HackerNews中文版

大型语言模型（LLMs）使得创建拉取请求（PR）的成本低于审查的成本（即多对一的关系）。作为一个代码库维护者，我通常会运行我的提示和工作流程来审查大量提交的代码，因此我不得不为每一个未请求的拉取请求付费。我在想：为什么大型语言模型的供应商不支持由贡献者资助的短期、针对拉取请求的使用密钥呢？就像一种审查保证金。例如：仅供维护者使用，仅限于这个代码库/拉取请求，金额或代币数量上限设置得很低，并且有效期短。这样可以足够支持维护者的审查工作流程，而不会使审查成本完全偏向一方。为什么没有任何主要供应商提供这种计费方式？附注：这个想法受到了一条讨论的启发（https://news.ycombinator.com/item?id=48497609），讨论内容是关于请求人类关注和展示人类努力的，类似于一种“工作证明”，这个案例可以被归类为“关注保证金”。

查看原文

LLMs make PRs cheaper to create than to review (many to one relationship).In my case, as a repo maintainer, I usually run my prompts and workflow to review a lot of the code that comes in, and I end up paying for every unsolicited PR.I was wondering: why don't LLM suppliers support ephemeral, PR-scoped usage keys funded by the contributor? Like a review bond.For example: usable only by the maintainer, only for this repo/PR, hard capped at a low $ amount or N tokens, with a short expiry. Enough to run the maintainer's review workflow without making review cost entirely one sided.Why isn't this billing type offered by any major player?ps: inspired by the thread (https://news.ycombinator.com/item?id=48497609) about asking for human attention and demonstrating human effort, a kind of "proof of work", this case could be classified as an "attention bond".

大型语言模型供应商应提供与公关相关的临时密钥。