Event ·Multi-source

Salesforce 用 8B 小模型训练 AI 科研创意生成，创意性超 Claude Opus 4.8

Salesforce 团队将一个 8B 推理模型训练成一次性生成多个科研创意的 AI-scientist 管线。奖励机制会对整组创意打分，按每个创意与已有创意的距离加权质量，让模型学会拉开新创意与前面创意的差异。在其无数据污染的基准上，这个 8B 模型在衡量质量与多样性的 creative utility 指标上超过 Claude Opus 4.8 和 GPT-6 Astra。论文已发布在 arXiv（编号 2610.22324）。

Current read

Salesforce 只用 8B 小模型就教 AI 一次写出一堆不重复的科研点子，创意性还赢了 Claude Opus 4.8 和 GPT-6 Astra，做法挺巧的。

11 sources43° AI heatLast updated 2026/10/17 22:04:02

Evidence

11 sources

Primary sourceelvis

2026/10/17 22:04:02View source

Related source 1OpenRouter

2026/10/15 22:24:41View source

Related source 2Teortaxes

2026/10/17 03:40:54View source

Related source 3kimmonismus

2026/10/18 07:23:35View source

Related source 4阮一峰的网络日志

2026/10/16 00:02:56View source

Related source 5Tw93

2026/10/16 01:12:10View source

Related source 6Jeff Dean

2026/10/16 17:29:45View source

Related source 7Browser Use

2026/10/18 05:49:18View source

Related source 8Bindu Reddy

2026/10/18 21:00:57View source

Related source 9TestingCatalog

2026/10/19 20:07:11View source

Related source 10AI Will

2026/10/16 02:47:42View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story