我也支持这个选择。一定要买第二版，并且一定要读一读高德纳（Knuth）的《计算机程序设计艺术》（Art of Computer Programming）第 4 卷的各个分册（fascicles）。第 2 卷里也有不少干货。
I second this choice. Be sure to get the second edition and to read the various fascicles for Knuth's Volume 4 of Art of Computer Programming. Volume 2 has some useful content as well.
…quite often we find that some old method that is comparatively unsatisfactory has blindly been passed down from one programmer to another, and today’s users have no understanding of its limitations. — The Art of Computer Programming, Volume 2, by Donald Knuth
…improvement of Knuth's earlier "reservoir sampling" from the first edition). > All in all, Algorithm R was known to Knuth and Waterman by 1975, and to a wider audience by 1981, when the second edition of The Art of Computer Programming volume 2 was published.
BTW that one isn't a joke made up for this site: it's a verbatim quote from The Art of Computer Programming (Volume 2, page 2). TAOCP is full of silly-but-serious jokes like that.
…Can't they run an example prompt and verify they get the exact same output token probabilities for all prompts? You don’t even get that with GPUs in general, or really floating point in general. The Art of Computer Programming. Volume 2: Seminumerical Algorithms section 4.2.2 with explain where it loses floating addition associativity property. Apartness relations are another possible lens.
我也支持这个选择。一定要买第二版，并且一定要读一读高德纳（Knuth）的《计算机程序设计艺术》（Art of Computer Programming）第 4 卷的各个分册（fascicles）。第 2 卷里也有不少干货。
I second this choice. Be sure to get the second edition and to read the various fascicles for Knuth's Volume 4 of Art of Computer Programming. Volume 2 has some useful content as well.
……我们常常发现，某种相对来说不尽如人意的旧方法被盲目地在程序员之间代代相传，而如今的使用者对其局限性一无所知。——《计算机程序设计艺术》第 2 卷，唐纳德·高德纳（Donald Knuth）
…quite often we find that some old method that is comparatively unsatisfactory has blindly been passed down from one programmer to another, and today’s users have no understanding of its limitations. — The Art of Computer Programming, Volume 2, by Donald Knuth
……对高德纳在第一版中提出的“水塘抽样”（reservoir sampling）的改进）。> 总而言之，算法 R 早在 1975 年就已被高德纳和沃特曼知晓，而到了 1981 年《计算机程序设计艺术》第 2 卷第二版出版时，它才被更广泛的受众所了解。
…improvement of Knuth's earlier "reservoir sampling" from the first edition). > All in all, Algorithm R was known to Knuth and Waterman by 1975, and to a wider audience by 1981, when the second edition of The Art of Computer Programming volume 2 was published.
顺便说一句，那可不是为了这个网站现编的段子：那是《计算机程序设计艺术》（第 2 卷，第 2 页）里一字不差的原话。TAOCP 里到处都是这种看似一本正经胡说八道的幽默。
BTW that one isn't a joke made up for this site: it's a verbatim quote from The Art of Computer Programming (Volume 2, page 2). TAOCP is full of silly-but-serious jokes like that.
……难道他们就不能跑一个示例 prompt，验证一下所有 prompt 是不是都能得到完全相同的输出 token 概率吗？在 GPU 上通常根本做不到这一点，或者说广义上的浮点数计算本来就做不到。《计算机程序设计艺术：第 2 卷 半数值算法》第 4.2.2 节解释了浮点数加法结合律是在哪里失效的。分离关系（Apartness relations）是另一个可行的切入视角。
…Can't they run an example prompt and verify they get the exact same output token probabilities for all prompts? You don’t even get that with GPUs in general, or really floating point in general. The Art of Computer Programming. Volume 2: Seminumerical Algorithms section 4.2.2 with explain where it loses floating addition associativity property. Apartness relations are another possible lens.