为此，加州大学伯克利分校的研究团队提出了一种颠覆性的解决方案—— Prompt-to-Leaderboard （P2L）。 这项技术能够“见微知著”：仅凭一个用户输入的提示（如“帮我生成一段Python代码”或“写一篇悬疑小说”），就能实时生成该任务下的 专属模型排行榜 ，并 ...

To address this, we propose Prompt-to-Leaderboard (P2L), a method that produces leaderboards specific to a prompt or set of prompts. The core idea is to train an LLM taking natural language prompts as input to output a vector of Bradley-Terry coefficients which are then used to predict the human preference vote.

2025年2月22日 · 本文主要介绍了p2l的工作，值得细读，p2l提供了一个模型选型部分解决方案，可以使用大型语言模型自动对这些提示进行分类；在每个类别内生成一个偏好排行榜，针对每个模型进行分析。

To address this, we propose Prompt-to-Leaderboard (P2L), a method that produces leaderboards specific to a prompt. The core idea is to train an LLM taking natural language prompts as input to output a vector of Bradley-Terry coefficients which are then used to …

2025年2月20日 · To address this, we propose Prompt-to-Leaderboard (P2L), a method that produces leaderboards specific to a prompt. The core idea is to train an LLM taking natural language prompts as input to output a vector of Bradley-Terry coefficients which are then used to predict the human preference vote.

这就是竞技场（lmarena.ai）最新推出的排名方式，叫做 Prompt-to-leaderboard（P2L）。 主打的就是找到最能直击你Prompt“灵魂”的那款大模型。 话不多说，我们来看下效果。 例如给一个算数的Prompt： 在竞技场的P2L排行榜中，针对这道算数Prompt，得分最高的模型就是o3-mini-high了。 再来一个： Be...

2025年2月20日 · 为了解决这个问题，我们提出了 Prompt-to-Leaderboard （P2L），这是一种生成特定于提示的排行榜的方法。 核心思想是训练一个LLM以自然语言提示作为输入来输出 Bradley-Terry 系数向量，然后用于预测人类偏好投票。

There are 27 games in six categories (The normal distribution, Correlations, Linear regression, Z test, Inferential statistics, and Effect size). Site: P2L.io. Alternative Menu: View list of all games, sorted by category. Proportions: From the Middle.

2019年8月20日 · We use this measure, which we call "Predict To Learn" ("P2L"), in the two very different domains of images and semantic relations, where it predicts, from a set of "source" models, the one model most likely to produce effective transfer for training a given "target" model.

We use this method, "Predict To Learn" (P2L), to predict the most likely "source" dataset to produce effective transfer for training on a "target" dataset. We validate our approach extensively across 21 tasks, including image classification tasks and semantic relationship prediction tasks in the linguistic domain.