Directly competing with OpenAI for users? microsoft's GitHub AI assistant introduces Anthropic and Google model selection.

Hard AI · Oct 30 12:56

来源：硬AI 作者：李丹

除了GPT-4o，GitHub Copilot还将提供Gemini 1.5 Pro和Anthropic的Claude 3.5 Sonnet作为替代选择。GitHub会继续将OpenAI的模型作为默认设置，但开发者可以根据需要切换到谷歌和Anthropic的模型。

OpenAI“金主”微软正在让OpenAI面临的竞争越来越激烈，劲敌Anthropic和谷歌已经把擂台打到了OpenAI的老客户——微软旗下的软件项目托管平台GitHub。

美东时间10月29日周二，GitHub发布公司博客称，除了目前使用的OpenAI旗舰AI模型GPT-4o，该平台的人工智能（AI）编写代码助手GitHub Copilot还将提供几种新的模型供开发者选择，包括OpenAI的o1预览版和迷你版o1-preview和o1-mini，以及谷歌Gemini 1.5 Pro和Anthropic的Claude 3.5 Sonnet。

这是GitHub首次让用户使用Anthropic和谷歌的模型，意味着GitHub同意将Anthropic和谷歌各自的新AI模型整合到拥有至少数百万开发者用户的代码助理中。

此举对GitHub而言可以强化自家AI助手的能力，对与GitHub合作了几年的OpenAI来说，就是直接面对可能被其他公司抢走用户的威胁，因为现在如果GitHub平台的开发者不想用GPT-4o支持GitHub Copilot，就可以选Anthropric或者谷歌的模型支持。

GitHub的CEO Thomas Dohmke称，GitHub会继续将OpenAI的模型作为默认设置，但开发者可以根据需要切换到谷歌和Anthropic的模型。Dohmke这样诠释GitHub提供新选择的意义：

“我们坚信单一模型的时代已经结束。”
“没有一种模型可以适用于所有情况，开发人员希望机构采用最适合他们的模型进行构建。很明显，下一阶段的AI 代码生成不仅将由多个模型功能定义，还将由多个模型选择定义。今天，我们实现了这一点。”

Dohmke 表示，过去一个月里，GitHub与Anthropic 和谷歌的工程团队就其模型的安全性、保密性和可扩展性进行了合作，GitHub计划未来扩大可选模型的范围，“但目前还没有宣布任何合作伙伴关系”。他还说，GitHub不想让开发者的选择过多。

据GitHub介绍，新模型选择将首先适用于GitHub Copilot 的聊天功能Copilot Chat。OpenAI的o1-preview 和 o1-mini现在已经Copilot Chat中推出，Claude 3.5 Sonnet 将在下周逐步上线，Gemini 1.5 Pro 将在未来几周推出。

然后，新模型将融合到GitHub Copilot的主要部分。从Copilot Workspace到多文件编辑、代码审查、安全自动修复和 CLI，GitHub Copilot 的许多表面区域和功能都将很快提供多模型的选择。届时开发者只要用几个简单的提示词，就可以得到代码输出。

GitHub周二还宣布推出名为GitHub Spark的AI原生工具，开发者用它能完全以自然语言构建应用程序App，也就是说，可以直接用英语提示词、而不是编程代码创建App。GitHub称，GitHub Spark是功能齐全的微型App，可以集成AI 功能和外部数据源，无需管理任何云资源。

一些媒体提到，一些媒体提到，微软今年有一些与OpenAI对手合作的动作。比如有媒体称，今年2月，微软宣布与早期AI 模型开发商 Mistral 建立合作伙伴关系，3月，微软称在招募前 DeepMind 联合创始人 Mustafa Suleyman 及其初创公司 Inflection 的员工。Inflection推出的Pi 助手就是ChatGPT的竞品。今年夏季披露的监管文件中，微软还将OpenAI列为竞争对手。

还有媒体指出，自 2018 年收购 GitHub 以来，微软就一直让GitHub保持广泛的合作伙伴关系，即使从技术上讲，这些合作伙伴也是微软的竞争对手。GitHub之前就为用户提供了谷歌和亚马逊云服务的选择。微软还允许其 Azure 云客户使用 OpenAI 以外的模型，包括 Meta和 Mistral 的模型，但没有提供谷歌和Anthropic 的模型。

对于GitHub提供OpenAI以外模型选择的消息，社交媒体X有评论称，微软是不是意识到为什么 VS Code 分支会如此有趣，所以通过在 GitHub Copilot 中提供 Anthropic和谷歌的模型加大对 OpenAI 的压力。

还有评论称这是明智之举，边缘应用层公司的优势在于，他们可以自由切换到最新、最佳的模型，不会出现任何问题。

编辑/lambor

Source: Hard AI, Author: Li Dan.

