A powerful new supercomputer is set to come online in September, with the goal of accelerating the development of human-level artificial intelligence. This modular supercomputer, built by SingularityNET, forms a "multi-level cognitive computing network" designed to host and train the architecture needed for AGI (Artificial General Intelligence).
The system will incorporate advanced components including NVIDIA L40S GPUs, AMD Instinct and Genoa processors, Tenstorrent Wormhole server racks with H200 GPUs, and NVIDIA GB200 Blackwell systems. It aims to enable a transition from AI based on large-scale data learning and simple context reproduction to more advanced non-imitative machine thinking involving multi-step reasoning algorithms and dynamic world modeling.
Key capabilities the system is targeting include:
- Continuous learning: AI that can constantly learn and update knowledge
- Seamless generalization: Applying learned knowledge to new, unseen situations
- Reflexive AI self-modification: AI that can autonomously improve and adjust its own algorithms and structure
The first supercomputer is scheduled to begin operation in September, with the full project expected to be completed by late 2024 or early 2025, depending on vendor delivery timelines.
SingularityNET's goal is to provide data access for the development of AI, AGI and future artificial superintelligence. The new hardware infrastructure is specifically built to implement OpenCog Hyperon, an open-source software framework for AI system architecture, and its AGI ecosystem environment.
To grant users access to the supercomputer, a token system common in AI will be used. Users can access and add data to existing datasets that others rely on to test and deploy AGI concepts.
While estimates vary, some experts predict AI systems may reach or exceed human-level intelligence by 2027-2028. This new supercomputing network represents a significant step toward developing more human-like artificial intelligence capabilities.