OpenAI hit by Security Breach in 2023- never bothered to report
++ Amid Tech Advances, Waymo Self-Driving Incident and Korean Robot 'Suicide' Stir Controversy, SenseTime and Odyssey AI Unveil Game-Changing Technologies, Amazon Spurs AI Security in Academia, and Mo
Today's highlights:
🚀 AI Breakthroughs
SenseTime Launches Advanced AI Models at World AI Conference
• SenseTime unveiled SenseNova 5o, China's first real-time multimodal AI model, at the World Artificial Intelligence Conference
• The updated SenseNova 5.5 language model shows a 30% improvement, boasting enhanced capabilities in English, Mathematical Reasoning, and Prompt Following
• SenseTime's Vimi AI avatar video generator in SenseNova 5.5 can create up to one-minute video clips from a single photo with customizable features.
Odyssey AI Unveiled as Hollywood-Grade Visual Control System for Enhanced Storytelling
• Odyssey introduces a Hollywood-grade AI system aiming to revolutionize storytelling in films, TV shows, and video games with advanced visual controls
• Compared to competitors like OpenAI's Sore and Google's Veo, Odyssey offers enhanced customization, allowing users to direct detailed shapes, materials, lighting, and motion
• Behind Odyssey is a team comprising Hollywood artists and AI experts from industries like Cruise and Tesla, backed by a $9M investment from Y-Combinator and Google Ventures.
Amazon Launches Global University Challenge to Enhance Security in Generative AI
• Amazon's Trusted AI Challenge to enhance LLM security in AI-assisted software tools through a global university competition
• Contest structure includes tournament-style challenges where teams develop or test AI models for coding security, ending in live finals
• Up to $950,000 in prizes and sponsorships, aiming to push forward secure, trustworthy generative AI development.
Robotic Cardiac Surgery Enhances Recovery and Slashes Costs at KFSH&RC
• Robotic surgeries at King Faisal Specialist Hospital & Research Centre KFSH&RC have decreased blood transfusions and mechanical ventilation times, boosting recovery and reducing complications
• These procedures cut hospital stays by over 50%, significantly lowering patient recovery time
• Overall costs have been reduced by 40% due to shorter hospital stays and more efficient surgery methods.
Med42 AI Assistant Advances with New Models to Enhance Healthcare Decision-Making
• M42 releases Med42 v2 8B and Med42 v2 70B, advanced AI models for healthcare, on Hugging Face
• Built on Meta's LLaMA-3, these models provide high-quality medical question-answering capabilities
• Open-access deployment aims to enhance global medical knowledge and healthcare response accuracy.
⚖️ AI Ethics
AI-Driven Hyper-Personalization Set to Transform Chronic Disease Management and Healthcare Spending
• Chronic diseases dominate U.S. health spending, absorbing 90% of the annual $4.1 trillion costs, with projected increases posing sustainability challenges.
• Thrive AI Health collaborates with OpenAI and Thrive Global to develop a personalized AI health coach aimed at promoting healthier daily habits through a mobile app.
• AI's role extends beyond efficiency, potentially enhancing health spans and lifespans by personalizing behavior change and leveraging vast data analytics capabilities.
Waymo Self-Driving Car Incident: Vehicle Pulled Over by Police in Phoenix
• A self-driving Waymo car was pulled over by police in Phoenix for running a red light and entering an oncoming lane
• Bodycam footage revealed the vehicle weaving before stopping in a parking lot, with no driver present to receive a citation
• Waymo stated the incident, lasting about one minute with no passengers aboard, was caused by unclear construction signage.
South Korea Investigates Alleged Robot 'Suicide' After Unusual Incident
• South Korea witnessed a peculiar incident where a robot, employed by Gumi City Council, was deemed to have committed "suicide" after falling down stairs
• The robot, known as 'Robot Supervisor', displayed unusual behavior shortly before the incident, circling in one spot atop a building, then plummeting
• Despite the robot's sophisticated abilities to navigate floors and operate autonomously, local authorities have no immediate plans to replace the machine following its destruction.
OpenAI Confirms Internal Breach, Hacker Accessed AI Design Discussions
• Early last year, a hacker infiltrated OpenAI's internal messaging system, stealing details on AI tech designs
• The breach was disclosed to OpenAI employees and board members in April 2023 but was not made public
• Executives assessed the hack as non-threatening to national security, given the hacker's lack of ties to foreign governments.
🎓AI Academia
LivePortrait: New AI Tech for Efficient, Controlled Portrait Animation Unveiled
• LivePortrait, developed by Kuaishou Technology and affiliated universities, enables efficient animation of static portraits with enhanced control over facial features
• This new AI technology allows for precise eye and lip movement adjustments, making animations more lifelike and expressive
• The framework achieves a generation speed of 12.8ms on an RTX 4090 GPU, demonstrating high efficiency and performance in portrait animation.
Study Reveals Flaws in Code Generation by Large Language Models
• A study revealed that large language models (LLMs) often generate code that is shorter but more complex compared to standard solutions
• Researchers developed a detailed bug taxonomy specifically for LLM-generated code, identifying 3 major categories with 12 sub-categories
• A newly proposed iterative method showed a 29.2% improvement in bug resolution and code passing rates without additional training.
LLMBox: A Novel Comprehensive Library for Streamlining Large Language Model Research
• LLMBox is a new library designed to streamline the development, use, and assessment of large language models (LLMs)
• This tool features a unified data interface supporting different training strategies, enhancing flexibility and efficiency in LLM deployment
• It includes a comprehensive evaluation framework covering 18 downstream tasks and 56 datasets to facilitate detailed performance comparisons.
A Blueprint for Auditing Generative AI: Ethical and Social Perspectives
• A new blueprint for auditing generative AI systems proposes a structured, three-layered approach to enhance governance, model integrity, and application safety
• This auditing framework aims to tackle ethical and social risks by integrating audits at multiple levels, ensuring comprehensive oversight of AI technologies
• Despite the innovative approach, the blueprint also discusses the inherent limitations of auditing generative AI, stressing realistic expectations and outcomes.
Survey Reviews Large Language Models' Impact on Healthcare Efficiency and Patient Care
• Large Language Models (LLMs) are transforming healthcare by enhancing decision-making and improving patient care through advanced data analysis
• A comprehensive review indicates LLMs in medicine effectively manage vast amounts of clinical data, offering significant support to healthcare professionals
• Challenges remain in the adoption of LLMs within the sector, primarily due to issues related to data complexity and model interpretability.
About us: We are dedicated to reducing Generative AI anxiety among tech enthusiasts by providing timely, well-structured, and concise updates on the latest developments in Generative AI through our AI-driven news platform, ABCP - Anybody Can Prompt!
