. . AI: Gemini, Gemini, Apple, and more Google (12.6.23)

Author

Marshall Kirkpatrick
December 06, 2023

Friends, Google's Gemini release dominated the AI conversation today and I've selected two of the links most-discussed below.  If you do data analysis, you'll probably enjoy the short video in the second link.  Then there's Apple AI news and an unrelated Google project.

Cheers,

First impacted: Bard users, AI researchers
Time to impact: Short

Sundar Pichai, Google's CEO, and Demis Hassabis, CEO of Google DeepMind, have penned a blog post launching Gemini, which they say is their most advanced and versatile AI model yet. Gemini comes in three sizes: Ultra, Pro, and Nano. It was tested on various tasks, with Gemini Ultra reportedly outperforming human experts on MMLU, a test of 57 subjects for world knowledge and problem-solving abilities, reportedly the model to do so. You can test Gemini Pro at bard.google.com today; Ultra is still being tested for safety before it's released.

One cool detail about this story: Google co-founder Sergey Brin was reportedly contributing every day to Gemini, often pair programming with other developers. [Introducing Gemini: our largest and most capable AI model] Share by email

First impacted: AI researchers, Data analysts
Time to impact: Short

Google's new AI model Gemini, which they claim has strong capabilities in handling image, audio, video, and text, is demonstrated in a pretty remarkable 3 minute video in which it reads and extracts data from over 200,000 research papers while its human partners eat lunch. Of all the links being shared about Gemini today, this is the one that leaders of the AI community are engaging with the most. [Gemini: Unlocking insights in scientific literature] Share by email

First impacted: Machine learning researchers, Software developers
Time to impact: Short

Apple's machine learning research team has introduced a new software, MLX Data, tailored for Apple silicon. "MLX Data is a framework agnostic data loading library ...for machine learning training or on its own for data pre-processing.... The goal of this library is to allow users to leverage multiple threads for data processing pipelines without the inflexibility of dealing with multiple processes or having to write in a symbolic language." [MLX Data — MLX Data 0.0.1 documentation] Share by email

First impacted: 3D Modelers
Time to impact: Short

Columbia University, Google Research, and Google DeepMind have developed a method called ReconFusion for 3D reconstruction. "3D reconstruction methods such as Neural Radiance Fields (NeRFs) excel at rendering photorealistic novel views of complex scenes. However, recovering a high-quality NeRF typically requires tens to hundreds of input images, resulting in a time-consuming capture process. We present ReconFusion to reconstruct real-world scenes using only a few photos." [ReconFusion: 3D Reconstruction with Diffusion Priors] Share by email

That’s it! More AI news tomorrow.