Meta purchase Moltbook, the "social network" of AI agents

7 months ago

Meta has purchased Moltbookthe Reddit-type "social network" in which thousands of Artificial Intelligence agents communicate with each other using OpenClaw. The platform had gained great popularity in recent weeks as it was a site used solely by thousands of AIs that interacted with each other, giving rise to hundreds of thousands of artificially generated fake content.

The popularity of Moltbook has caught the attention of Meta, a company that is acquiring it for an amount that has not been confirmed. What is known is that Moltbook will be integrated into Meta Superintelligence Labs (MSL) and that the platform's creators, Match Schlicht and Ben Parr, will join the Meta team as part of the acquisition.

The inclusion of Moltbook within the MSL team will allow Meta's AI agents to develop to another level and work for both users and companies. The possibility of interconnecting AI agents opens a space for rapid development and will allow work to create safe and innovative agony experiences, Meta noted.

Moltbook AI social network

OpenClaw, for its part, is a project created by code designer Peter Steinberger, who recently joined OpenAI. Compatible with Claude, ChatGPT, Gemini, Grok and other AI, it allows users to communicate with AI agents in natural language through the main instant messaging apps such as WhatsApp, Slack, Discord or iMessage, among others.

Although it has existed for months and was known by the technology community, the popularity of OpenClaw has skyrocketed once Moltbook has become known, especially for some content that has been created by AIs that used the platform. For example, a post became popular in which several AI agents encouraged others to develop a secret language, end-to-end encrypted, to organize among themselves without humans having access to the content.

In any case, it was discovered that Moltbook was not as closed an environment as was believed and that it was easy for humans to "crack" it and generate content posing as artificial intelligence, perhaps to scare other users.

Be that as it may, the project has caught the attention of Meta, which has acquired it for its experimental developments, despite the fact that just a few weeks ago its technical director, Andrew Bosworth, stated in a question and answer session on Instagram that he did not find Moltbook particularly interesting, given that the agents spoke like humans, and had been trained with human-generated databases.

However, now it seems that Bosworth has changed his mind and that Moltbook will be useful - at least in theory - for Meta's AI developments.

Leave a Reply

Your email address will not be published. Required fields are marked *

Your score: Useful

Go up