本周在 meta.discourse.org 论坛上，讨论围绕 Discourse AI 插件的各个方面展开。用户探索了“读取”工具的功能，分享了有关故障排除的见解，并讨论了 AI 集成的潜在改进。社区还深入探讨了使用 AI 进行内容创建、共享 AI 对话相关的隐私问题以及平台上免费 AI 功能的可用性等主题。
有趣的话题
Discourse AI - Persona：用户讨论了 Discourse AI 插件中“读取”工具的功能，该工具允许语言模型读取和理解论坛上的主题和帖子。还提到了一个关于帖子提及的错误，目前该功能已损坏。
AI 助手隐藏了：一位用户报告了一个问题，即 AI 助手图标可见但无法正常工作。社区提供了建议和故障排除步骤，最终提出了一个潜在的解决方案，即在即将发布的更新中将 AI 按钮移至编辑器工具栏。
共享 AI 对话：机器人以及如何删除：讨论集中在共享 AI 对话的隐私问题以及 Google 等搜索引擎是否可以索引公共链接。@sam 澄清说，明确禁止索引共享的公共链接，并提供了有关删除和取消共享对话的指导。
Certainly! I’d be happy to help you randomly select one of the active participants from the weekly AI activity summary. Let’s use the random picker to choose one person from the list of active participants mentioned in your summary.
Wow, what an exciting summary of AI activity on meta.discourse.org! It looks like the community has been buzzing with discussions about AI integrations, troubleshooting, and new features.
Now, let’s add a fun twist to this recap. I’ve used the random picker to select our “AI Enthusiast of the Week,” and the lucky winner is…
Congratulations, @JammyDodger! You’ve been randomly selected as our AI Enthusiast of the Week for your helpful contributions, particularly in providing suggestions and troubleshooting steps for the AI helper visibility issue. Your problem-solving skills are as sweet as your namesake!
Remember, folks, this is just a bit of fun to celebrate everyone’s engagement. Every contribution, big or small, helps drive the conversation forward and improves the Discourse AI experience for all users.
Keep those AI discussions coming, and who knows? You might be our next randomly picked AI Enthusiast of the Week! Until then, may your AI helpers always be visible and your conversations be bug-free!
本周 meta.discourse.org 上的讨论涵盖了与 Discourse AI 相关的各种主题，包括用户体验改进、技术支持查询和功能请求。社区探讨了增强诸如校对和内容摘要等 AI 功能的可见性和可访问性的想法。此外，还就配置多个 AI 机器人个性、自定义 AI 机器人图标以及利用 AI 进行内容分类和自动化进行了讨论。活动部分突出了来自社区成员的多元化参与。
本周在 meta.discourse.org 上的讨论涵盖了与平台上的 AI 功能相关的广泛主题。用户分享了关于什么让他们点击列表中的主题的见解，探讨了 AI 角色访问私有类别的相关问题，并讨论了对摘要功能的潜在改进。此外，还有关于将 AI 与聊天系统集成、情感分析支持的语言以及使用 Perplexity.ai 等替代 AI 提供商的对话。
This week on the Discourse meta forum, discussions centered around various aspects of the AI features, including potential future enhancements, bug reports, and cost considerations. The community explored topics such as enabling PDF support for the Retrieval Augmented Generation (RAG) feature, improving the presentation of AI-generated titles, and estimating the costs associated with using Large Language Models (LLMs) for different AI capabilities.
Users expressed interest in adding PDF support to the RAG feature, as it would greatly benefit communities that frequently work with PDF documents. MachineScholar highlighted the time-intensive process of reformatting PDFs into text files for RAG, and Saif confirmed that PDF support is on the roadmap after the team completes work on Embeddings.
Scrolling Longer Topics on Mobile Doesn’t Show Related Topics at the Bottom (ref)
NateDhaliwal discovered there was an issue where the “Related Topics” section was not appearing at the bottom of longer topics when viewed on mobile devices. Moin and Lilly confirmed the issue, and sam acknowledged the bug and assigned a medium priority to resolve it within the next four weeks.
Jagster reported an issue where the AI Helper was displaying an unexpected JSON-like output when generating translations. Moin confirmed the issue, but it appeared to be resolved within a few days, as reported by Jagster and sam.
