TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Ask HN: Why does OpenAI intentionally keep its randomness/seed value a secret?

1 点作者 kappuchino大约 2 年前
So any LLM works with an input&#x2F;prompt, parameters (like the so-called temperature) and randomness aka seed for its &quot;magic&quot; decision making based on a modell. Think of the randomness like the magic number for minecraft maps: It can create repeatably the same world just from one number and the same version. Or the seed number in stable diffusion: Same prompt, same seed, same modell = always the same result.<p>But not for GPT3, ChatGPT and GPT4. I wonder why.<p>I&#x27;m quite mad that this random seed is not presented to be preserved or available as an input. It would speed up scientific exploration, produce differen&#x2F;better fidelity tests.

暂无评论

暂无评论