In addition to GPT-4o, GitHub Copilot will also provide Gemini 1.5 Pro and Anthropic's Claude 3.5 Sonnet as alternative options. GitHub will continue to use OpenAI's models as the default setting, but developers can switch to Google and Anthropic's models as needed.

OpenAI's 'master' Microsoft is intensifying the competition faced by OpenAI, with rivals Anthropic and Google bringing the battle to OpenAI's old customer - the Microsoft-owned software project hosting platform GitHub.

On Tuesday, October 29, Eastern Time, GitHub announced in a company blog that in addition to the current flagship AI model GPT-4o by OpenAI, its artificial intelligence (AI) code-writing assistant GitHub Copilot will provide several new models for developers to choose from, including OpenAI's o1 preview and mini versions, as well as Google's Gemini 1.5 Pro and Anthropic's Claude 3.5 Sonnet.

This is the first time GitHub has allowed users to use Anthropic and Google's models, meaning GitHub has agreed to integrate the new AI models from Anthropic and Google into its code assistant, which has at least millions of developer users.

This move can strengthen GitHub's own AI assistant capabilities, but for OpenAI, which has been collaborating with GitHub for several years, it is a direct response to the threat of potentially losing users to other companies because now if developers on the GitHub platform do not want to use GPT-4o to support GitHub Copilot, they can choose Anthropic or Google's models instead.

GitHub's CEO Thomas Dohmke stated that GitHub will continue to use OpenAI's models as the default setting, but developers can switch to Google and Anthropic's models as needed. Dohmke interpreted the significance of GitHub providing new choices as follows:

"We believe the era of a single model has ended."
"No one model can fit all scenarios, developers hope that institutions will adopt the most suitable model for them to build. It is obvious that the next phase of AI code generation will not only be defined by multiple model features, but also by the selection of multiple models. Today, we have achieved this."

Dohmke stated that in the past month, GitHub has collaborated with engineering teams from Anthropic and Google on the security, confidentiality, and scalability of their models. GitHub plans to expand the range of optional models in the future, "but currently, no partnerships have been announced." He also mentioned that GitHub does not want developers to have too many choices.

According to GitHub, the new model selection will first be applied to the chat feature Copilot Chat of GitHub Copilot. OpenAI's o1-preview and o1-mini are now available in Copilot Chat, Claude 3.5 Sonnet will gradually be launched next week, and Gemini 1.5 Pro will be released in the coming weeks.

Next, the new models will be integrated into the main part of GitHub Copilot. From Copilot Workspace to multi-file editing, code reviews, security auto-fixes, and CLI, many surface areas and features of GitHub Copilot will soon offer the choice of multiple models. Developers will be able to get code outputs with just a few simple prompts.

GitHub also announced the launch of an AI-native tool called GitHub Spark on Tuesday, which developers can use to build applications entirely in natural language. This means that apps can be created directly using English prompt words instead of programming code. According to GitHub, GitHub Spark is a fully-featured micro app that can integrate AI capabilities and external data sources without managing any cloud resources.

Some media outlets have mentioned that Microsoft has taken some actions this year to rival OpenAI. For example, in February, Microsoft announced a partnership with early AI model developer Mistral, and in March, Microsoft said it was recruiting employees from the former DeepMind co-founder Mustafa Suleyman and his startup company Inflection. Inflection's Pi assistant is a competitor to ChatGPT. In regulatory filings disclosed this summer, Microsoft also listed OpenAI as a competitor.

Some media reports have pointed out that since acquiring GitHub in 2018, Microsoft has always maintained extensive partnerships for GitHub, even though technically these partners are Microsoft's competitors. GitHub previously offered users choices of Google and Amazon cloud computing services. Microsoft also allows its Azure cloud customers to use models other than OpenAI, including Meta and Mistral's models, but did not provide Google and Anthropic's models.

Regarding the news of GitHub offering model choices beyond OpenAI, social media X commented whether Microsoft realizes why the VS Code branch is so interesting, so by providing Anthropic and Google models in GitHub Copilot, it increases the pressure on OpenAI.

Other comments also believe that this is a wise move. The advantage of edge application layer companies is that they can freely switch to the latest and best models without any issues.

Editor/Lambor

The translation is provided by third-party software.

The above content is for informational or educational purposes only and does not constitute any investment advice related to Futu. Although we strive to ensure the truthfulness, accuracy, and originality of all such content, we cannot guarantee it.