Microsoft’s Designer App Brings AI-Powered Design to Mobile Users
Microsoft’s Designer app, an AI-powered design tool, is now available on iOS and Android, offering users the ability to create various designs using text prompts. The app, which rivals Canva, supports over 80 languages and allows users to generate stickers, greeting cards, emojis, and more. Designer features prompt templates to aid creativity and will soon include a background replacement feature. Microsoft is also integrating Designer into Word and PowerPoint through Copilot, and enhancing Windows 11’s Photos app with AI editing capabilities.
Salesforce Introduces AI-Powered Customer Service Agent
Salesforce unveils Einstein Service Agent, a generative AI-driven self-service tool for customer interactions. This new agent can handle complex, multi-step processes, including product returns and refunds, by leveraging existing Salesforce workflows. Unlike the employee-facing Einstein Copilot, this service is designed for end-customers. The key advantage is its seamless integration with Salesforce’s customer data and approved business processes. Initially focused on customer service, Salesforce plans to expand the agent’s capabilities to other areas in the future.
Google Unveils Project Oscar: AI Agents for Open-Source Development
Google has introduced Project Oscar, an open-source platform for creating AI agents to manage software development lifecycles. Announced at Google I/O Bengaluru, the platform aims to assist development teams in monitoring issues and bugs. Currently focused on open-source projects, Oscar allows developers to create various AI agents that can interact using natural language. The Go programming language team is already using Oscar to manage bug reports and contributor engagements. Google plans to expand Oscar’s deployment to other open-source projects and potentially closed-source projects in the future, aiming to make AI more accessible and helpful for developers.
OpenAI Develops Game-Based Method to Improve AI Explanation Skills
OpenAI researchers have introduced a new algorithm to enhance the ability of large language models (LLMs) to explain their reasoning. Based on the “Prover-Verifier Game,” the method pairs two AI models—a more powerful “prover” and a less capable “verifier”—to challenge each other. Through multiple rounds of gameplay, the models improve their ability to provide and evaluate explanations. This approach aims to increase the trustworthiness and transparency of AI systems, particularly in critical fields like healthcare and law. The research demonstrates potential for aligning future AI models that may surpass human intelligence, addressing the crucial “legibility” problem in AI development.
Deepfake Detection Firm Pindrop Secures $100M Loan for Expansion
Pindrop, a company specializing in deepfake detection and multi-factor authentication, has obtained a $100 million, five-year loan from Hercules Capital. The funding will be used for product development and hiring as the company aims to combat the rising threat of deepfakes.
The company addresses potential biases in voice recognition by focusing on acoustic and spectro-temporal features rather than pronunciation or language. The loan allows Pindrop to raise capital without diluting equity, positioning the company to meet growing demand for deepfake protection solutions. Pindrop plans to expand its services into new sectors such as healthcare, retail, media, and travel.
Live2Diff AI Enables Real-Time Video Stylization
Live2Diff, a new AI system developed by international researchers, can transform live video streams into stylized content in near real-time. This groundbreaking technology processes live video at 16 frames per second on high-end consumer hardware, marking the first successful implementation of uni-directional attention modeling in video diffusion models for live-stream processing. Live2Diff overcomes limitations of current models that require future frame data, demonstrating its capabilities by transforming live webcam input of human faces into anime-style characters in real-time.
The system outperforms existing methods in temporal smoothness and efficiency, as confirmed by both quantitative metrics and user studies. Potential applications for Live2Diff span various industries, including entertainment (live streaming and virtual events), augmented and virtual reality experiences, content creation, gaming, and virtual tourism.
