Meta AI for Web

About Meta's AI Research Group

Meta's artificial intelligence laboratory (the organization that used to operate under the Facebook name) develops advanced machine-learning systems for research applications. The group focuses on building models and infrastructure for tasks that range from visual analysis to audio transformation and strategic reasoning.

Core research technologies

• Full-body keypoint detection: systems that locate and track body, hand, face and foot positions in two-dimensional media.
• Music-style and instrument conversion: tools that translate a musical piece into different instruments or stylistic arrangements.
• Image de-identification: methods for obscuring or removing personally identifying information from photos and videos.
• Facial recognition and analysis: technology for detecting and matching faces within imagery.

CICERO — strategic language agent

CICERO is Meta’s high-profile agent that merges large-scale language capabilities with planning and game-theoretic reasoning. It reasons about others’ intentions from conversations and forecasts their likely moves, which enables it to play Diplomacy at a level comparable to skilled human opponents. CICERO can plan ahead, sustain multi-turn dialogues, and build cooperative relationships within a game environment.

Additional products and platforms

• Imagine (paid): a complementary offering positioned as an optional, paid creativity tool.
• MultiRay: a scalable runtime and deployment platform for running many AI models efficiently. It is used to accelerate workflows such as automated moderation and content labeling.

Example tasks supported by the platform:

• Hate-speech and safety detection
• Topic classification and metadata tagging

Access, usage and design constraints

Meta’s AI research systems are primarily available for internal and academic research at this time. Engineers are working to improve model efficiency, scalability, and safety.

CICERO was purposely constructed with design constraints: it is not intended for autonomous self-play, and the system is restricted from achieving strategic advantage through covert manipulation or deceptive dialogue aimed at exploiting human players.