AI News

The rapid deployment of AI chatbots exposes companies to insider threats. Research highlights the vulnerability of large language models (LLMs) to ASCII art attacks that bypass security. As businesses integrate generative AI, robust defenses and security measures are essential to mitigate risks and protect sensitive data.

President Biden's new executive order on AI mandates federal agencies to protect citizen rights, ensure transparency in AI systems, and establish internal oversight. The policy aims to mitigate algorithmic bias while leveraging AI for public service improvements, amidst ongoing regulatory efforts.

Google is expanding its Circle to Search tool on Android, adding a translation feature that allows users to translate text easily without drawing. Updates to Google Maps include AI-generated summaries for businesses and a new lists feature for personalized recommendations. These updates will roll out for select devices soon.

Intel announced that Microsoft's Copilot AI will shift to local PC processing, requiring next-gen PCs to have NPUs delivering over 40 TOPS, surpassing current processors. This change aims to reduce lag, enhance performance, and increase privacy, with Intel outlining a strategic roadmap for AI PC development.

Stanford researchers and Notbad AI, Inc. introduced Quiet-STaR, enhancing AI reasoning by mimicking human thought processes. Implemented on the Mistral 7B model, it improved question-answering and math problem-solving accuracy. Quiet-STaR enables iterative learning and general reasoning, bridging the gap between AI and human cognition.

ValidMind secured $8.1 million in seed funding, led by Point72 Ventures, to enhance AI model risk management for banks facing compliance challenges. The company aims to automate manual processes, boosting efficiency and enabling AI adoption while aligning with evolving regulations.

Enterprises are heavily investing in AI and ML tools, with transactions increasing by nearly 600% from April 2023 to January 2024. To combat rising cyber threats, organizations blocked 18.5% of these transactions. Key sectors include manufacturing and finance, with growing AI-related risks requiring robust cybersecurity strategies and risk management.

Amazon has invested $4 billion in Anthropic, enhancing its collaboration in AI development. This funding strengthens AWS's offerings, particularly with Anthropic's Claude 3 models, positioning Amazon competitively against Microsoft Azure in the AI cloud services market.

Observe, a San Mateo startup, raised $115 million in Series B funding led by Sutter Hill Ventures. The investment aims to enhance its data observability platform using Snowflake's data cloud, enabling better incident management. The company reported significant revenue growth and plans to expand its development and sales teams.

MyShell, a generative AI platform for creating chatbots and "AI girlfriends," recently secured $11 million in funding, raising its total to $16.6 million. Users can interact with and create personalized bots using various AI models, including its own, ShellLLM. The platform features a blockchain token system for monetization, encourages creative engagement, and supports diverse chatbot uses beyond personal companionship. MyShell aims to enhance its open-source model and foster community involvement.

MLCommons has launched MLPerf 4.0, a key benchmark for AI inference, reflecting advancements in AI from major companies like Nvidia and Intel. The update introduces new metrics and models, emphasizing Nvidia's lead in performance and Intel's improved Xeon CPUs, aiding informed AI solution choices for enterprises.

The Ray framework, crucial for scaling AI workloads, has a critical security flaw dubbed "ShadowRay," allowing attackers to access sensitive data since it lacks formal patching. This vulnerability raises concerns for organizations using Ray, including major companies like Amazon and OpenAI, highlighting the need for better security practices.

Indico Data has introduced an innovative platform that allows businesses to assess and share the document understanding capabilities of their large language models (LLMs). Many current benchmarks for language models focus predominantly on chatbot-related tasks. In response, Indico has developed a solution that enables businesses and developers to evaluate how effectively their AI models extract data from various inputs, as well as perform crucial tasks such as summarizing and classifying documents.

Adobe has launched Photoshop on the web, allowing users to edit and share projects without installation. New AI tools, including generative fill and expand, enhance usability. The updated interface simplifies navigation for beginners, though some features will be added later.

Stanford University's DeepSolar AI system has mapped 1.47 million solar panels in the U.S. using satellite imagery, identifying installations rapidly and accurately. It highlights regional solar adoption trends and the impact of socioeconomic factors, though it faces challenges with data freshness and resource demands.