研究人员开发了一种新颖的方法，通过在测试时优化提示嵌入来增强AI模型中的上下文学习（ICL）。该技术利用模型自身演示输出的对数概率作为一种自监督置信度代理。通过最大化此代理，系统无需微调或外部数据即可进行校准，在各种ICL任务中表现出一致或改进的性能。 AI
影响 该方法提供了一种在不要求额外训练数据或微调的情况下增强AI模型在各种任务上性能的方法。
排序理由 该集群包含一篇详细介绍一种新方法以提高AI模型性能的学术论文。
AI 生成摘要 · Google Gemini · 来自 5 个来源。 我们如何撰写摘要 →
arXiv:2605.26827v1 Announce Type: cross Abstract: Recent benchmarks reveal that despite strong reasoning capabilities, large language models (LLMs) still struggle to faithfully apply complex contextual knowledge. These failures are often not wholesale reasoning collapses: in cont…
Recent benchmarks reveal that despite strong reasoning capabilities, large language models (LLMs) still struggle to faithfully apply complex contextual knowledge. These failures are often not wholesale reasoning collapses: in context-rich tasks, models may follow the central reas…
Recent benchmarks reveal that despite strong reasoning capabilities, large language models (LLMs) still struggle to faithfully apply complex contextual knowledge. These failures are often not wholesale reasoning collapses: in context-rich tasks, models may follow the central reas…
arXiv:2605.23180v1 Announce Type: new Abstract: We propose to improve in-context learning (ICL) by optimizing the continuous embeddings of a fixed few-shot prompt at test time. The key observation is that the log-probabilities a model assigns to its demonstrated outputs$\unicode{…
We propose to improve in-context learning (ICL) by optimizing the continuous embeddings of a fixed few-shot prompt at test time. The key observation is that the log-probabilities a model assigns to its demonstrated outputs$\unicode{x2013}$available from a single forward pass with…