ChatGPT Just Got Smarter: Can Now Be Set Up to Do Scheduled Tasks
IBM and L’Oréal harness AI for sustainable beauty, USPTO unveils strategies for AI-driven innovation. Studies highlight AI’s potential in healthcare and ethics while addressing biases in dialects..
Today's highlights:
🚀 AI Breakthroughs
OpenAI Rolls Out ChatGPT Tasks for Proactive Scheduling and Automated Alerts
• OpenAI unveils Tasks, a beta feature for ChatGPT, enabling proactive scheduling of reminders and actions, enhancing its role as a proactive AI assistant
• Users can set up daily or recurring notifications for events like weather updates or concert reminders, receiving alerts across platforms, though task management is web-exclusive
• Tasks aim to boost efficiency by evolving ChatGPT's functionality, offering personalized schedules and timely updates for premium users at a $20/month price, globally available soon..
IBM and L’Oréal Collaborate on AI-Powered Sustainable Beauty Solutions for 2030 Goals
• IBM and L’Oréal have collaborated to harness AI in advancing sustainable cosmetics, using AI to uncover insights for utilizing sustainable raw materials, aiming at reducing energy and material waste;
• The partnership marks the creation of the first AI model designed for sustainable cosmetics, aiding L’Oréal in sourcing most formulas from bio-sourced materials by 2030;
• The AI model will accelerate product formulation processes, utilizing extensive data to enable L’Oréal's 4,000 researchers to develop eco-friendly cosmetics and innovate inclusively and sustainably.
Microsoft Releases AutoGen v0.4, Enhancing Agentic AI with Asynchronous Framework
• AutoGen v0.4 introduces an asynchronous, event-driven architecture to enhance flexibility in agentic workflows, addressing previous user feedback on architectural constraints and limited debugging functionality
• The update features cross-language support and a modular design, enabling developers to create scalable, distributed agent networks and custom multi-agent applications
• AutoGen Bench, Studio, and Magentic-One provide tools for benchmarking, prototyping, and completing open-ended web tasks, supporting developers in building advanced agentic systems;
OpenAI and Axios Partner to Enhance Local Newsrooms with AI Technology
• OpenAI's collaboration with Axios funds AI newsrooms in cities like Pittsburgh, enhancing local news coverage and representing OpenAI's first direct investment in newsroom expansion;
• With the partnership, Axios content becomes accessible via ChatGPT, offering summaries and links, while Axios incorporates OpenAI's AI for journalism development and distribution;
• The collaboration is part of broader OpenAI efforts, supporting 160+ news outlets in 20+ languages, with ChatGPT reaching 300 million weekly users, transforming global news dissemination;
⚖️ AI Ethics
USPTO's AI Strategy to Tackle AI's Influence on IP and Innovation Ecosystem
• The USPTO unveiled a comprehensive AI strategy targeting improvements in IP policy, agency operations, and the broader innovation ecosystem, aiming to maintain U.S. leadership in AI innovation;
• Core objectives include fostering inclusive AI innovation, enhancing computational infrastructure, and promoting responsible AI use, while developing internal expertise and collaborating with governmental and global partners;
• Over 6,000 stakeholders engaged since 2022, driven by USPTO's AI and Emerging Technology Partnership, signaling an ongoing commitment to guiding AI-driven advancements across industries.
Biden-Harris Administration Releases an Interim Final Rule on Artificial Intelligence Diffusion
• The U.S. emphasizes its AI leadership, releasing an Interim Final Rule to control AI diffusion and strengthen security
• Six key mechanisms in the rule facilitate chip sales to allies, streamline order processing, and expand U.S. AI presence globally
• The rule sets security standards to protect advanced AI models while restricting access for countries of concern, aiming to prevent misuse of U.S. technology
White House Issues Executive Order to Strengthen U.S. Leadership in AI Infrastructure
• The White House issued an executive order to establish the U.S. as a leader in AI infrastructure, focusing on national security and economic competitiveness;
• The order emphasizes collaboration with the private sector for building advanced computing clusters and secure supply chains essential for AI development;
• The initiative also prioritizes developing clean energy technologies and ensuring local community benefits from the construction and operation of AI infrastructure.
