Today's highlights:
๐ AI Breakthroughs
Samsung Launches Galaxy Ring, Aiming to Popularize Smart Ring Market
โข Samsung launched its first smart ring, the Galaxy Ring, at its Galaxy Unpacked event, targeting health and fitness enthusiasts
โข The Galaxy Ring features sleep tracking capabilities and integrates with Samsung's AI-powered ecosystem
โข Analysts suggest the move could popularize smart rings, projecting a global market of four million units by 2025.
AMD Acquires Silo AI to Boost Global Enterprise AI Solutions Development
โข AMD acquires Europe's largest private AI lab, Silo AI, for $665 million to boost its AI solution offerings globally
โข The acquisition aims to enhance AMD's open-source AI software for more efficient training and inference on its computing platforms
โข Silo AI specializes in end-to-end AI solutions, impacting diverse markets with clients like Allianz and Philips.
OpenAI Develops New Scale to Track Progress Towards Artificial General Intelligence
โข OpenAI introduces a new scale to chart progress towards AGI, breaking advancements into five distinct levels, starting with technologies like ChatGPT
โข At Level 2, OpenAI's models aim to match human PhD-level problem-solving capabilities, hinting at the potential of their upcoming GPT-5
โข Despite strides towards AGI, substantial challenges remain, including significant computing requirements and unresolved ethical and safety concerns.
OpenAI Develops 'Strawberry': Enhancing AI's Reasoning and Autonomous Research Abilities
โข OpenAI's "Strawberry" project aims to develop AI models capable of conducting autonomous internet-based research to enhance reasoning abilities
โข Dubbed previously as Project Q*, Strawberry focuses on Long-Horizon Tasks (LHT), allowing AI to plan and execute complex activities over extended periods
โข The technology includes a "deep-research" dataset and a novel post-training phase, enhancing the base modelsโ ability to perform sophisticated calculations and reasoning tasks.
Washington Post Launches AI Chatbot for Enhanced Reader Interaction on Climate Issues
โข The Washington Post introduces an AI-powered chatbot to engage readers by answering queries based on its climate reporting from the past eight years
โข This chatbot, an experiment, draws solely on the publicationโs climate content, reinforcing answers with related articles to ensure information accuracy
โข In a broader trend, major news outlets are increasingly integrating generative AI technology, with significant partnerships and legal actions shaping the landscape.
KoBold Metals Discovers Largest Copper Deposit in Over a Decade Using AI
โข KoBold Metals, using AI technology, discovered a major copper deposit in Zambia, potentially the largest in over a decade
โข Estimated annual production of at least 300,000 tons could generate billions in revenue yearly for decades
โข Independent assessments confirm the size of the deposit, with KoBold anticipating further valuable findings.
โ๏ธ AI Ethics
EU's AI Act gets published in bloc's Official Journal, starting clock on legal deadlines
The EU AI Act will come into force on August 1, 2024, with full applicability by mid-2026.
The law takes a phased approach, with different provisions becoming applicable at various times between 2025 and 2027.
Key elements include bans on certain AI uses, transparency requirements for AI chatbots and powerful AI models, and obligations for high-risk AI applications.
New Timeline:
โกย ๐ญ๐ฎ ๐๐๐น๐ ๐ฎ๐ฌ๐ฎ4: EU AI Act published in Official Journal
โก ๐ญ ๐๐๐ด๐๐๐ ๐ฎ๐ฌ๐ฎ๐ฐ: Enters into force
โก ๐ฎ ๐๐ฒ๐ฏ๐ฟ๐๐ฎ๐ฟ๐ ๐ฎ๐ฌ๐ฎ๐ฑ: Chapters I and II apply (prohibited practices)
โก ๐ฎ ๐๐๐ด๐๐๐ ๐ฎ๐ฌ๐ฎ๐ฑ: Rules on general-purpose AI models and governance start
โก ๐ฎ ๐๐๐ด๐๐๐ ๐ฎ๐ฌ๐ฎ๐ฒ: General application begins
โก ๐ฎ ๐๐๐ด๐๐๐ ๐ฎ๐ฌ๐ฎ๐ณ: High-risk AI system classification rules apply
NATO Revamps AI Strategy to Enhance Defense Capabilities and Ensure Responsible Usage
โข NATO's AI Strategy accelerates the adoption of generative AI technologies, focusing on interoperability and responsible use within defense
โข AI's role in defense escalates as NATO updates strategic aims, highlighting the necessity for enhanced AI-ready workforce and standards
โข Aligning with global norms, NATO seeks to safeguard intellectual property and foster alliances with academia and industry, mitigating adversarial AI risks.
EU Claims Elon Musk's X Misleads Users with "Verified" Account Status
โข The European Commission claims Elon Musk's X misleads users by selling blue check marks that once signified verified authenticity now available to any paying subscriber.
โข Violations against the EUโs Digital Services Act by X could lead to fines up to 6% of its global annual revenue for facilitating misinformation.
โข Musk challenges the preliminary findings, alleging the Commission proposed a censored speech deal to other platforms that X refused to accept.
COPIED Act Sets New Guardrails for AI, Backed by Broad Creative Coalition
โข COPIED Act introduced to enforce new federal transparency guidelines for AI-generated content, enhancing protection against harmful deepfakes
โข Legal framework enhanced by granting individuals and organizations the right to sue for unauthorized use of AI-generated content
โข Major endorsers, including SAG-AFTRA and the Recording Academy, applaud the Act for establishing protections for artists and journalists against AI-driven content theft.
๐AI Academia
Making LLMs Safer Through 'Machine Unlearning'
In the rapidly evolving field of generative artificial intelligence, ensuring the safety and reliability of large language models (LLMs) has become a critical challenge. A groundbreaking paper titled "Towards Safer Large Language Models through Machine Unlearning
Large Language Models Enhance Real-Time Anomaly Detection in Robotic Systems
โข A new two-stage reasoning framework using large language models enhances real-time anomaly detection and safety in robotic systems
โข The system's initial stage quickly classifies anomalies, while the second stage, though slower, deliberates on safe fallback actions
โข Demonstrations in simulated and real-world settings show the system can outperform traditional models in dynamic environments like autonomous vehicles.
'A Detailed Taxonomy of Data Contamination in Large Language Models'
โข A new taxonomy categorizes data contamination in large language models, enhancing understanding of influences on AI performance
โข Contamination types like in-domain data during training show similar benefits to direct test data contamination, affecting tasks such as summarization and question answering
โข Recommendations developed for mitigating contamination during model training to ensure more reliable evaluations and outcomes.
Study Reveals Large Language Models Still Susceptible to Bias Elicitation Through Jailbreak Prompts
โข A recent study from the University of Calabria reveals that Large Language Models (LLMs) still produce biased responses when manipulated with specialized prompts
โข The research demonstrates that biases in LLMs, such as gender or ethnic stereotypes, persist despite advanced model training and alignment techniques
โข Enhancing bias mitigation strategies is crucial as per the findings, to develop more equitable and reliable artificial intelligence systems.
Study Reveals Key Linguistic Differences Between AI and Human-Written Texts
โข A study utilizes Open Brain AI to compare linguistic features in human-written and AI-generated essays
โข Significant differences found in language elements like word stress and usage of complex words
โข Results highlight the need for enhanced AI training to improve generation of human-like text.
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!
