michael-dean-k/

If OpenAI were to build a social media company, they should have an AI-first approach. Grok is just a dongle. If you think about it, chatbots have so much context on the emerging thoughts of a user. There has to be a way to pluck something from a private conversation and make it a “post.” Now, when others are having chats, they don’t just get LLM answers, but through embeddings, they can surface relevant posts from other users. This is a good example of a semantic social graph, not an engagement one. And consider talking to a chatbot is the primary experience, it beats the lurker paradox—100% of users are creators. Discovery is guided by your prompts. There’s an opportunity to do this right, I just have little trust.

Some notes on AI-based social media.

• During conversation, the chat will compress your dialogue into nodes that you can save or reject.
• You can easily make a node private or public. Private nodes affect your system memory (for future chats). Public nodes are discoverable by others.
• As you chat, it’s not just showing AI responses, it’s using embeddings to pull in related social media posts of other people’s public nodes. These can be displayed as a horizontal swappable carousel.
• Chats are not fleeting, they’re like text threads that update through time. New social nodes can be added via agents or others in the network.
• Conversations are auto-segmented; if you have 3 tangents in one chat, it will separate them (and add hyperlinks).
• AI group chats let me tag 1-10 friends. By doing it, their social nodes are factored into the response. Their AIs are credited when relevant, and at any point they can jump in themselves.
• In an AI social networks, replies should be shorter and skimmable, but can be clicked to be expanded.