AI-Powered Scammers Exploit Trust, Convincing Victim of Fake Brad Pitt Romance
• A French interior designer was deceived into a fake romance with an AI-generated Brad Pitt, leading to financial and emotional devastation
• The 18-month scam exploited generative AI technology to create convincing deepfakes, ultimately costing the victim over €800,000
• As AI-powered scams surge, experts stress the increasing difficulty in differentiating real interactions from fake AI-generated personas and voices.
🎓AI Academia
New Instrument Validates Documentation Summaries by Large Language Models in Healthcare
• Researchers developed the Provider Documentation Summarization Quality Instrument (PDSQI-9) to evaluate the efficacy of Large Language Models in generating clinical summaries from electronic health records;
• Validation methods for PDSQI-9 involved multiple statistical techniques, including Pearson correlation and factor analysis, confirming structural validity and generalizability with high inter-rater reliability;
• The study assessed 779 summaries from various specialties, finding PDSQI-9's internal consistency and reliability sufficient for integration in healthcare workflows using advanced LLMs.
Study Highlights Convergent Ethical Reasoning in Six Major Generative AI Models
• Recent research explores the ethical logic of six leading large language models, including GPT-4o and LLaMA 3.1, using moral dilemmas like the Trolley Problem and Heinz Dilemma;
• The study reveals these models predominantly adopt a rationalist, consequentialist ethical stance, focusing on harm minimization and fairness, akin to graduate-level moral discourse;
• Despite architectural similarities, significant ethical reasoning differences among models were noted due to fine-tuning variations, highlighting challenges in aligning AI ethics with human values.
SAIF Offers a Systematic Framework to Address Generative AI Risks in Public Sector
• A comprehensive framework, SAIF, has been developed to evaluate the risks of generative AI in the public sector, addressing challenges in applications like immigration and welfare services
• SAIF operates through stages including risk breakdown, scenario design, and the application of jailbreak methods, ensuring a robust assessment of AI risks in public applications
• The framework also accommodates emerging methods and prompt types, providing a dynamic approach to effectively respond to unforeseen generative AI risk scenarios in the public sector.
Enhanced AI Models Improve Screening for Depression and Anxiety Disorders Effectively
• Researchers from The University of Hong Kong and Tsinghua University developed enhanced large language models to improve the accuracy of screening for depression and anxiety
• The collaborative research aims to leverage AI technology to assist healthcare professionals in diagnosing mental health conditions more effectively and promptly
• The study underscores the potential of integrating advanced AI models in neuropsychology for early detection and intervention of mental health issues;
Study Reveals Bias in Language Models for Non-Standard Dialect Reasoning Tasks
• Research reveals that Large Language Models (LLMs) display significant bias and lack robustness in understanding African American Vernacular English (AAVE) during reasoning tasks such as math and logic;
• The study introduces ReDial, a benchmark comprising over 1,200 parallel queries in Standardized English and AAVE, to evaluate dialectal fairness in widely used LLMs like GPT and Claude;
• Findings indicate that mainstream LLMs often provide unfair and brittle responses to AAVE queries, highlighting a need for improved dialect inclusivity in model training and evaluation.
Large Language Models Exhibit Regional Bias in Recruiting Software Engineers
• Research reveals biases in GitHub recruitment conducted by Large Language Models, displaying regional preferences and differing roles assigned based on candidate location.
• Experiments using ChatGPT-4 highlight implicit biases in selecting software developers, favoring certain countries even in counterfactual scenarios.
• Study underscores the need to address and mitigate societal biases in AI recruitment processes to ensure fair and equitable outcomes.
About ABCP: 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!
