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。

Absolute Zero: Reinforced Self-Play Reasoning with Zero Data

7 点作者 dave1010uk8 天前

2 条评论

pk-protect-ai5 天前
Wow, so empty here, and it was 3 days ago... I wonder why?<p>I have a question if someone can answer. On github they are stating 8x80GB for 14B models but I found no information on how long this fine-tuning takes?<p>Given the toolchain, it probably takes significant time.<p>Another question wouldn&#x27;t it be fun to hijack the training loop with some tasks set by humans? Would it improve results or opposite?<p>I wonder if at some point all tasks will degrade to the &quot;uh-oh moment&quot; tasks, which will be most complex and perplexing with no actual productive yield?
dave1010uk8 天前
Paper: <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2505.03335" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2505.03335</a><p>Code: <a href="https:&#x2F;&#x2F;github.com&#x2F;LeapLabTHU&#x2F;Absolute-Zero-Reasoner">https:&#x2F;&#x2F;github.com&#x2F;LeapLabTHU&#x2F;Absolute-Zero-Reasoner</a><p>Thread: <a href="https:&#x2F;&#x2F;xcancel.com&#x2F;AndrewZ45732491&#x2F;status&#x2F;1919920459748909288" rel="nofollow">https:&#x2F;&#x2F;xcancel.com&#x2F;AndrewZ45732491&#x2F;status&#x2F;19199204597489092...</a>