AI startup SingularityNet plans to deploy a “multi-tiered cognitive computing network” in the coming months. It is designed to host and train the models that will form the basis of an artificial general intelligence (AGI) capable of matching – or even surpassing – human cognition, the company announced Monday.
The realization of AGI is widely seen as the next major milestone in the development of artificial intelligence. While today’s top models such as GPT-4o and Gemini 1.5 Pro are immensely powerful and can perform certain tasks at superhuman levels, they are unable to apply these skills across disciplines. AGI, although still theoretical at this point, would be free of these limitations and able to think and learn on its own, regardless of the task.
SingularityNet is working to build the compute base needed to train and deploy such a system, using some of the most advanced components currently available on the market. According to a report from LiveScience, the startup’s modular supercomputer will include Nvidia L40S GPUs, AMD Instinct and Genoa processors, Tenstorrent Wormhole server racks with Nvidia H200 GPUs, and Nvidia’s 1,500W+ GB200 Blackwell systems.
Supercomputer architectures differ from traditional desktop configurations in that they run multiple sets of processors (both CPUs and GPUs) grouped into individual nodes. These nodes are then chained together in the tens of thousands to form larger arrays of the parent supercomputer.
“This supercomputer in itself will be a breakthrough in the transition to AGI. While the novel neural-symbolic AI approaches developed by the SingularityNET AI team somewhat reduce the need for data, processing, and energy compared to traditional deep neural networks, we still need significant supercomputing facilities,” Ben Goertzel, CEO of SingularityNet, told LiveScience. “The mission of the computer we are developing is to ensure a phase transition from learning from large data sets and then reproducing contexts from the neural network’s semantic memory to non-imitative machine reasoning based on multi-stage reasoning algorithms and dynamic world modeling.”
“We are witnessing a paradigm shift toward continuous learning, seamless generalization, and reflexive self-modification of AI,” he continued.
The company plans to give the public access to the supercomputer once it is fully online in late 2024/early 2025. This will be done through a token system. Users can purchase tokens as they would in a classic arcade, and then spend those tokens to gain a certain number of opportunities to play with the system. The data generated by these interactions will then be fed back into the system for further AGI experimentation and development.
SingularityNet is far from the only company striving to develop and deploy the first AGI. The pursuit of such systems is one of OpenAI’s founding principles, and Meta’s Mark Zuckerberg has committed more than $10 billion to his company’s AGI research and development.
