Meta abandons open-source Llama for proprietary Muse Spark (thenewstack.io)

Meta has announced a significant shift from its open-source Llama models to a proprietary AI language model called Muse Spark. This transition, prompted by concerns over Llama's competitiveness compared to leading models like ChatGPT, involves building Muse Spark from scratch with new infrastructure and architecture, backed by substantial investments in AI talent. While Meta promises that existing Llama models will remain available, it has not committed to any future development or enhancements, effectively leaving current Llama users in a challenging position. The implications of this pivot are stark for developers previously reliant on Llama’s open weights, as they face a lack of migration options due to Muse Spark's cloud-only deployment model and absence of downloadable weights. Developers can either continue using Llama, switch to other open-source models, or migrate to proprietary APIs, which can be costly and complex. The emergence of forks and alternatives, such as llama.cpp and OpenLLaMA, offers some respite, but the community's overarching concern remains: Meta's abandonment of open-source AI undercuts the foundation it once laid, raising questions about the future accessibility and innovation in the AI landscape.