← 回總覽

构建 Gemini API 技能以克服知识截止限制

📅 2026-03-26 15:57 Philipp Schmid 人工智能 2 分鐘 1286 字 評分: 88
Gemini GoogleDeepMind LLM 智能体技能 开发者体验
📌 一句话摘要 Philipp Schmid 分享了一篇关于构建 Gemini API 技能的新博客,旨在帮助模型保持最新状态并提升性能,在评估测试中取得了 95% 的成功率。 📝 详细摘要 这条推文重点介绍了一种解决 LLM 知识截止问题的技术方法,即为 Gemini API 开发“技能”。通过教会模型了解最新的 SDK 和模型能力,团队提升了 Gemini 3.1 Pro 的性能,在 117 项评估测试中达到了 95% 的通过率。这为使用 LLM 的开发者提供了一种提升模型实用性的实用方案。 📊 文章信息 AI 评分:88 来源:Philipp Schmid(@_philschmid
![Image 1: Philipp Schmid](https://www.bestblogs.dev/en/tweets?sourceId=SOURCE_89ab8d)

We just published a blog on how we built the Gemini API skill. LLMs have fixed knowledge cutoffs, so we need to teach them about our newest models and how to use the SDK. In our evaluations, it helped Gemini 3.1 Pro pass 95% of 117 eval tests. Skills and Blog below

!Image 2: Tweet image

8 Replies

4 Retweets

52 Likes

1,558 Views

One Sentence Summary

Philipp Schmid shares a new blog post on building Gemini API skills to help models stay updated and improve performance, achieving 95% success on evaluation tests.

Summary

This tweet highlights a technical approach to addressing LLM knowledge cutoffs by developing 'skills' for the Gemini API. By teaching the model about the latest SDKs and model capabilities, the team improved Gemini 3.1 Pro's performance, achieving a 95% pass rate on 117 evaluation tests. This offers a practical solution for developers working with LLMs to enhance model utility.

AI Score

88

Influence Score 19

Published At Today

Language

English

Tags

Gemini

GoogleDeepMind

LLM

AgentSkills

DeveloperExperience

查看原文 → 發佈: 2026-03-26 15:57:50 收錄: 2026-03-26 18:00:21

🤖 問 AI

針對這篇文章提問,AI 會根據文章內容回答。按 Ctrl+Enter 送出。