🚀AI Updates: iOS 18 Beta & New Claude App

Meet Codestral Mamba 7B

Mistral has just released the Codestral Mamba 7B, a groundbreaking Code LLM built on the innovative Mamba2 architecture.

This model, available under the Apache 2.0 license, achieves an impressive 75% on HumanEval for Python Coding.

Alongside it, Mistral has also launched a math-focused model, Mathstral 7B, which scores 56.6% on MATH and 63.47% on MMLU.

What makes Codestral Mamba 7B stand out?

• Efficiency and Performance: Unlike traditional Transformer models, the Mamba architecture offers linear time inference, allowing it to handle sequences of infinite length efficiently. This makes it ideal for extensive code productivity applications, delivering quick responses regardless of input length.

• Advanced Capabilities: Trained with advanced code and reasoning capabilities, Codestral Mamba 7B is designed to perform on par with state-of-the-art transformer-based models, making it a formidable tool for developers.

• In-Context Retrieval: The model has been tested for in-context retrieval capabilities up to 256k tokens, significantly enhancing its utility as a local code assistant.

• Deployment Flexibility: Developers can deploy Codestral Mamba using the mistral-inference SDK or through TensorRT-LLM. Support for local inference will be available in llama.cpp, with raw weights downloadable from HuggingFace.

For easy testing, Codestral Mamba is available on Mistral’s la Plateforme, alongside its larger counterpart, Codestral 22B.

The open-source community can now access, modify, and distribute Codestral Mamba to explore new horizons in architecture research.

Anthropic Releases Claude App for Android

Anthropic has released a new Android app for its AI assistant, Claude, which is now available for free on Google Play.

The app features Anthropic's advanced AI model, Claude 3.5 Sonnet, and offers the same functionality as its iOS and web versions.

Key features include multi-platform support, vision capabilities for image analysis, real-time language translation, and advanced reasoning for complex tasks.

The app is designed for versatile use, suitable for both work and personal tasks. This release follows the iOS app launch in May and aims to make Claude more accessible, competing with other AI chatbots like OpenAI's ChatGPT.

Despite lacking some features like image generation and voice interaction, the latest Claude model is noted for its speed and capabilities in writing, editing, and executing code independently.

Apple Releases iOS 18 Public Beta for iPhone

Apple has released the first public beta of iOS 18, allowing iPhone users to test new features such as home screen customization, a redesigned Control Center, Safari improvements, enhanced Apple Maps, updates to the Journal app, a new Passwords app, and additional features in Messages.

To install the beta, back up your data, enroll your device on Apple's Beta Software Program website, download the beta profile, and install it via Settings.

Note that the beta may contain bugs and is best used on a secondary device. The final version is expected this fall.

AI-Equipped, Energy-Efficient Air Conditioners Gain Popularity in Japan

In Japan, AI-equipped, energy-efficient air conditioners are becoming increasingly popular.

Leading brands like Mitsubishi Electric and Panasonic are at the forefront of this trend, offering models that use AI to enhance comfort and reduce energy usage.

These advanced systems predict temperature changes, optimize operations based on building properties, and adjust settings according to the user's mood.

Some units include sensors that measure occupants' pulses and connect to solar power systems, further maximizing energy efficiency and significantly reducing power consumption.

Bring Static Images to Life with Magic

Ever wondered how we can achieve seamless and temporally consistent human image animation? It's time to bring static images to life with the magic of diffusion models!

The best part? MagicAnimate is open source, and you can try it yourself!

MagicAnimate integrates temporal attention blocks into the diffusion model, ensuring that temporal information is encoded and maintained throughout the animation. It also introduces a novel appearance encoder to retain the intricate details of the reference image, resulting in smooth transitions and high animation fidelity. 

Project page is here.

New Merch Drop - Head Of The Class

Show your support for the AI movement with our new merch. Check out our new design —Head Of The Class — and more at shopgenai.works

5 Best AI Voice Automation Tools

🎬 Descript

📹 DupDub

🧠DeepBrain

🏛️ Colossyan

🎞️ Vyond