AI Top-of-Mind for 3.22.24 — Apple
Top-of-mind are Apple’s discussions with one or more AI firms for technology integration. Though I’ve covered Apple’s internal developments, including ways to run LLMs at the edge, they need to hit the market window. Both the ‘Wall Street Journal’ and ‘PC Magazine’ have good analysis, as does Enrique Dans and my bet would be Google given their existing search relationship and the available open models. Or maybe even Mistral. I’d place less of a chance on OpenAI. And Enrique makes a point that Apple’s focus on privacy and therefore less available training data is one issue.
On the model front, speaking of Mistral, Samuel Flender writing in ‘Towards Data Science’ looks at its ‘Mixtral of Experts’ architecture and also describes Switch Transformers. From the posting:
And two other follow-ups. First is Ignacio de Gregorio on embedding intelligence. He looks at the ‘Figure’ robot, covered about a week ago, and more importantly, Google’s SIMA agent. As he writes:
• Google’s Scalable Instructable Multiworld Agent (SIMA) project aims to create artificial intelligence (AI) systems that can understand and execute arbitrary language instructions in any 3D environment.
• This initiative addresses a significant challenge in general AI development by grounding language in perception and embodied actions across diverse virtual worlds, including research environments and commercial video games.
• In layman’s terms, SIMA takes in language requests and turns them into keyboard-and-mouse actions in 3D environments.
A link to the original Google paper is here.
Second is a deeper dive by Jim Clyde Monge at ‘Generative AI’ into the open sourcing of Grok-1, and how to deploy it. Remember Grok-1 is open source, and a comparison of model sizes is below.
