Commenting as a mathematician who sometimes uses the Bourbaki books as reference for research: this is probably not a good idea. These books have a very dry style and in particular rarely contain detailed examples, motivation for definitions, or intuition for theorems.
They are great for reference if you already know the subject, but if you want to learn something, you'd be better off using introductory textbooks for the various fields you're interested in. I'm happy to make recommendations, and you can also look up syllabi for math undergrad at Stanford, Princeton, etc.
As a professional mathematician, I see it the same way a musician views musical notation. Sure, writing the name of each note on the stave would be easier to read at first, but the seemingly complex notation then allows for a much quicker understanding (see how talented pianists can sight read very complicated scores; I doubt this could be possible with a more verbose notation). Similarly, I have no trouble parsing (correctly written) LaTeX mentally.
I feel that the author saying LaTeX is bad because it is (in particular) not based on a GUI/point-and-click system, and preferring Microsoft Equation Editor/Outline to it, reveals a big misunderstanding.
I also don't see how Mathematica's notation is any better; I view it as way worse.
This is exactly one of the points raised in the text. Non-TT assistant professorships are indeed postdocs, which often come with more teaching load than other fellowship-based postdocs. This is for example the case of the "Hills Assistant professorship" at Rutgers
- Gemma3 12B: ~100 t/s on prompt eval; 15 t/s on eval
- MistralSmall3 24B: ~500 t/s on prompt eval; 10 t/s on eval
Do you know what different in architecture could make the prompt eval (prefill) so much slower on the 2x smaller Gemma3 model?