Baidu Wenxin Yiyuan App introduces new "Create My Sound" feature for users to customize their exclusive voice assistant in seconds. Enhancing personalized experience and interaction with digital avatar, users can enjoy unique tone for voice broadcasting and real-time conversations.

Alibaba Cloud's Tongyi Qianwen open sources 32B parameter model Qwen1.5-32B, offering a cost-effective choice for enterprises and developers. It has achieved impressive performance balance compared to previous models, showing excellent capabilities in various evaluations. This move enhances Tongyi Qianwen's influence and technical strength in large language models field.

Google AI team launches visual language model ScreenAI, enhancing user experience. ScreenAI can understand UI, graphics, navigate screens, answer questions, and summarize content. It offers potential applications in education, training, and marketing.

Google is adding a "Search" button to its Phone app to help identify unknown callers. It will also update the emergency call screen. Additionally, a Gemini email summary feature is in the works for the Gmail app on Android.

Gretel releases world's largest open-source Text-to-SQL dataset to accelerate AI model training, bridging gap between natural language queries and SQL generation. Dataset addresses data quality challenges, with wide application domains from finance to healthcare. Gretel emphasizes data privacy and accessibility in generating high-quality synthetic data. Release marks a key step in accelerating data-centric AI adoption.

Interacting with cute BD-X robots at Disney parks during the "Season of the Force" celebration, with movements inspired by child-like behavior and trained through reinforcement learning. Apple also reportedly developing AI-driven home robots.

Google Books and Google Scholar now index low-quality, AI-generated books among human-authored works. Some AI-generated books mimic real authors, presenting shallow narratives that resemble Wikipedia articles. This trend raises concerns about the credibility of these platforms.

AI offers benefits in education like personalized learning and efficient administrative tasks. Challenges include job security for teachers and maintaining student authenticity. Solutions involve collaboration with the EU, implementing safeguards, and balancing AI with human interaction in teaching. Ireland is leading in adapting AI in education.

Stability AI introduces Stable Audio 2.0, enabling creation of full-length songs, audio-to-audio generation, and enhanced sound effects. The model uses a diffusion transformer for coherent track production. Collaboration with Audible Magic ensures copyright protection. Free access on website and upcoming API availability.

OpenAI expands Custom Model program, introduces assisted fine-tuning capabilities to meet demands of enterprise customers. SK Telecom sees significant improvements in AI customer service. OpenAI emphasizes the importance of customized AI models for specific industries and businesses.

Higgsfield AI, founded by former Snap AI head Alex Mashrabov, raised $8M for personalized AI video creation. Their app Diffuse lets users create content with custom avatars in various scenes. The mobile-first approach aims to simplify video creation for everyone.

Meta announced an update in handling AI-generated content on its platform after consulting a supervisory committee and reviewing public opinion. Content will be labeled as "AI-generated" to increase transparency and avoid restricting freedom of speech. Consultations with stakeholders and public opinion surveys support this approach. Meta will keep AI-generated content but add informative labels and background information. The policy will be implemented in May 2024.

Apple unveils MobileCLIP, an image-text model series optimized for mobile devices with a hybrid CNN-transformer architecture. Multimodal reinforcement training method boosts learning efficiency. Expect AI-powered user experiences on iPhone with intelligent image search and object detection offline.

Cohere launches Command R+ in AI language models for enterprise workloads, offering enhanced performance in RAG, multilingual coverage, and tool usage. Oracle VP praises the model's productivity-driving capabilities. Command R+ supports 10 languages, prioritizes data privacy, and will be available on various cloud platforms. Price set at $3.00 per million input tokens and $15.00 per million output tokens.

AI companies face challenges in acquiring quality training data. OpenAI transcribes YouTube videos to train GPT-4, facing legal concerns. Google and Meta also struggle with data scarcity, considering various solutions.

Neal Mohan, CEO of YouTube, stated that using videos for training AI models would violate terms. OpenAI's Sora tool raised concerns on data sources including YouTube. OpenAI faces lawsuits for using copyrighted materials in training data.

SiMa.ai secures $70M in funding from Dell, highlighting edge AI growth. Dell's investment shows confidence in SiMa.ai's approach. Generative AI transforms industries with personalized experiences. SiMa.ai's MLSoC tech simplifies edge AI deployment, aligning with Dell's strategy for powerful AI at the edge.

Google Research and Johns Hopkins University researchers explore the efficiency of AI models in image generation tasks, finding that smaller models can outperform larger ones within a fixed budget, with implications for developing more efficient AI systems. This study on Latent Diffusion Models (LDMs) highlights trade-offs between model size and performance, paving the way for creating balance between efficiency and quality in AI systems.

Anthropic Corporation discovered a novel "jailbreaking" technique affecting large language models. The attack, named "multi-round jailbreaking," manipulates models by providing harmful examples. It is effective against popular models and can lead to undesirable outputs. Anthropic's mitigation efforts include fine-tuning models and modifying prompts to reduce the attack's success rate. This research highlights the need for better alignment methods and responsible AI development practices.

OctoAI launches OctoStack for hosting AI models on internal infrastructure, eliminating the need to share data with external vendors. The platform supports various hardware, accelerators, and AI optimizations, promising increased efficiency and reduced operating costs. It automates the optimization process for different chips, supporting popular LLMs like Llama and Mixtral.