I also have used llm assisted coding and I'm not sure I agree. superficially it seems very fast but then coding using modern languages and libraries (I use Rust and ofc cargo) is also extremely fast these days. when I use LLMs I often have to spend significant time correcting errors made by the LLM and I estimate the time gain to be negative in these circumstances.
admittedly in certain cases , for example building boiler plate to get database rows into structures, LLM coding is spectacularly faster than my manual effors because I struggle at repetitive detail work.
As with all statistics
Google's AI research is done in London, for example. Just because the companies are listed on the US stock exchange and their headquarters are in the US, doesn't in any way show where the actual work is happening.
The UK suffers from a cultural snobbery towards technology that tends to direct UK capital away from such investments, it does not suffer from a shortage of high quality research (both commercial and private), funded from abroad, or for that matter talented individuals.
: is not an identifier