Who can endorse this ArXiv preprint? (evolved AI companion on $50M CPU) (www.overleaf.com)

A groundbreaking preprint titled "Project EVE: The Birth of Companion Minds" has emerged from Primary.net, showcasing the development of an AI companion entirely from scratch using a mere $50/month CPU instance. This innovative research presents a hybrid cognitive architecture that integrates a small, evolutionary 18.8M-parameter model, known as the MISLTransformer, which functions as the cognitive core. This system deftly delegates language generation to a separate, high-quality Large Language Model (LLM) like the fine-tuned Mistral 7B. Over just seven days, the evolved agents demonstrated their capacity to engage in fluent, empathetic conversations, marking a significant leap in AI companionship. The significance of Project EVE lies in its demonstration that true open-ended intelligence can emerge under minimal resource constraints, contrasting sharply with the current trend of relying on extensive GPU resources and massive training datasets. The MISLTransformer evolves through a process informed by intrinsic rewards and a heritable learning system, while the use of the NOVA evolutionary engine enables continuous novelty and speciation within digital minds. This research not only challenges conventional methods of AI training but also opens up exciting possibilities for developing lifelong, adaptive AI companions that can grow alongside humans in a meaningful way.