Magic AI
Magic is advancing towards safe Artificial General Intelligence (AGI) by automating software engineering and research through frontier code models, leveraging their resources of 8,000 H100s, substantial funding, and a collaborative team of engineers and researchers.
What is Magic AI?
Magic's mission is to automate the fields of AI research and code generation to enhance model capabilities and address the alignment problem more effectively than humans can. Their strategy involves utilizing frontier-scale pre-training, domain-specific reinforcement learning, ultra-long context understanding, and significant compute power at inference time. Backed by a substantial investment of $515 million from prominent investors and equipped with a state-of-the-art hardware setup, they are tackling a select set of core research challenges that lie on the path to AGI. Magic also emphasizes a commitment to responsible innovation by instituting an AGI readiness policy to manage and mitigate the existential risks associated with advancing AI capabilities. Additionally, they recently announced a collaboration with Google Cloud, secured more funding, and made progress with ultra-long context models that can process context windows of 100 million tokens.