Regarding obs: what about atypical archive of obs such as the vertically pointing MST radar near Aberystwyth?

Regarding reanalyses: they’re fantastic for setting baselines when evaluating forecasts. Also case studies of old events with as close to “like for like” as possible.

Yeah! It’s a function of practical tests’ trustworthiness. I can’t vibe-code a prediction model and pray it doesn’t have a snowfall bias. But then our new website is entirely supervised “coding with natural language” allowing high school students to be trained, contribute, learn. Mantra: caution! 😥

So we aren’t using the same word then. Inference is the act of proposing the next symbol based on current state(s). LMs in general don’t understand what you mean on Google Translate either. It can still be correct. That’s inference too.

Cited two of those papers recently after deliberately writing with a LLM sceptic on LLMs, haha, and he was right there with his Bullshit papers.

Nope, my first big use of LLMs was fixing a six year math problem I couldn’t work out. It’s Shannon at the core: information transfer. I don’t care what it sounds like and neither does info entropy quantification.

Truth is a pattern, a sequence, that can be true by accident. I don’t think truth must be intentional. Maybe that’s way off into philosophy at this point

We definitely have different definitions of inference…

(I did a first round and it hit me I haven’t seen an email about R2. Oh no…)

Lovely plot, nice work.

You’ve posted this a lot I see. How do you know?

Moving! Extremely moving.