Kylie Robison and Alex Heath, reporting for The Verge:

OpenAI is working on its own X-like social network, according to multiple sources familiar with the matter.

While the project is still in early stages, we’re told there’s an internal prototype focused on ChatGPT’s image generation that has a social feed. CEO Sam Altman has been privately asking outsiders for feedback about the project, our sources say. It’s unclear if OpenAI’s plan is to release the social network as a separate app or integrate it into ChatGPT, which became the most downloaded app globally last month. An OpenAI spokesperson didn’t respond in time for publication.

Only one thing comes to mind for why OpenAI would ever do this: training data. It already collects loads of data from queries people type into ChatGPT, but people don’t speak to chatbots the way they do other people. To learn the intricacies of interpersonal conversations, ChatGPT needs to train on a social network. GPT-4, and by extension, GPT-4o, was presumably already trained on Twitter’s corpus, but now that Elon Musk shut off that pipeline, OpenAI needs to find a new way to train on real human speech. The thing is, I think OpenAI’s X competitor would actually do quite well in the Silicon Valley orbit, especially if OpenAI itself left X entirely and moved all of its product announcements to its own platform. That might not yield quite as much training data as X or Reddit, but it would presumably be enough to warrant the cost. (Altman is a savvy businessman, and I really don’t think he’d waste money on a project he didn’t think was absolutely worth it.)

OpenAI might also position the network as a case study for fully artificial intelligence-powered moderation. If the site turns to 4chan, it really doesn’t benefit OpenAI unless it wants to create an alt-right persona for ChatGPT or something. (I wouldn’t put that past them.) Content moderation, as proven numerous times, is the most potent challenge in running a social network, and if OpenAI can prove ChatGPT is an effective content moderator, it could sell that to other sites. Again, Altman is a savvy businessman, and it wouldn’t be surprising to see the network be used as a de facto example of ChatGPT doing humans’ jobs better.

In a way, OpenAI already has a social network: the feed of Sora users. Everyone has their own username, and there’s even a like system to upvote videos. It’s certainly far from an X-like social network, but I think it paints a rough picture of what this project could look like. When OpenAI was founded, it was created to ensure AI is beneficial for all of humanity. In recent years, it seems like Altman’s company has abandoned that core philosophy, which revolved around publishing model data and safety information openly so outside researchers could scrutinize it and putting a kill switch in the hands of a nonprofit board. Those plans have evaporated, so OpenAI is trying something new: inviting “artists” and other users of ChatGPT to post their uses for AI out in the open.

The official OpenAI X account is mainly dedicated to product announcements due to the inherent seriousness and news value of the network, but the company’s Instagram account is very different. There, it posts questions to its Instagram Stories asking ChatGPT users how they use certain features, then highlights the best ones. OpenAI’s social network would almost certainly include some ChatGPT tie-in where users could share prompts and ideas for how to use the chatbot. Is that a good idea? No, but it’s what OpenAI has been inching toward for at least the past year. That’s how it frames its mission of benefiting humanity. I don’t see how the company’s social network would diverge from that product strategy Altman has pioneered to benefit himself and place his corporate interests above AI safety.