AI Weekly: What's New in AI? – 05/02/2024

?? Curious about the latest models and AI-powered software releases? Bring yourself up to speed with Bluebird’s AI Weekly newsletter!

Does OpenAI have offices outside the US? How small can AI models get while still being super useful? Does Sora have any notable competitors? We have the answers to all these questions and more.

From this week’s selection: OpenAI announces new office in Japan, Microsoft launches the Phi-3 family of open models, Ray-Ban Meta smart glasses get new AI features, Apple introduces CoreNet and OpenELM, Vidu AI from China… and a lot more:

OpenAI’s new office in Tokyo & GPT-4 optimized for the Japanese language

??OpenAI announces its first office in Asia and releases a GPT-4 custom model optimized for the Japanese language. Tadao Nagasaki, new President of OpenAI Japan will lead commercial and market engagement efforts. OpenAI provides local businesses with early access to the optimized model. The company plans to release the custom model more broadly in the API in the next months.

Read the original announcement from OpenAI CEO Sam Altman.

Microsoft launches the Phi-3 small language models

?? Microsoft launched its smallest AI model yet: Phi-3-mini was designed to operate directly on devices like iPhone 14.

The Phi-3 family of small language models were designed to perform well on a variety of language, coding, and math benchmarks—offering a more cost-effective and capable alternative to larger models.

Inspired by bedtime stories, these small models, such as the newly released Phi-3-mini with 3.8 billion parameters, leverage high-quality, educational data to deliver high-performance results. Upcoming models in the Phi-3 series promise further enhancements and choices for users, catering to different needs and resource availabilities.

Read Microsoft’s informative, long-format introduction for the Phi-3 models.

Ray-Ban Meta smart glasses get an update & Firefly Image 3 from Adobe

?? Meta has updated their Ray-Ban Meta smart glasses by introducing new AI features and a video calling option, which are currently in beta and available to everyone. These glasses are equipped with AI technology that uses the cameras to recognize objects, landmarks, translate languages, and even suggest clothing options from the user's wardrobe.

Adobe has introduced its third-generation image-generation model, Firefly Image 3, which is now integrated into Photoshop. This update features advanced AI tools such as 'generative fill with reference image', which enhances creative capabilities.

OpenAI introduces upgrades to its business API.

More in this video here.

Apple introduces CoreNet and OpenELM

?? OpenELM (Open-source Efficient Language Models) is an open-source language model that significantly outperforms earlier models like OLMo by being 2.36 percent more accurate—while using only half the pretraining tokens.

CoreNet is a deep neural network toolkit supporting various tasks, including object detection and semantic segmentation.

Read more about CoreNet & OpenELM.

Vidu AI model developed by ShengShu-AI and Tsinghua University

?? Vidu is the first text-to-video large AI model in China, developed by Tsinghua University and ShengShu Technology.

Vidu can generate 16-second videos in 1080p, similar to OpenAI's Sora. Although much shorter than Sora's 60-second video capability, Vidu is currently the best in China. The new model was unveiled at the Zhongguancun Forum in Beijing over the weekend.

Find out more about Vidu here.

See what the model is capable of in this video.

Useful? If you liked our newsletter, don’t forget to subscribe & find out more about what we do at Bluebird. See you next week!