大模型
x.com ·
@hwchase17
GPT Researcher 用 Jev 替换 embeddings：相关上下文 73% vs 46%
Harrison Chase 转述 Assaf Elovic 的实验：在 GPT Researcher 的 RAG 流程中以 Jev 替代 embeddings，在 28 个 SimpleQA 与开放式研究任务上，相关上下文比例由 46% 升至 73%，盲评报告偏好 15 比 3，单报告成本相同；GPT Researcher 现已默认使用 Jev 且不再需要 embeddings。该结论来自团队自测，尚待独立验证。
AI推荐理由 LangChain 创始人 Harrison Chase 转述 Assaf Elovic 的实验：在 GPT Researcher 的 RAG 流程中用 Jev 替换 embeddings，在 28 个 SimpleQA 与开放式研究任务上，相关上下文从 46% 提升到 73%，盲评报告偏好 15:3，成本相同，并称 GPT Researcher 已默认使用 Jev。给出了具体指标、任务集与仓库链接，但数据来自单一团队自测，尚无第三方复现。
72
浏览 1,806
点赞 13
评论 12
转发 2
正文
retrieval is a decision problem, not just a similarity problem
cool experiment from [@assaf\_elovic](https://x.com/assaf_elovic) swapping embeddings for Jev in GPT Researcher - 73% vs 46% relevant context
decision models will show up all over the harness
[x.com/assaf\_elovic/s…](https://x.com/assaf_elovic/status/2104562754774303203)
Assaf Elovic@assaf\_elovic
Didn't expect this 🤯
We replaced embeddings with Jev in GPT Researcher's RAG pipeline and tested both on 28 research tasks from SimpleQA and open ended research.
Jev beat embeddings on every quality measure we ran:
- 59% more relevant context (73% vs 46%)
- Reports preferred 15 to 3 in blind comparisons
- Same cost per report
GPT Researcher now runs on Jev by default, and no longer needs embeddings at all.
All you need is [@LangChain](https://x.com/LangChain) + [@tavilyai](https://x.com/tavilyai) +Jev for the perfect RAG system.
Check out the repo here:[github.com/assafelovic/gp…](https://github.com/assafelovic/gpt-researcher)q
Research:[docs.gptr.dev/docs/gpt-resea…](https://docs.gptr.dev/docs/gpt-researcher/gptr/context-filter)q
[🔗 View Quoted Tweet](https://x.com/assaf_elovic/status/2104562754774303203)