I believe that integrating AI with forums is a friendly approach, especially for smaller forums, as it can better assist users and add fun to the community. I know some friends are already working on this, and I want to express my gratitude and hope to contribute my modest efforts to accelerate the process. Here are some suggestions I have for using ChatGPT AI:
- Limitations: ChatGPT AI is a paid resource, so there should be reasonable restrictions to fully utilize it. For example, it could be available only under specific tags or for specific user groups, with token limits like using the money plugin or limiting usage per user group per day.
- Features: Some settings can optimize ChatGPT AI's performance, such as
model temperature presence_penalty max_tokens. Corresponding settings should be provided, like long text from model=gpt-3.5-turbo-16k, which might be suitable for forums.
- Flexibility: ChatGPT AI's strength lies in its versatile prompts, allowing the AI to play various roles and provide diverse assistance:
- Service Mode: Preset prompts in settings, with each user having their own AI conversation context. Users can send special text to start a new conversation. This is suitable for service-oriented forums where AI can act as customer support to help users.
- Communication Mode: The first post of a thread serves as the prompts for the conversation (with an option to hide the text from everyone except admins, as some people may not want to share, but I don't mind). This post serves as the context for a group AI conversation (no longer distinguishing between different users), and new posts can start new conversations. This is suitable for forums focused on AI research, where users can study, exchange, and share AI experiences or let AI play more roles to provide assistance.
Here are some additional suggestions that might improve the experience:
- Filters: When a user or AI replies with specific keywords, refuse the service to avoid AI-generated harmful content that may affect the forum.
- Statistics: Get a rough idea of which users utilize more AI resources to better plan AI costs.
- Prompts: Use specific formats to distinguish
system assistant user and make the prompts' performance more in line with expectations.
- Simple Packaging: Allow setting AI avatars, names, disclaimers, etc., to help users better understand the AI.
- If necessary, it is advisable to enable CAPTCHA for AI usage to reduce accidental waste and prevent malicious bots.
- Common error messages, such as those indicating that dialogue memory has been exhausted or that the AI's response time has exceeded the limit, should not be recorded in the conversational context.
Furthermore, I noticed some Github projects using Cloudflare workers or Vercel for redirection. I'm not sure if this could help reduce server pressure for smaller forums. I hope AI can bring more assistance to our lives.