但要真正实现这一切，却并不容易。刘强东之前，已先后有多位企业家以不同方式入局其中，然而进展与成就却可以说是乏善可陈。
The tradeoff is complexity. The microcode must be carefully arranged so that the instructions in delay slots are either useful setup for both paths, or at least harmless if the redirect fires. Not every case is as clean as RETF. When a PLA redirect interrupts an LCALL, the return address is already pushed onto the microcode call stack (yes, the 386 has a microcode call stack) -- the redirected code must account for this stale entry. When multiple protection tests overlap, or when a redirect fires during a delay slot of another jump, the control flow becomes hard to reason about. During the FPGA core implementation, protection delay slot interactions were consistently the most difficult bugs to track down.
。51吃瓜对此有专业解读
2024年12月25日 星期三 新京报
let maxVal = -Infinity;
Returning back to the Anthropic compiler attempt: one of the steps that the agent failed was the one that was more strongly related to the idea of memorization of what is in the pretraining set: the assembler. With extensive documentation, I can’t see any way Claude Code (and, even more, GPT5.3-codex, which is in my experience, for complex stuff, more capable) could fail at producing a working assembler, since it is quite a mechanical process. This is, I think, in contradiction with the idea that LLMs are memorizing the whole training set and uncompress what they have seen. LLMs can memorize certain over-represented documents and code, but while they can extract such verbatim parts of the code if prompted to do so, they don’t have a copy of everything they saw during the training set, nor they spontaneously emit copies of already seen code, in their normal operation. We mostly ask LLMs to create work that requires assembling different knowledge they possess, and the result is normally something that uses known techniques and patterns, but that is new code, not constituting a copy of some pre-existing code.