Welcome back!

Microsoft isn't just collaborating with Open AI, now it's also creating its own super-sized AI model, MAI-1. Led by former Google AI chief Mustafa Suleyman, this model has a massive scale and ambitious goals, and could signal new power dynamics in AI development. Let's unpack this...

Today’s Summary:

• Microsoft challenges with MAI-1

• Apple's M4 AI chip

• OpenAI enhances creator control

• 3 new tools

TOP STORY

Microsoft Working on Massive AI Model to Challenge OpenAI and Google

The Summary: Microsoft is developing a new, large-scale AI language model called MAI-1, led by former Google AI chief Mustafa Suleyman. With around 500 billion parameters, MAI-1 aims to compete with state-of-the-art models like GPT-4 from OpenAI and Google's Gemini.

The tech giant may preview MAI-1 at its upcoming Build developer conference later this month.

Key details:

• Microsoft acquired most of Inflection AI staff and IP for $650 million in March.

• The company is allocating a large cluster of Nvidia GPUs for training.

Why it matters: The development of MAI-1 signals Microsoft's intent to compete head-on with other AI giants in the realm of large language models. Despite its multi-billion dollar investment in OpenAI, Microsoft is also developing its own in-house AI capabilities. This move could further speed up innovation in the field.

APPLE

The new Apple M4 Chip is an AI Powerhouse

The Summary: Apple unveiled its M4 chip, designed from the ground up for AI tasks. The M4 boasts an “outrageously powerful” Neural Engine outperforming the neural processing unit of any AI PC NPUs today (though high-end Nvidia GPUs are still faster).

The chip delivers up to 1.5x faster CPU and 4x faster GPU performance over M2. With its 3nm process and display engine, M4 enables already the iPad Pro and will be integrated in the laptop and desktop product line in the coming months.

Key details:

• The first M4 version features a 10-core CPU and 10-core GPU with hardware-accelerated ray tracing and mesh shading.

• The 16-core NPU is capable of 38 trillion operations per second for AI workloads.

• Up to 50% faster CPU, 4x faster GPU performance over M2. 120GB/s memory bandwidth, up from 100GB/s in M2.

Why it matters: Apple's M4 chip demonstrates Apple continued leadership in custom silicon design, enabling superior performance and capabilities in a power-efficient device. With its AI prowess, M4 opens up new possibilities for on-device AI, enhancing pro workflows and user experiences.

OPENAI

OpenAI Unveils 'Media Manager' for Creator Control & Provenance Tools

The Summary: OpenAI announced it is developing Media Manager, a tool allowing creators to specify if their works can be included or excluded from AI training data by 2025.

The company also joined the Coalition for Content Provenance and Authenticity (C2PA) to contribute to standards for digital content certification. Additionally, OpenAI released an image detection classifier to identify AI-generated images and introduced watermarking for audio originating from its tools.

Key details:

• Media Manager will let creators control use of their copyrighted text, images, audio, and video for AI training.

• OpenAI joined C2PA steering committee to develop standards verifying how digital content was created/edited.

• An image classifier detecting DALL-E 3 generated images with 98% accuracy was released to researchers.

• Audio watermarking has been incorporated into OpenAI's Voice Engine model for transparency.

• C2PA metadata is being added to DALL-E 3 and upcoming video model Sora's outputs.

Why it matters: These initiatives demonstrate OpenAI efforts to balance accessing training data with respecting creators' rights as AI-generated content grows. Provenance standards and tools can promote transparency and trust as generative AI content becomes more frequent.

TOOLS

🥇 New tools

• Eraser AI - The first copilot for technical design

• QRev - Open Source AI-first alternative to Salesforce

• Touch AI - AI email for sales