. . AI: Microsoft Gets Top AI Talent from Inflection AI (3.19.24)

Friends, today Microsoft has made headlines with the appointment of their new CEO for Microsoft AI, Mustafa Suleyman, and Chief Scientist, Karén Simonyan. It has been interesting to see the number of rumors of partnerships, acqui-hires and paradigm changing industry developments in the last 18 months. It really does feel like a year is passing by every month! In other news we see new and improved models, like RAG 2.0!

As always, these are the stories the AI community is talking about the most, according to our weighted analysis of community engagement. Thanks for reading and sharing. I hope you find these stories useful and interesting.

Here’s today’s news…

-Marshall Kirkpatrick, Editor

Mustafa Suleyman is the New Microsoft AI CEO and Inflection AI to be Available on Azure

First impacted: Developers, Businesses
Time to impact: Short

Mustafa Suleyman, cofounder of DeepMind and Inflection AI announced that he has joined Microsoft as CEO of Microsoft AI, leading all consumer AI products and research. He also mentioned that Karén Simonyan will be joining as Chief Scientist. Meanwhile, Inflection AI will continue its mission under new leadership, aiming to make its API widely available to developers and businesses globally. It plans to expand its business by launching Inflection-2.5 on Microsoft Azure and extending to other cloud platforms, according to a seperate blog post by the company. [via @mustafasuleyman] Share this story by email

Project GR00T Launches to Develop Humanoid Robots

First impacted: Robotics Engineers, AI Researchers
Time to impact: Medium

Project GR00T was announced during GTC and is a general-purpose foundation model for humanoid robots, according to Jim Fan. The model will be able to understand multimodal instructions, including language, video, and demonstration, enabling it to perform a variety of tasks. The project, which is a collaboration with leading humanoid companies globally, will use NVIDIA's technology, be simulated in Isaac Lab, trained on OSMO (Nvidia's cloud workflow orchestration platform), and deployed to Jetson Thor (a new edge GPU chip designed specifically for GR00T). If you missed this part during Nvidia CEO Jensen's presentation, check out the videos in the provided link. [via @DrJimFan] Share this story by email

Databricks Acquires Lilac to Boost Unstructured Data Capabilities for AI

First impacted: Data Scientists, AI Application Developers
Time to impact: Medium

Databricks has announced its acquisition of Lilac, a tool aimed at simplifying the evaluation of unstructured data for generative AI. According to Databricks, integrating Lilac's capabilities into their platform will expedite the development of generative AI applications, as Lilac enables data scientists to search, cluster, and analyze text datasets to enable customers to create high-quality generative AI applications using their own data. [Lilac Joins Databricks to Simplify Unstructured Data Evaluation for Generative AI] Share this story by email

Stability AI Launches Improved 3D Generative Model

First impacted: 3D technology developers, commercial users with Stability AI Membership
Time to impact: Short

Stability AI has launched Stable Video 3D, a model that the company says improves 3D technology using Stable Video Diffusion. You put in a 2D image and it generates a 3D video. The model, which can be used commercially with a Stability AI Membership or non-commercially via Hugging Face, is said to generate detailed multi-views from any angle and outperform previous versions and other open source alternatives. Click the link to check out a few examples, they are pretty cool. [Introducing Stable Video 3D: Quality Novel View Synthesis and 3D Generation from Single Images — Stability AI] Share this story by email

Contextual AI Launches Improved RAG 2.0

First impacted: AI developers, AI researchers
Time to impact: Short

Contextual AI has launched RAG (retrieval augmented generation) 2.0, a question-answering system based on your provided data that they say outperforms previous AI models in efficiency and accuracy. According to the company, RAG 2.0 was used in the final stages of developing Contextual Language Models (CLMs), which they claim surpassed strong RAG baselines based on GPT-4 and the best open-source models by a large margin. RAG 2.0, unlike its predecessor, integrates all components as a single system, backpropagating through both the language model and the retriever to maximize performance, moving beyond the limitations of the previous RAG system. [Introducing RAG 2.0 - Contextual AI] Share this story by email

That’s it! More AI news tomorrow.