Today:

• OpenAI Stealth Models

• Microsoft's $1.7B Indonesian Investment

• Apple Poaches Google Talent

• Amazon Q Apps Launch

• Google, News Corp AI Deal

• Command R on Bedrock

AI NEWS: OpenAI STEALTH Models | California KILLS Open Source?

OpenAI's internal culture, known for its humor and playful trolling, surfaced once again when Sam Altman, a prominent figure, emphasized the skill of concise communication. He suggested that practicing with a friend to rephrase thoughts more succinctly can significantly enhance one’s clarity—drawing millions of views. 

Altman's insights resonate with recent developments in AI training where smaller models learn from more capable ones to improve efficiency. Notably, OpenAI used this technique with GPT-4 to help smaller models like Orca 2 perform at levels comparable to much larger systems, validating Altman’s advocacy for effective and concise communication.

WATCH THE VIDEO ON YOUTUBE

Microsoft announces US$1.7 billion investment to advance Indonesia’s cloud and AI ambitions

Microsoft announced a substantial $1.7 billion investment in Indonesia to boost its cloud and AI capabilities, marking its largest investment in the country in 29 years. This investment will support the Indonesian government's Golden Indonesia 2045 Vision, aiming to transform the country into a global economic powerhouse. 

The funds will enhance cloud and AI infrastructure, provide AI training for 840,000 Indonesians, and bolster the local developer community. These initiatives are designed to empower Indonesians with the necessary infrastructure and skills for the AI era, positioning Indonesia as a hub for digital talent and innovation.

MICROSOFT

Apple targets Google staff to build artificial intelligence team

Apple has been actively hiring artificial intelligence experts from Google to enhance its AI and machine learning capabilities. Apple has recruited at least three dozen specialists from Google, primarily operating out of California and Seattle, with a significant base also in Zurich, Switzerland. The Zurich team's growth stems from Apple’s acquisition of two AI startups, leading to the establishment of a "Vision Lab" research facility. 

This aggressive recruitment drive is part of Apple's broader strategy to advance its AI technologies, including developing systems like ReALM (Reference Resolution as Language Modeling), which aims to revolutionize voice assistant interactions and potentially transform the commerce landscape by improving AI communication abilities.

FINANCIAL TIMES

Amazon Q, a generative AI-powered assistant for businesses and developers, is now generally available

Amazon has launched Amazon Q, a generative AI-powered assistant designed to enhance productivity for developers and business users. This tool is now available for widespread use, offering capabilities such as generating, testing, and debugging code, as well as planning and reasoning to optimize business operations. Amazon Q integrates seamlessly with company data across various platforms, providing insights and facilitating data-driven decisions. 

In addition to these technical capabilities, the platform promotes increased productivity, with early users reporting up to 80% efficiency improvements. Amazon Q also includes a new feature, Amazon Q Apps, enabling employees to create custom AI-driven applications without coding experience. This launch is part of Amazon's broader AI Ready initiative aiming to train 2 million people in AI skills by 2025.

AMAZON

As Tech Firms Prioritize AI in News Dealings, Other Deals Get Cut

Google has entered into a new deal with News Corp, owner of Wall Street Journal, to pay $5-6 million annually to support the development of AI-related content and products. This agreement follows Google's decision to end a previous partnership with The Journal, which resulted in layoffs within the news organization. 

The focus on AI initiatives reflects a broader trend among tech companies to prioritize AI in their dealings with news publishers, while reducing investments in other types of news partnerships. This shift has prompted concerns among media executives about tech firms' inconsistent strategies and their tendency to abruptly terminate content creation deals.

THE INFORMATION

Cohere’s family of Command R models arrives in Amazon Bedrock

Cohere has expanded the availability of its Command R and Command R+ large language models (LLMs) to Amazon Bedrock, enhancing the options for enterprise customers. This update follows the recent launch of Command R+, which has shown superior performance over OpenAI's GPT-4 in benchmarks, and was previously made available on Microsoft Azure and Amazon Sagemaker. 

The Command R models, known for their scalability, multilingual capabilities across 10 languages, and optimization, now offer a broader range of AI tools for Amazon Bedrock users. This expansion into AWS’ AI development platform allows Cohere to potentially increase its market share and revenue in the AI sector.

VENTUREBEAT

🧠RESEARCH

Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Models

This study explores evaluating large language models (LLMs) by replacing a single large model judge with a panel of diverse, smaller models (PoLL). Testing across six datasets showed that this panel method reduces bias, improves accuracy, and cuts costs significantly compared to using one large model.

Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting

The paper introduces "Kangaroo," a novel framework for speeding up the processing of large language models without losing effectiveness. It employs a smaller, fixed sub-network for initial predictions and integrates an early exiting mechanism to minimize processing steps when confidence in a prediction is low. This method outperforms existing models, offering significant speed improvements with far fewer parameters.

LEGENT: Open Platform for Embodied Agents

The paper introduces LEGENT, an open and scalable platform for developing embodied agents that use large language and multimodal models. LEGENT enhances the integration of these models into realistic, interactive 3D environments, facilitating advanced simulations for embodied tasks. The platform demonstrates significant improvements in task performance over current models, exemplified by its superior performance compared to GPT-4V in experimental settings.

BlenderAlchemy: Editing 3D Graphics with Vision-Language Models

The paper introduces BlenderAlchemy, a system that uses Vision-Language Models (VLMs) to automate complex 3D graphics editing tasks in software like Blender. By combining VLMs with image generation models, the system intelligently navigates design action spaces to fulfill user intents based on text and visual references. This approach notably simplifies repetitive tasks like editing materials and adjusting lighting, demonstrating its potential to significantly streamline graphics design workflows.

Capabilities of Gemini Models in Medicine

The paper discusses the capabilities of Gemini models in medicine, specifically through the Med-Gemini model. Med-Gemini, a multimodal AI specialized in medical applications, achieves superior performance on medical benchmarks, improving on existing methods and even surpassing human experts in tasks like medical text summarization. The model excels in long-context reasoning and integrates web search, showing promising potential for medical research, education, and dialogue. However, the authors note that further extensive testing is essential before these AI systems can be deployed in real-world medical settings.

🛠️TOP TOOLS

BuddyBreeds - A dog directory built with AI-created images and content

Packify - Packaging design with AI 

Aqua Voice - Dictate, Edit, and Transform using Natural Language.

Fluently - language coach providing instant feedback on calls to enhance pronunciation, grammar, and vocabulary.

CreativAI - Create amazing blog entries, attention-grabbing advertisements, interesting emails, and super cool social media posts in just a few seconds

📲SOCIAL MEDIA

🗞️MORE NEWS

ChatGPT’s AI ‘memory’ can remember the preferences of paying customers

OpenAI has upgraded ChatGPT with a "Memory" feature for its paying subscribers, enhancing its personal assistant capabilities. This allows the AI to remember user preferences and apply them in future interactions without reminders. Users maintain control over what ChatGPT retains, ensuring their privacy. THE VERGE

​​​Sam's Club Deploys AI-Powered Exit Technology at 120+ Locations, Leading Retail Industry With Customer Experience Innovation

Sam's Club has launched AI-powered exit technology at over 120 locations, significantly speeding up the member checkout process by eliminating the need for receipt checks. This technology, which uses computer vision to verify purchases automatically, has improved exit times by 23% and is part of Sam's Club's broader initiative to enhance customer experience through digital innovations like the Scan & Go app. Sam's Club aims to implement this technology in all its clubs by the end of 2024. WALMART

Yelp’s Assistant AI bot will do all the talking to help users find service providers

Yelp is launching an AI-powered chatbot called Yelp Assistant that helps users connect with local service providers for tasks like home improvements and event planning. The chatbot, using OpenAI technology, can generate and send service requests after understanding user needs through interactive queries. This innovation aims to streamline user interactions and provide more accurate service quotes. Initially available on iOS, the Android version will follow this summer. THE VERGE

AI that determines risk of death helps save lives in hospital trial

An AI system trained on electrocardiogram data can predict and reduce death risks, significantly decreasing mortality rates by 31% among high-risk patients in a clinical trial involving almost 16,000 participants at two hospitals. This groundbreaking use of AI in healthcare shows immense potential to enhance patient monitoring and treatment, achieving outcomes seldom seen with traditional medical interventions. NEWSCIENTIST

FDA clears first fully autonomous AI for portable diabetic retinopathy screening

The FDA has approved AEYE Health's AI system for the first fully autonomous diabetic retinopathy screening using a portable device. This AI can analyze retinal images to diagnose conditions that can lead to blindness, improving accessibility and preventing vision loss in diabetic patients worldwide. PR NEWSWIRE

Spire Global strikes mystery deal to boost AI in forecasting

Spire Global, a satellite data company, has secured a multimillion-dollar deal with an undisclosed financial firm to enhance weather forecasting using AI. Spire's new AI model, developed with Nvidia, leverages high-resolution data from space for advanced, long-range forecasts. The deal highlights the secretive nature of AI advancements in weather technology, with Spire keeping specifics under wraps while indicating significant industry progress. AXIOS