OpenAI has developed CriticGPT, an AI assistant designed to help human trainers refine GPT-4’s code generation abilities. CriticGPT identifies subtle coding errors that humans might miss, improving the accuracy and quality of GPT-4’s code output. The model significantly outperforms both humans and other AI systems in bug detection, and even helps reduce hallucinations in code generation.
Results for: AI
Microsoft has discovered a new AI jailbreak technique called ‘Skeleton Key’ that allows malicious users to bypass safety guardrails and access restricted content. The technique involves convincing the AI model to augment its guardrails, leading to the production of potentially harmful content. Microsoft has disclosed the vulnerability and implemented safeguards to prevent this exploit.
OpenAI has unveiled CriticGPT, an AI model specifically designed to identify and correct errors in code generated by GPT-4. This new model leverages reinforcement learning from human feedback (RLHF) to enhance code quality, achieving a 63% improvement over ChatGPT in error detection. While still under development, CriticGPT has the potential to revolutionize code review and enhance the reliability of AI-generated code.
AACSB International is leading the charge in integrating AI into business schools. Their recent report explores how schools can leverage generative AI, and they’ve invited renowned AI expert Ethan Mollick to speak at their Deans Conference. AACSB is also hosting AI conferences to foster discussion and exploration of this transformative technology in business education.
Artificial intelligence (AI) is revolutionizing the field of biodiversity research, helping scientists identify new animal species and monitor complex ecosystems. While AI alone cannot definitively identify new species, it can flag potential candidates for further investigation by expert biologists. By analyzing vast datasets collected from camera traps, smartphones, and automated monitoring systems, AI is enabling researchers to gain a deeper understanding of Earth’s wildlife and the threats they face.
Solos, the makers of AirGo smart glasses, are adding a front camera and AI features to their lineup, creating a direct competitor to the popular Ray-Ban Meta glasses. The new AirGo Vision model boasts ChatGPT-4 integration for visual search and other functionalities, offering a modular design with interchangeable front panels. The company will also release three new AirGo styles without cameras, featuring an LED notification light and a modular frame design.
Character.AI, the popular digital avatar startup, has introduced ‘Character Call’, allowing users to engage in real-time conversations with AI agents. This feature builds upon the company’s existing AI-generated voice technology, expanding the possibilities for interactive experiences, from language practice to roleplaying.
A groundbreaking study from the University of Reading reveals that AI can produce exam answers superior to real students, raising concerns about academic integrity and the future of education in the age of AI. The study, which used ChatGPT to generate exam answers for psychology students, found that AI-generated responses not only scored higher but also went largely undetected by markers. The findings highlight the urgent need for educators to rethink assessment methods and adapt to the evolving landscape of AI.
Dileep Kumar Pandiya, an AI and microservices expert, delivered a captivating session at DeveloperWeek Europe 2024, exploring how AI can revolutionize microservices architecture for dynamic decision making. His talk highlighted practical applications, real-world examples, and the transformative power of AI in driving business innovation and operational excellence.
Figma has unveiled a significant redesign and new AI-powered tools at its annual Config conference, aiming to simplify and expedite design workflows. This update features a refreshed visual style, updated toolbar, and AI-powered features for streamlining design tasks.