news to literally anyone on a mobile internet connection (which is most internet connections!)

Let's stipulate that what pretraining does is train next token prediction over a gigantic corpus. You can then sample from this distribution repeatedly (cf the Large Language Monkeys paper) and count how often it passes some deterministic verifier.

What GRPO-style RLVR does is precisely this, but then reward the trajectories which passed the verifier. These distributions are _by construction_ within the accessible output space of the pretrained model; you're reweighting the distribution so that pass@k goes up, because that's (for applications like programming) very useful. RLVR is about making sampling more efficient; the only new information being added to the system is the presence of the verifier, and note that you only get a reward when the verifier passes, so there's essentially no mechanism for "teaching new facts" here.

Take the NHS; it will have to deal with ten plus separate unions - https://nhsunions.org/#about – of which the biggest powers are the British Medical Association and the Royal College of Nursing, but the cleaners are GMB or Unite and they're huge pan-working-class institutions.

Education has to deal with the NEU, the NASUWT, and the NAHT, each of which has a different political slant. Some unions in the UK have been historically rather centrist in their politics (a good example of that is Prospect, https://prospect.org.uk/about/, which is a roll-up of a number of scientific and finance unions), some are firebreathing communists, but all of them work across employers.

There's also no such thing as a closed shop in the UK – because there are much stronger worker protections, there's less of a need for one.

(I was, at one time, in a majority-UCU workplace; https://www.ucu.org.uk.)

you literally can, it's the entire point of eg management buyouts

In pop music this has been true since the 60s. For independent music it has mostly never been true. This hasn't changed much.