Perplexity 详细介绍了其专有的后训练流程，该流程可增强基础模型以进行搜索增强的问答。此过程包括用于指令遵循和安全性的初始微调，然后进行策略内强化学习以提高搜索准确性和效率。该公司的奖励设计优先考虑正确性和用户偏好，防止模型生成看似合理但不正确的响应。Perplexity 声称，当此方法应用于阿里巴巴的 Qwen 模型时，其事实准确性可与 GPT 模型相媲美或更优，同时成本更低。 AI
影响 Perplexity 的研究详细介绍了一个流程，可提高模型在搜索增强答案方面的准确性和效率，从而可能降低运营成本。
排序理由 Perplexity 发布了关于其模型后训练流程的新研究。
AI 生成摘要 · Google Gemini · 来自 5 个来源。 我们如何撰写摘要 →
This pipeline is why the same base model produces more accurate, better-cited, and more efficient answers inside Perplexity than out of the box. Read our research: https://t.co/pYjUTnkPMW
Our reward design combines correctness, preference, and efficiency. Preference only counts when the answer is correct. This keeps the model from optimizing for better-sounding wrong answers. https://t.co/VbJ1M4o26w
@Alibaba_Qwen We first fine-tune the model to follow instructions, stay within guardrails, and keep language consistent. Then we run on‑policy RL to improve search accuracy and tool efficiency while preserving those behaviors. https://t.co/KaVs7h5Ixa
We've published new research on how we post-train models for accurate search-augmented answers. Our SFT + RL pipeline improves search, citation quality, instruction following, and efficiency. With Qwen models, we match or beat GPT models on factuality at a lower cost. https://t…
RT Perplexity<br />We've published new research on how we post-train models for accurate search-augmented answers.<br /><br />Our SFT + RL pipeline improves search, citation quality, instruction following, and efficiency.<br /><br />With Qwen models, we match or beat GPT models o…