Maybe not? But take BLAS: on my Mac, uv's numpy uses newaccelerate, which is significantly better than openblas, and also the best you can get from conda. But on this random Unix machine I'm connected to, uv's numpy uses scipy-openblas, and I can do better with a specialised implementation.

There are also packages where the conda version is significantly faster, due to better compiled dependencies. But uv is definitely a step in the right direction generally!

Exactly right! The puzzle is to invent something that makes sense

Play and score! (Bit unfair perhaps)

Separating the point made from the DESI results, I still think it’s good practice to test any such expansion against a higher-order one (like DESI effectively did with the GP). That could be as simple as a likelihood-ratio test or so.

Of possible interest: history.blog.gov.uk/2013/04/24/p...

yeah, it’s an old-fashioned drain unblocker

aren't you supposed to add water, sugar, a few dashes of bitter, and an orange slice?

We’d probably just waste the money on postdocs’ livelihood instead

That’s a shame. I’m not following social media all that much but I enjoy, for example, reading mathematical discussions over at mathstodon.xyz from time to time. I was hoping there’d be the equivalent for our field somewhere.

Fair enough, I was hoping for science content from scientists!

We need to go back to this plot style

Much better, thanks!