I can’t comment that he will fail to review LLM code, but code churned by LLMs has different failure modes than human written code, so reviewing it is different. Notably LLMs easily use good terms and the code looks plausible, disregarding whether it is correct or wrong; while sloppy (no pun intended) code made by a person has often tells – i.e noticeable habits, wrong terms pointing out that they are new to the domain etc.
I think Linus Torvalds has rejected quite a few patches where a competent person implemented under protest the outcome of a corporate decision-making process. Which might not have low-level tells but might be absurd globally.
Corporations are often human-hostile alien-logic decision-making machines, just like LLMs. They are just slower.
procedurally these may be similar, but my point is that LLMs get to the goal differently than humans, so the review skills are not obviously transferable between different types of authors. corpo logic may be alien, but that applies to outcomes, not means of coding.
«You ship your orgchart» definitely includes corporate logic leaking into the code. And for drivers, lies a corporation tells about their product sometimes also become constraints on the code.
Interesting. I review (and reject) a lot of LLM code, much of it in Rust or other typed languages. Less than 10% of the problems I have found are caused by subtle bugs. At least 90% of problems could be described as “over-expedient architecture” or “badly factored architecture.” This may be because I use a long list of linters and other automated checkers, and I’m not shy about asking LLMs to write proptests.
Honestly, I like to imagine I’m careful. But the LLM probably finds as least as many of my bugs as I find of its. The biggest complaints I have all lie higher up.
(People who don’t read the code, who don’t design good test strategies, or who use agents swarms mostly get what they deserve.)
sure; my point is that reviewing llm code and human code is not the same thing, even when we put aside the sheer volume and velocity - so the masterclass review skills of Torvalds of human coda are not necessarily as good guardrail as the grandparent suggests.
edit: to add a metaphore - a top class swimmer is not necessarily a top class runner and vice versa
I can’t comment that he will fail to review LLM code, but code churned by LLMs has different failure modes than human written code, so reviewing it is different. Notably LLMs easily use good terms and the code looks plausible, disregarding whether it is correct or wrong; while sloppy (no pun intended) code made by a person has often tells – i.e noticeable habits, wrong terms pointing out that they are new to the domain etc.
score 1 · canonical
I think Linus Torvalds has rejected quite a few patches where a competent person implemented under protest the outcome of a corporate decision-making process. Which might not have low-level tells but might be absurd globally.
Corporations are often human-hostile alien-logic decision-making machines, just like LLMs. They are just slower.
score 2 · canonical
procedurally these may be similar, but my point is that LLMs get to the goal differently than humans, so the review skills are not obviously transferable between different types of authors. corpo logic may be alien, but that applies to outcomes, not means of coding.
score 1 · canonical
«You ship your orgchart» definitely includes corporate logic leaking into the code. And for drivers, lies a corporation tells about their product sometimes also become constraints on the code.
score 1 · canonical
Interesting. I review (and reject) a lot of LLM code, much of it in Rust or other typed languages. Less than 10% of the problems I have found are caused by subtle bugs. At least 90% of problems could be described as “over-expedient architecture” or “badly factored architecture.” This may be because I use a long list of linters and other automated checkers, and I’m not shy about asking LLMs to write proptests.
Honestly, I like to imagine I’m careful. But the LLM probably finds as least as many of my bugs as I find of its. The biggest complaints I have all lie higher up.
(People who don’t read the code, who don’t design good test strategies, or who use agents swarms mostly get what they deserve.)
score 1 · canonical
sure; my point is that reviewing llm code and human code is not the same thing, even when we put aside the sheer volume and velocity - so the masterclass review skills of Torvalds of human coda are not necessarily as good guardrail as the grandparent suggests.
edit: to add a metaphore - a top class swimmer is not necessarily a top class runner and vice versa
score 1 · canonical