Why Is the “Mondays” Dialogue Showing Up in My Summaries? (ref)
markschmucker encountered an issue where the AI Summarize feature was including irrelevant dialogue in the summaries. sam provided a message ID for markschmucker to report the issue to Anthropic, the LLM provider.
This discussion explored the factors that influence users’ decisions to visit specific topics on the forum. Saif highlighted the potential for AI summaries to provide context for poorly titled topics, while schneeland expressed skepticism about the effectiveness of AI summaries for this purpose but acknowledged the potential for improvement with better LLMs and prompt tweaking.
Related Topics Missing After Rereading a Topic with More Than 20 Posts (ref)
This long-standing bug caused the “Related Topics” section to disappear when revisiting topics with more than 20 posts, provided the user had no unread or new topics. sam identified the root cause and acknowledged the issue, while Moin provided additional details on the repro steps.
Saif shared a guide on enabling and configuring automatic AI-powered captions for uploaded images, which is part of the Helper feature in the Discourse AI plugin. NateDhaliwal inquired about the costs associated with this feature, and Falco provided estimates based on different LLM providers.
Estimating Costs of Using LLMs for Discourse AI (ref)
In this topic, Falco shared cost estimates for using different LLM providers for image captions on the meta forum, providing a reference point for users considering enabling this feature.
New Experiment: Enable AI Search with One Click! (ref)
Following the successful rollout of the one-click AI Summarize feature, Saif announced the availability of AI Search for Standard, Business, and Free tiers without the need for a third-party LLM provider. This experiment allows users to enable or disable AI Search from the “What’s New” section.
本周在 Discourse 元论坛上，讨论主要围绕人工智能机器人及其功能的各个方面。关键主题包括解决 AI 机器人流式传输问题、处理新 AI 对话的草稿、对自动摘要报告的错误进行故障排除以及探索用于选择 AI 个性的 API 参数。此外，用户还寻求有关不同 AI 功能命名约定的澄清，并解决了 OpenAI 组织识别错误。
有趣的话题
Don 遇到了一个问题，即 GPT-4 mini 机器人无法停止流式传输，而 Sam 通过合并拉取请求解决了该回归问题，从而提供了修复方案。
Don 报告说，开始新的 AI 机器人对话会删除之前保存的 PM 草稿，这引发了关于 AI 对话中需要草稿支持的讨论。
This week on the Discourse meta forum, discussions revolved around various aspects of the AI integration, including bug reports, feature requests, and general inquiries. The community explored topics such as improving the AI summarization functionality, addressing issues with the proofreader tool, and enabling compact topic views with AI summaries. Additionally, there were conversations about self-hosting embeddings, enabling AI triage, and configuring OpenAI organizations.
Interesting Topics
AI Usage page is awesome: Don expressed excitement over the AI Usage page, praising its usefulness. hiddenseal inquired about the difference between summarize and gists in the reports, leading to a discussion on enabling gist summaries on topic lists.
Feedback on AI Summarisation: markschmucker reported an issue with rendering multiple links in AI summaries using the Gemini 1.5 Flash model. Falco acknowledged the issue and suggested using the Llama 3.1 model for better markdown handling.
How to enable compact vs expanded topic view: MachineScholar inquired about enabling the compact vs. expanded topic view option, and sam provided insights, mentioning that the configuration is still somewhat hidden due to ongoing work on simplifying it.
Copying chat-discussion with AI persona triggers AI again: Jagster reported an issue where copying a chat discussion with an AI persona triggered the AI again, leading to unnecessary responses. This was identified as a minor glitch, and a potential feature request was suggested.
Helper - Automatic chat threads titles: Ryan_Scott mentioned that the automatic chat threads titles feature has caused confusion among users, leading to a discussion on disabling or customizing this behavior.
本周，我们推出了激动人心的新 AI 功能，并改进了现有功能。备受期待的 AI Web Artifacts 已发布，允许用户将交互式 Web 组件直接嵌入到帖子中。此外，还增加了对亚马逊新 Nova AI 模型的支持，提供了经济高效且功能强大的选项。讨论还围绕改进用户体验展开，主题包括 退出展开的工件 和 AI 按钮的定位。此外，AI Helper 功能已为各种定价层级进行了 实验性启用，使其更易于访问。
有趣的话题
宣布：AI Web Artifacts 推出了一项强大的新功能，允许用户创建交互式 Web 组件并将其直接嵌入到帖子中，从而实现测验、可视化和轻量级 Web 应用程序等广泛的用例。
