The resignation of OpenAI CEO Sam Altman on November 18th has caused a stir in the global AI community. ATYUN has compiled a timeline of events from start to finish, preserving all HTML tags.

Microsoft officially introduces Cocreator image generation AI feature in Paint for Windows 11, powered by OpenAI's DALL-E 3 model. Users can now generate images from text descriptions. The feature is also available in other Microsoft services and Copilot AI assistant. Copilot integration across Microsoft products can be overwhelming, but Cocreator and Paint integration seems deliberate, aiming to become the new clip art.

A new research paper reveals that large language models could unintentionally expose terabytes of training data through "extractable memory." Researchers extracted data from models like ChatGPT and found potential privacy risks in widely used systems. AI models are exploring encryption to protect training data.

NVIDIA and AWS announced strategic initiatives in AI supercomputing. NVIDIA's GH200 chip will be deployed in AWS's cloud infrastructure. Collaboration with Microsoft, Google, and Oracle further expands NVIDIA's AI reach. NVIDIA's diverse partnerships enhance its position in the AI field across various aspects.

At AWS re:Invent, AWS launched Model Evaluation on Bedrock for better model evaluation. Automated and manual evaluation help choose the right model. AI models like Llama 2, Claude 2 are included. Customized pricing and schedules for manual evaluation. Benchmark testing service in preview, charges only for model inference used. Aim is to measure model impact on projects, guide development.

Confirmation bias is a feature of human decision-making replicated in AI models. RIKEN team published results in Nature Comm., showing overconfidence rooted in subtle cues. Despite evidence, familiarity leads to unwarranted certainty. AI model exhibited similar bias, learning from natural image noise.

Dell signs $150 million contract with AI startup Imbue for server services. Imbue stands out by independently building AI models. Dell emphasizes server benefits for AI field. Imbue values flexibility with Dell hardware purchase. Collaboration enables AI system experimentation and foundational model development. Imbue develops autonomous AI agents for various tasks. Dell partnership offers alternative to public cloud providers for AI startups.

Bill Gates predicts a 3-day work week with AI advancement. "Silent resignation" trend emerges as AI takes over tasks. Debate on future work hours intensifies. People hope AI will do the work while they enjoy life. Universal Basic Income concept gains traction.

The Advanced Technology Research Council (ATRC) launched AI71 in Abu Dhabi to rival top AI labs. AI71, a result of TII's AI research, focuses on marketing Falcon LLM. ATRC's VentureOne played a key role, solidifying Abu Dhabi's AI competitiveness. UAE drives AI innovation with TII and G42 launching LLM domestically and Core42 introducing Jais 30B for Arabic. UAE aims for a leading AI position with continuous developments.

POLIGON, the operator of Helo, launched "ChitChop", an AI assistant tool with 200+ services, similar to Douyin Group's "Xiao Wukong". Despite occasional inaccuracies, it shows strong practicality. The global AI large-scale model market is projected to hit $21 billion by 2023. ByteDance's Resso, CapCut, and Lemon8 have performed well overseas. The trend of large-scale model products going global is growing, with companies like Huawei Cloud, Alibaba, and Baidu expanding overseas.

RunwayML introduces new features like Motion Brush for enhanced creativity and Director Mode updates for finer camera control. It also improves image model fidelity and merges motion and camera control for higher precision. Its collaboration with Canva integrates generative AI videos into Canva's ecosystem.

During AWS re:Invent, NVIDIA and AWS partnered on cloud GPUs. AWS introduced AI chatbots, new AI chips, and Amazon Bedrock features. Salesforce products are now on AWS Marketplace. AWS and Salesforce aim to enhance data access and generative AI tech.量表 for preview. New AWS data integration Zero-ETL options include DynamoDB with OpenSearch. Amazon One Enterprise provides secure physical access via biometric authentication. NVIDIA launched new GPUs on AWS and announced NeMo Retriever for accurate

Stability AI introduces SDXL Turbo, a text-to-image model using Adversarial Diffusion Distillation. It generates high-quality images in real-time from short text prompts, outperforming other diffusion models. SDXL Turbo is fast, but has limitations like fixed resolution and occasional rendering issues. Available for free on Clipdrop for non-commercial research.

The article discusses DeBERTa, a model introducing "decoupled attention" in Transformer architecture."Decoupled attention" improves attention calculation by considering the relationship between content and position of tokens. The model also includes an "Enhanced Masked Decoder" and "Scale-Invariant Fine-Tuning," leading to significant performance gains.

Netweb partners with NVIDIA to manufacture cutting-edge AI systems, aligning with "Make in India" initiative. AI systems based on NVIDIA MGX platform offer multifunctional solutions for enterprises, driving AI adoption across industries in India and Asia-Pacific region.

Amazon Web Services (AWS) enhances Amazon Transcribe with advanced speech models for improved accuracy in over 100 languages. Features include generative call summaries, automatic punctuation, and custom vocabulary filtering, benefiting various industries. No migration required for existing customers.

AMD predicts MI300X will generate $2B revenue in 2024. It is expected to be the fastest-growing product in AMD's history. MI300X will be utilized for AI workloads by Microsoft. AMD's ROCm software stack has reached production-level performance for high-performance computing.

Amazon Web Services (AWS) launched Amazon WorkSpaces Thin Client for secure cloud-hosted virtual desktops. It addresses device security concerns and offers enhanced functionality, user experience, and cost-effectiveness. Managed via AWS Management Console, it boosts appeal for their virtual desktop products.

Significant international agreement aims to safeguard AI systems, prioritize security and public safety. Not legally binding, but emphasizes monitoring AI for abuse prevention. Key for ethical AI development. Signed by 18 countries, highlighting the global importance of AI security. Future regulations crucial for safety, ethics, and public welfare.

Large language models struggle with long sequences. StreamingLLM, a breakthrough technology, expands context to millions of tokens without high resource demands. Attention sinks play a crucial role in LLM stability. StreamingLLM retains initial tokens to maintain performance. Sink tokens can improve attention allocation in pretrained models. StreamingLLM's code is available on GitHub and Hugging Face.