@rapha_gl we probably shouldn't even talk about o3-mini-cross-entropy-faded

2024 squared away

@karthikkalyan90 @martin_casado you might like the new error messages i added for this situation to python 3.13 :-)

@sama "there is no wall?" "then you'll see that it is not the scaling law that bends, it is only yourself"

nothing like running into an actual bug in git to make you doubt your own sanity

@charliermarsh g then n is the keyboard shortcut

tfw you find yourself googling dead snakes action

bb(5) = 47,176,870

@miru_why @ezyang and that's the minimised repro :-)

@nearcyan hm that number is higher than i expected. i checked and it looks like my openai commits are counted twice, for reasons. so maybe divide by two :-)

@HenrySchreiner3 yeah, it's great. in typing_extensions, we run the tests of a few downstream packages. previously these would break all the time. now uv lets us just use pypi as of the date of the commit whose tests we're running

@charliermarsh when i got one jennifer explicitly said she was asking in an informal sense, so felt more self-audit-y and less regulation-y

@FelledMan @ryxcommar Python didn’t have bool until v2.2.1, only int. Prior boolean logic just used aliased ints. This is also why bool is a subclass of int

@llanga @mitsuhiko Recent versions of mypy should complain about the dumb case of using method in conditional. In addition, you can also turn on —enable-error-code truthy-bool to get complaints in several more cases (with risk of some false positives)

@srchvrs This might be a fact, but it's not fun — it's quadratic, and you will be sad if you use it in real world code

@mitsuhiko It’s not wrong. In 3.11 the optimisation no longer happens for globals:

@mycoliza that’s N.Y.T. to you

��