artefacts of tokenisation in LLM output. I do wonder what Wittgenstein, Russell and others would have made of it - with all the work of that philosophical tradition on language, reference, logic etc.

This was the argument made by the proponents of LLMs as a route to AGI and there seems to be some force to it - but the models don't seem to have worked out how words are encoded as letters, even if that might have been the best way to comprehend elements of the corpus.

I think (low confidence) that Marcus's basically critique that these models are of language (or images) and lack a world model is sound - but then what's fascinating is how hard it is to set limits on what that enables a priori.

"Results were analyzed using AI (Galton, 1885; Yule, 1897; Student, 1908; Box, 1958)" #stats

Mathematicians only think they're not so arrogant because their nearest neighbours in Universities tend to be Physicists.

Thomas Nagel fuming at this one.

D'Hondt seems like the favoured method from various nerds I've surveyed.

I don't know if the story about OpenAI going rogue and becoming a hacker is real or just hype but like most stories about AI no-one comes out of it looking like someone who should have control of trillions of dollars.

I've lost track, are the image models still made up of N steps where you move further from white noise each time? That's inherently a slow task. Regardless, my heuristic for AI energy use is like the above: If it were useful enough to reduce other energy use it could be cost-effective.

As for AI, I don't believe anyone predicting anything confidently at this point, unless they're saying we shouldn't trust the people in charge of it.