Microsoft Unveils New Phi-Silica

Today:

• Microsoft Unveils New Phi-Silica

• Khan Academy, Microsoft Join Forces

• Meta Unveils Chameleon AI Model

• Google's AI Search Includes Ads

• Suno Raises $125M for AI

• Scale AI Hits $14B Valuation

• Europe's AI Laws Set Benchmark

Microsoft introduces Phi-Silica, a 3.3B parameter model made for Copilot+ PC NPUs

Microsoft introduced Phi-Silica, a small language model with 3.3 billion parameters, at its Build conference. Designed for Copilot+ PCs, Phi-Silica leverages powerful Neural Processing Units (NPUs) for efficient performance. It boasts a first token latency of 650 tokens per second and low power usage, enhancing local AI capabilities without burdening the CPU or GPU. 

Phi-Silica is part of the Phi-3 series, which includes models with up to 14 billion parameters. This new model aims to improve productivity and accessibility in the Windows ecosystem, marking a significant advancement in AI for personal computers.

VENTUREBEAT

Khan Academy and Microsoft partner to expand access to AI tools that personalize teaching and help make learning fun

Microsoft and Khan Academy announced a partnership to bring AI tools to teachers, enhancing classroom experiences and easing administrative tasks. Using Azure AI, Khanmigo for Teachers will be available for free to all K-12 educators in the U.S. This AI assistant helps create engaging lesson plans, offers real-time tutoring, and customizes content for diverse student needs. 

The partnership also aims to develop new small language models for scalable, affordable math tutoring. Teachers report significant time savings and improved student engagement, making education more accessible and effective.

MICROSOFT

Meta introduces Chameleon, a state-of-the-art multimodal model

Meta introduced Chameleon, a new multimodal AI model designed to integrate text, images, code, and more. Unlike traditional models that combine separate components, Chameleon uses an "early-fusion" approach, transforming all inputs into unified tokens. This architecture allows it to seamlessly handle and generate interleaved text and images. 

In experiments, Chameleon demonstrated state-of-the-art performance in tasks like image captioning and visual question answering, while also remaining competitive in text-only tasks. Meta aims to offer Chameleon as an open alternative to private models, potentially inspiring advancements in AI applications, including robotics.

VENTUREBEAT

Google Search’s New AI Overviews Will Soon Have Ads

Google is testing the inclusion of ads in its new AI-generated search answers, called AI Overviews. These summaries, designed to retain users who might switch to alternatives like ChatGPT, will feature ads labeled as "sponsored." The ads will be integrated from existing campaigns, requiring no additional input from advertisers. 

This move aims to adapt Google's dominant search advertising model to the era of generative AI. While the exact rollout timeline is unclear, the initiative reflects Google's strategy to maintain its revenue from search ads while leveraging AI to enhance user experience.

WIRED

As AI Race Heats Up, A ChatGPT For Music Just Raised $125 Million

Suno, an AI music generator capable of creating songs with both music and vocals, has raised $125 million in funding. This tool, similar to ChatGPT, allows users to input phrases and keywords to produce AI-generated tracks. Suno aims to democratize music creation, making it accessible for everyone. The funding, led by Lightspeed Venture Partners and others, will be used to accelerate product development and expand their team. 

While the rise of AI in music sparks concerns about unauthorized use of artists' work, many musicians are also exploring AI's creative possibilities.

ROLLING STONE

Scale AI secures $1B funding at $14B valuation as its CEO predicts big revenue growth and profitability by year-end

Scale AI has secured $1 billion in funding, raising its valuation to $14 billion. The Series F round was led by Accel, with participation from investors like Wellington Management, Y Combinator, and Amazon. CEO Alexandr Wang revealed that the company’s annual recurring revenue is expected to hit $1.4 billion by the end of 2024, driven largely by demand for generative AI. 

Scale AI, which helps companies label and test data for AI model training, aims to be profitable by year-end. The company plans to use the funds to enhance its data services and expand its team.

FORTUNE

Europe sets benchmark for rest of the world with landmark AI laws

Europe's landmark AI Act will come into effect next month after being endorsed by EU countries, setting a global standard for AI regulation. The AI Act emphasizes trust, transparency, and accountability, imposing strict transparency requirements on high-risk AI systems and lighter ones on general-purpose AI models. It restricts government use of real-time biometric surveillance to specific serious crimes. 

The Act's global reach will require non-EU companies using EU customer data to comply, influencing global AI regulation much like the GDPR. The legislation will be implemented in phases, with fines for violations ranging from €7.5 million to €35 million.

REUTERS