Weekly AI - 10 Dec 2023
Hello readers! Welcome to Weekly AI. In this issue, we'll learn about major new AI capabilities unveiled by Google, Stability AI, Amazon, Meta, and Microsoft. We'll see how new models are pushing the boundaries of what's possible with image generation, video editing, natural language understanding, creation of new materials, and even strategic gameplay.
As always, I appreciate your thoughts and questions about these topics and the continued march of artificial intelligence progress. Please keep sending feedback so we can have an open, informed dialogue. For now, let's jump in to this week's top AI stories!
This newsletter is also available in Spanish and Catalan.
Google Unveils Gemini, Its Largest and Most Capable AI Model Yet
On December 6th, Google announced the launch of Gemini, its biggest and most advanced AI model to date. Built by Google DeepMind, Gemini sets new benchmarks in understanding text, images, video, audio, and code. It massively outperforms previous models across over 30 research benchmarks. Gemini will start rolling out across Google products like Search, Bard, and Pixel phones over the coming months.
Stability AI Unveils SDXL Turbo: A Real-Time Text-to-Image Generator
Stability AI introduced SDXL Turbo, a novel text-to-image model that can generate high-quality images from text descriptions in real time. Leveraging a new technique called Adversarial Diffusion Distillation (ADD), SDXL Turbo massively reduces the number of steps required to produce an image, from 50 down to just one. This enables unprecedented performance, with the model generating 512x512 images in around 200 milliseconds. You can try it here.
Amazon Unveils Amazon Q, An AI Assistant for the Workplace
Amazon recently announced the launch of Amazon Q, a new AI-powered assistant designed for workplace settings. Available in preview, Amazon Q allows employees to have natural conversations to get answers, generate content, and take actions relevant to their business using their company's data and systems. Amazon says Amazon Q has over 40 built-in connectors to integrate with company information and tailors interactions based on a user's role and permissions to provide secure, personalized support.
Amazon Unveils New AI Image and Text Capabilities
Amazon announced the launch of new artificial intelligence services powered by Amazon Titan, the company's suite of AI models. This includes Amazon Titan Image Generator for creating custom images from text prompts, Amazon Titan Multimodal Embeddings for combining images and text in machine learning models, and general availability of Amazon Titan Text Lite and Express models for natural language tasks.
Pika's AI-powered video editing platform raises $55M, launches new creative tools
Startup Pika recently closed a $55 million funding round to support the launch of Pika 1.0, a major update to its AI-powered video editing and generation platform. Dropping on November 28th, Pika 1.0 introduces new capabilities like extending videos, transforming styles from live action to animation, and using AI to edit content by changing clothes or adding characters.
Meta Launches Standalone AI Image Generator
Meta launched a new AI-powered image generation experience called Imagine With Meta. Similar to DALL-E and Stable Diffusion, it creates images from natural language text prompts. The model introduces AI watermarking for attribution and aims to support responsible image generation. You can try it here.
Meta’s Project CICERO Masters Complex Strategy Game Diplomacy
Meta AI has created CICERO, the first AI system to achieve expert-level performance in the complex strategy game Diplomacy, which requires negotiation and cooperation. By combining natural language processing and strategic planning, CICERO can collaborate, coordinate, and negotiate with humans at a high level.
New method for high-quality image-to-video synthesis enables consistent character animation
Researchers from Alibaba released a paper on November 23, 2023, presenting Animate Anyone, a novel AI framework for converting still images into high-quality, controllable videos of animated characters. It achieves state-of-the-art results by better preserving visual details and ensuring smooth motion over time. The method also allows animating a wide variety of character types.
AI and Robots Team Up to Discover New Materials
An AI system called the A-Lab, which combines robotics and artificial intelligence, has announced its first batch of newly created materials. The A-Lab devises and carries out materials synthesis completely autonomously, without any human intervention. It successfully produced 41 new inorganic compounds that could have applications in batteries, solar cells, and other clean technologies.
Major Countries Sign AI Safety Agreement
The U.S., U.K., and over 15 other countries signed an international agreement focused on keeping artificial intelligence safe and secure. The 20-page document lays out recommendations for companies designing and deploying AI, such as monitoring systems for misuse, protecting data from tampering, and vetting software suppliers. While the agreement itself is non-binding, its symbolic support from major world powers signals growing momentum and consensus around establishing protocols and best practices for mitigating risks as AI becomes more integrated across industries and society.
Tesla Rolls Out Major FSD Update, Claims True Self-Driving Coming by End of 2023
Tesla started releasing its FSD v12 software update to employees. CEO Elon Musk has previously said Tesla will achieve "true self-driving capability" by the end of 2023, linking this goal to the v12 update. Musk says vehicle controls will now be handled completely by neural nets rather than hard-coded by engineers. However, it's unclear if drivers will still need to monitor and be ready to take over control. The update is a critical step, but many experts remain skeptical of Tesla's timelines and claims around full autonomy.
Bing Launches "Deep Search" Powered by GPT-4 for More Relevant Answers
Microsoft announced the launch of a new "Deep Search" feature for Bing powered by OpenAI's latest AI system, GPT-4. Deep Search is designed to provide users with more comprehensive and relevant answers to complex search queries by expanding the initial query to better capture the user's intent. It then searches across a wider range of pages and sources to find pertinent results. However, Deep Search is optional and can take up to 30 seconds to generate results.
That wraps up this week's AI news! If you found this informative, don't forget to share it with friends and colleagues. And be sure to subscribe to get next week's news straight to your inbox. Thanks for reading!