Demonstrate secure AI practices with ISO 42001

As AI becomes more ubiquitous, so does the need to prove your AI practices are secure.

ISO 42001 was introduced to fulfill this need. This framework helps companies demonstrate their organization’s security practices around AI, in a verifiable way.

Vanta provides 95% of the document templates required for ISO 42001. Through powerful automation and 300+ integrations, Vanta is a fast and easy way for companies to become ISO 42001 certified and maintain continuous compliance.

Join Vanta and A-LIGN for a live webinar on 8 May on ISO 42001 —what it is, what types of organizations need it, and how it works. (If you can’t make the time, register anyway, and you’ll receive a link to the recording.)

Register here

Today:

• Adobe Premiere Sora AI Integration

• Microsoft and G42 $1.5B AI Deal

• Amazon Maestro’s AI-Powered Playlist

• Baidu's Ernie Hits 200M User 

• Deepmind Commits $100 Billion to AI

• Adobe Launches Acrobat AI for PDFs

Adobe Premiere's STUNNING Integration With OpenAI's SORA | Coming this year?

Adobe Premiere Pro will soon integrate OpenAI's Sora video generation model, a breakthrough in editing technology. The Adobe Firefly video model allows precise object addition and removal with simple text prompts, enhancing editing efficiency. 

Additionally, generative extend intelligently adds frames for seamless footage extension. Adobe prioritizes transparency, introducing content credentials to indicate AI usage. Collaborations with third-party AI models like Pika and Runway expand editing possibilities. 

WATCH THE VIDEO ON YOUTUBE

Microsoft and G42 partner to accelerate AI innovation in UAE and beyond

Microsoft and G42 are teaming up to supercharge AI in the UAE and beyond. Microsoft is investing $1.5 billion in G42, grabbing a slice of the company and a seat on its board. Together, they'll pump $1 billion into an AI skills fund for the UAE and beyond. This move expands their existing partnership to spread advanced AI solutions using Microsoft Azure across various industries. 

With this deal, they're aiming to boost AI innovation in emerging markets, while keeping things secure and above board with governments in the loop. Plus, they're bringing Arabic AI models to Microsoft Azure, pushing for digital transformation in banking and healthcare.

MICROSOFT

Amazon Music’s Maestro lets listeners make AI playlists

Amazon Music is rolling out Maestro, an AI playlist maker currently in beta for select US subscribers. Users can create playlists using text prompts, including emojis, and share them with others. Similar to Spotify's AI Playlist, Maestro builds playlists based on user input. Beta users can access Maestro through the Amazon Music app, where they can type or speak prompts to generate playlists. 

Prime Music subscribers and those using the ad-supported version have limited access, while Amazon Music Unlimited users can save and listen to full playlists immediately. Amazon emphasizes that Maestro is a work in progress, implementing safeguards against offensive language and inappropriate prompts. The company plans to gather feedback before expanding Maestro's availability.

THE VERGE

Baidu Says Ernie AI Chatbot Now Has 200 Million Users

Baidu's Ernie AI chatbot, akin to ChatGPT, has surged in popularity, boasting over 200 million users and 85,000 business clients. Robin Li, CEO of Baidu, announced the milestone at a recent conference, highlighting the significant growth of the Chinese-language chatbot since its public release in August 2023. 

The chatbot's expansion underscores Baidu's foothold in the AI-driven digital landscape, reflecting the widespread adoption of conversational AI technologies in both consumer and enterprise domains.

THE WALL STREET JOURNAL

DeepMind CEO Says Google Will Spend More Than $100 Billion on AI

Google's AI division anticipates spending over $100 billion on artificial intelligence development, signaling the escalating investment competition in Silicon Valley. Demis Hassabis, CEO of Google DeepMind, revealed the ambitious financial commitment during a TED conference in Vancouver. 

This announcement comes in response to reports about Microsoft Corp. and OpenAI's collaboration on a potential $100 billion supercomputer project named "Stargate." The massive investment underscores the intense focus and substantial resources being poured into AI research and development by tech giants, fueling the ongoing AI arms race.

BLOOMBERG

Adobe releases AI assistant that can summarize all those complex PDFs for you

Adobe has launched its Acrobat AI Assistant, aimed at simplifying document management. Timed with tax season and the 25th anniversary of IRS-standardized tax documents on PDF, the tool is available across Acrobat Reader, desktop, and web for a $4.99 monthly subscription. 

With over three trillion PDFs worldwide, Adobe aims to democratize generative AI, offering features like content summarization and easier document navigation. Users can engage in conversational interactions with their PDFs and receive intelligent citations. Adobe assures user data privacy and offers early access pricing until June 5, 2024, with plans for multilingual support and a mobile version.

TECHRADAR

🧠RESEARCH

Learn Your Reference Model for Real Good Alignment

Researchers tackle instability in model alignment methods. They introduce Trust Region DPO (TR-DPO) to improve upon Direct Preference Optimization (DPO). TR-DPO updates the reference policy during training, outperforming DPO by up to 19%, enhancing model quality across various parameters.

Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length

"Megalodon" presents a novel neural architecture for efficient sequence modeling, addressing the limitations of Transformers in scaling to long sequences. The model, based on Mega with gated attention, incorporates innovations like complex exponential moving average, normalized attention, and pre-norm with two-hop residual configuration. It outperforms Transformers in pretraining efficiency and accuracy, showcased in comparisons with Llama2.

