Overview
- Granted in late December and first filed in 2023, the patent lists CTO Andrew Bosworth as the primary inventor.
- The filing outlines a large language model trained on user-specific data—posts, comments, likes, chats, and voice messages—to mimic a person’s online behavior.
- It describes simulations that could make posts, reply to direct messages, and potentially recreate audio or video calls using voice cloning and deepfake techniques.
- The system is framed to operate when a user is absent or deceased, with the document noting the effect is far more permanent if the user has died.
- Coverage highlights mounting ethical concerns over consent, privacy, and grief effects, notes prior similar work like Microsoft’s 2021 patent and grief-tech startups, and reports visible public backlash.