• We believe that the solution space for architectures and models is vastly bigger than the ones that define the current "Meta" of generative AI. There are more than just transformers and LLMs. We recognize its usefulness at the given moment as it accelerates progress, but the future holds solutions with a higher grade of efficiency and capabilities.
• We believe scalable and continual memorization is one of the next steps towards AGI.
• We believe that open source, edge, and confidential computing will be a key part of distributed AI systems.
• We believe that information flow in the crypto industry can be less fragmented. It's a natural and expected consequence of decentralization, yet it is possible and desirable to develop systems that can help improve it.
• We believe that complexity must always be justified, constantly re-evaluated, and reduced whenever possible.
• We believe that AI applications within the crypto industry must be designed with additional care. The space is much more adversarial than Web2, the stakes are higher, and risks like data poisoning can lead to more fatal outcomes.
• We believe the biggest value is created from tangible, simplest, and most straightforward things.
• We believe deconstructing complexity is the right next step.
• We believe highly integrated AI systems will emerge as new tool primitives akin to smartphones or browsers.
• We believe there is merit in doing and trying out things in a completely different way, starting from the bottom up.
• We focus on pushing the boundaries of compute efficiency.
• We focus on designing and developing systems for highly interactive and intuitive applications.
• We work on novel architectures for contextualization and memory.