TransformerFAM: Feedback attention is working memory

"TransformerFAM" introduces a novel Transformer architecture called Feedback Attention Memory (FAM), addressing the quadratic attention complexity of Transformers. FAM incorporates a feedback loop, enabling the network to attend to its own latent representations, effectively creating a working memory. This design enables processing of indefinitely long sequences without additional weights, enhancing Transformer performance on long-context tasks across various model sizes and showcasing the potential for Large Language Models (LLMs) to handle sequences of unlimited length.

Compression Represents Intelligence Linearly

"Compression Represents Intelligence Linearly" explores the relationship between compression and intelligence in large language models (LLMs). Treating LLMs as data compressors, the study correlates their ability to compress external text corpora with intelligence, measured through downstream benchmark scores. Across 12 benchmarks and 30 LLMs, results demonstrate a nearly linear correlation, suggesting that superior compression indicates greater intelligence. 

Video2Game: Real-time, Interactive, Realistic and Browser-Compatible Environment from a Single Video

"Video2Game" introduces a novel approach to automatically convert real-world videos into interactive game environments. Utilizing a Neural Radiance Fields (NeRF) module to capture scene geometry and appearance, a mesh module for faster rendering, and a physics module for modeling interactions, the system constructs realistic, interactive digital replicas. Benchmarked on various scenes, it demonstrates real-time rendering and interactive gameplay capabilities.

🛠️TOP TOOLS

Morphic - An AI-powered answer engine with a generative UI

OpenParse - Beyond text splitting - improved file parsing for LLM’s

Markprompt - AI infrastructure for customer support

Success AI - Meetings, & Revenue With AI-Powered Cold Emails

Sendspark - Book more meetings, close more deals, and accelerate the sales cycle

📲SOCIAL MEDIA

🗞️MORE NEWS

Anthropic's Claude on Amazon Bedrock

Amazon Bedrock introduces Anthropic's Claude 3 models for generative AI solutions. Claude offers intelligence, speed, and safety features for enterprises. The collaboration between Amazon and Anthropic aims to responsibly deploy AI. Claude's prompt engineering enhances performance across various sectors like customer service, legal, and coding. Claude models cater to different needs, from complex tasks to quick responses. AMAZON

Intel to launch two AI chips for China with reduced capabilities

Intel plans to launch two AI chips tailored for the Chinese market, complying with U.S. export controls. The HL-328 and HL-388 chips, part of Intel's Gaudi 3 line, will have reduced capabilities. Nvidia also follows suit with three China-specific chips amid tightened regulations on AI chip exports to China by the U.S. YAHOO!

AMD is taking on Nvidia and Intel with a new AI computer chip

AMD is upping its game in the AI chip race, unveiling new processors tailored for AI-enabled laptops and desktops. The AMD Ryzen PRO 8040 Series targets business laptops and mobile workstations, while the Ryzen PRO 8000 Series claims to be the first AI-enabled desktop processor for business users. These chips, built with cutting-edge 4-nanometer technology, promise enhanced performance and efficiency for handling AI tasks on personal computers. QUARTZ

Stanford report: AI surpasses humans on several fronts, but costs are soaring

The Stanford University AI Index 2024 report reveals significant advancements in AI in 2023, with AI surpassing humans in various benchmarks. However, costs for training AI models have skyrocketed, reaching unprecedented levels. The United States leads in AI model production, but standardized testing for AI responsibility is lacking. Emerging risks like political deepfakes raise concerns, reflecting growing public apprehension about AI's impact. VENTUREBEAT

Startup Rivos raises $250 million to develop RISC-V AI chips

Chip startup Rivos secures $250 million to develop AI server chip. Competing with Nvidia's dominant AI chips, Rivos plans RISC-V architecture for open source advantage. Focusing on software first, it aims for optimized performance in language models and analytics. Investors include Matrix Capital, Intel, MediaTek, and Dell. YAHOO!

Meta oversight board reviews handling of AI-created celebrity porn

Meta's Oversight Board reviews Meta Platforms' handling of sexually explicit AI-generated images circulating on Facebook and Instagram. Facing challenges in controlling fake content, particularly harmful "deep fakes," the board aims to assess Meta's policies. Industry executives call for legislative measures. Meta pledges to implement the board's decisions. REUTERS

Zendesk unveils AI-powered CX platform with sophisticated agents and intelligent copilots

Zendesk introduces advanced AI agents and copilots to enhance customer experience. With up to 20% of support interactions already automated, Zendesk's AI aims to resolve 80-90% of issues autonomously, backed by improved accuracy and personalized service. The platform prioritizes human-agent collaboration, offering predictive scheduling and AI-guided responses to optimize customer support. VENTUREBEAT

World’s First AI Pageant To Judge Winner On Beauty And Social Media Clout

The world's first AI beauty pageant, "Miss AI," judges AI-generated contestants based on looks, technical skill, and social media influence. Winners receive cash prizes and promotion. AI influencers are gaining traction, with some securing paying gigs. Critics raise concerns about perpetuating narrow beauty standards. FORBES

Zyphra releases Zamba, an SSM-hybrid foundation model to bring AI to more devices

Zyphra Technologies introduces Zamba, a decentralized AI model with 7B parameters, aiming to make AI more accessible. Zamba utilizes Mamba blocks and a shared attention layer for global connectivity. Smaller than competitors, Zamba aims for local device implementation, promising real-time responses and cost-effectiveness. Testing shows outperformance of other models. Available on Hugging Face. VENTUREBEAT