Daily AI Briefing

Meta’s Muse Spark Open brings an 816-billion-parameter model for coding and multi-step tasks to Mac users, with Apple helping optimize it through MLX. While Chinese open weight models, this new model brings new capacities to locally run hardware (even if it’s $40,000 in Mac). Also privacy risks in shared Muse conversations, AI’s effect on customer service and startups, and fresh moves by Amazon, Nvidia and Meta.

Meta releases Muse Spark Open for local Mac inference

Meta released Muse Spark Open, an open-weight model for coding and multi-step tasks, and worked with Apple to optimize it for Mac computers using the MLX toolkit, The Wall Street Journal reported. Meta says the 816-billion-parameter model can run across two M5 Ultra Mac Studios and is available through download and cloud services in its announcement. The hardware requirement is substantial: the Journal puts the cost of enough Mac memory above $40,000. Apple’s research team reports that four Mac Studios generate more than 36 tokens per second, using a 4-bit version of the model in its benchmark. Meta also acknowledges that Chinese open-weight models remain stronger in some areas, including software development.

Quote of the Day

This is the culmination of my life, and at a scale I never could have imagined. I am very high on the AI optimism scale, because I am using AI for good. I want millions of people to be able to use AI for good.”

Bill Gross, founder of Idealab, WSJ

Leave a comment

Your email address won’t be published. Name, email and comment are required.