During the earnings call, Apple CEO Tim Cook revealed plans for AI software features in upcoming iOS 18 update. Cook hinted at significant developments without giving specifics, suggesting a major push towards AI integration in Apple platforms later this year.

AI2 introduces OLMo, a transparent large-scale language model with open-source code, data, and training details, addressing the issue of "black box" AI models. OLMo surpasses commercial products in performance and aims to become the best open language model worldwide, driving AI innovation with unprecedented transparency and collaborations.

Synthesia, a London-based startup, launches AI Video Assistant, converting text to videos in minutes. Aimed at improving content delivery, it simplifies video creation process. It has a 4,500-word limit but aims to expand with enterprise growth.

A study found combining digital watermarking with AI speeds up copyright infringement resolution, easing lawsuit filing and providing more evidence. AI usage may lead to more disputes. AI can assist in arbitration, enhancing decision-making consistency and predictability in copyright disputes related to 3D printing.

GPT-4 and other AI systems can learn human language from massive data. NYU researchers trained AI with child's daily experiences, concluding that AI can learn words and concepts effectively with limited data, reshaping understanding of early language acquisition.

Amazon releases AI shopping assistant Rufus, trained on its product catalog and web data. Beta version in US, expanding soon. Offers personalized shopping experience, answers questions, gives recommendations on various products. Available in Amazon app, users can chat, ask questions, and receive advice. Feedback will be used to improve AI technology and responses. Potential to enhance shopping experience but market feedback needed.

SenseTime released the new large-scale model system "SenseNova 4.0" with upgraded knowledge coverage, reasoning, and text comprehension capabilities. It supports cross-modal interaction, deep client cooperation, and flexible APIs for developers. The model outperformed GPT-4 in evaluations and offers improved multimodal and multilingual capabilities for practical applications, including intelligent driving and creative design.

Alibaba introduces MobileAgent, an autonomous AI agent based on image analysis for mobile operations. It enhances flexibility and efficiency without the need for system code, making it ideal for various tasks and easy deployment in enterprises.

OpenAI's study on GPT-4's assistance in creating biological threats found only slight improvements in accuracy. Current LLMs like GPT-4 do not pose substantial risks, but future AI advancements could change this landscape. Ongoing research and regulation are crucial as AI technology evolves.

Dipendra Misra from Microsoft Research explained how LASER optimizes large language models, improving accuracy without sacrificing performance, even increasing accuracy by 20-30% in certain cases. Misra's team successfully applied LASER on open-source models like RoBERTa and GPT-J, showing significant improvements in tasks like gender prediction. Factual errors in AI models remain a concern, as inaccuracies can have harmful impacts.

Google Cloud and Hugging Face collaborate to enhance generative AI and ML development. Utilize Google Cloud's AI-optimized infrastructure for efficient model training, including Vertex AI integration and GKE deployment. Partnership aims to support A3 virtual machines and NVIDIA's H100 Tensor Core GPU. Available features expected by first half of 2024.

Shopify Magic, a generative AI tool suite, offers a new media editor allowing merchants to change photo backgrounds and match themes without Photoshop. The AI editor suggests backgrounds and styles. Other AI tools include chatbot integration and product description generation. Shopify's new semantic search interprets search intent for more relevant results. Shopify combines AI base models like GPT and Llama 2 for features like watermarking and moderation.

In recent days, the open-source AI community saw the emergence of a new large-scale language model "miqu-1-70b" on HuggingFace, similar to Mistral's Mixtral 8x7b. Speculation arose over Mistral's involvement, which was confirmed by Mistral's CEO, promising a competitive model to GPT-4. This development may signify a significant shift in the open-source AI landscape, challenging the dominance of GPT-4 and OpenAI.

Meitu Design Studio launched "AI Video Generation" in beta, allowing quick dynamic video creation from a single image. E-commerce professionals see potential in improving work efficiency and cutting costs. "AI Video Generation" aligns with the growth predicted in AI video creation for 2024, offering new possibilities for product display and marketing in e-commerce.

Meta AI released "Prompt Engineering with Llama 2" for the open-source community. Prompt engineering is a trending profession in 2023, with companies using open-source LLMs like LLaMA. Experts are needed to prompt chatbots effectively, highlighting the importance of prompt customization for optimal results.

Layier AI's LLM Guard enhances security for large language models, protecting against threats like prompt injection attacks. It optimizes CPU performance, integrates with security workflows, and has gained significant adoption. Protect AI aims to further develop LLM Guard for various industries. Protect AI is backed by investors like Evolution Equity Partners and Salesforce Ventures.

ByteDance launched domestic AI chatbot platform Coze (Kouzi), allowing users to build various bots easily. Despite initial issues, it offers diverse functionalities and plugins for customization. Coze, developed by ByteDance's AI department, shows promise for future growth.

CloudFrom Technology launched China's 1st AI-native data analysis tool, DataGPT. It uses a large model for accurate analysis, supports voice/text commands, integrates various data sources securely, provides personalized data services, helps build data visualization centers, and offers intelligent insights for decision-making. DataGPT overcomes challenges of traditional BI, accelerating digital transformation.

Shanghai AI Lab, Tsinghua Univ., CUHK, and SenseTime open-sourced InternVL-6B, a large visual encoder model with 6 billion parameters. It achieves fine alignment between visual and language models, handles complex images, text tasks, interprets page info, solves math problems, and shows promise in various fields.

At the Xunfei Spark Cognitive Model V3.5 upgrade release, iFlytek revealed enhancements in text generation, language understanding, question answering, reasoning, math, coding, and multimodal understanding. Surpassing GPT-4 Turbo in math, coding, and language. Jointly developed "Feixing One" platform with Huawei for high computing efficiency. Future plans to surpass GPT by mid-2024.