Global AI Weekly
Issue number: 91 | Tuesday, April 1, 2025
Highlights
OpenAI says “our GPUs are melting” as it limits ChatGPT image generation requests
OpenAI is introducing temporary limits on the number of image generation requests users can make with ChatGPT. This decision comes as the company works to optimize the feature and address the high demand that's putting a strain on its GPU resources. OpenAI aims to enhance user experience while balancing the growing workload on its systems.
theverge.com
Gemini 2.5 Pro is here with bigger numbers and great vibes
Google introduces the Gemini 2.5 Pro, a cutting-edge AI model designed to enhance your experience with smarter capabilities and impressive performance. Packed with larger parameters and refined technology, it promises more accurate and faster responses while maintaining an intuitive, user-friendly vibe. Aimed at setting new benchmarks in AI innovation, Gemini 2.5 Pro is ready to make your digital interactions seamless and smarter than ever.
arstechnica.comResearch
OmniScience: A Domain-Specialized LLM for Scientific Reasoning and Discovery
The paper introduces OmniScience, a large language model (LLM) specifically designed for scientific reasoning and discovery. It highlights how the model is tailored to handle the complexities of scientific domains, bridging gaps in traditional methodologies. The work emphasizes OmniScience's potential to enhance research workflows and contribute to advancements across various scientific disciplines.
arxiv.org
Fugatto 1 - Foundational Generative Audio Transformer Opus 1 | Research
Fugatto is an innovative audio synthesis and transformation model designed to follow free-form text instructions, with or without accompanying audio inputs. Unlike audio-only models that can’t inherently interpret the instructions behind audio data, Fugatto bridges this gap by leveraging unique training techniques. This approach positions it as a cutting-edge tool in audio generation and transformation, pushing the boundaries of how text and sound can interact.
research.nvidia.comVideo
How to Create Datasets for Individual Steps of Your Agent
Learn how to evaluate individual components of your agents by creating targeted datasets in LangSmith. This approach is akin to unit testing in software development, ensuring each step of your agent works as intended. The video guides you through the process, making it easy to optimize and fine-tune your applications effectively.
youtu.be
Scott Hanselman - Human Hearts, Silicon Minds
In this episode of *Silicon Minds Human Hearts,* Scott Hanselman from Microsoft discusses the intersection of human experience and artificial intelligence. He reflects on his 30-year journey in computing, the ethical challenges linked to rapid tech advancements, and how AI reshapes productivity. Scott also highlights meaningful AI applications in personal and educational contexts, offering practical insights for navigating the ever-evolving AI landscape. A thought-provoking conversation on the balance between innovation and ethics!
youtube.comArticles
DeepSeek-V3 now runs at 20 tokens per second on Mac Studio, and that's a nightmare for OpenAI
DeepSeek's 685B-parameter AI model showcases impressive performance by running at 20 tokens per second on Apple's Mac Studio, using only 200 watts of power. This efficiency surpasses competitors like Claude Sonnet and presents a significant challenge to OpenAI's reliance on cloud-based solutions. The model’s ability to operate locally on accessible hardware raises questions about the future of AI deployment and accessibility.
venturebeat.com
Alibaba launches new open-source AI model for "cost-effective AI agents"
Alibaba Cloud has introduced a new addition to its "Qwen series" of AI models, designed to support the development of cost-effective AI agents. This release comes as competition among large language models in China intensifies. The open-source model aims to make AI-powered solutions more accessible and versatile for various applications.
cnbc.com
Nowcasting on Search is bringing AI-powered weather forecasts to people across Africa
Google Research teams are making AI-powered short-term weather forecasts, known as Nowcasting, available to people across Africa through Search. This technology provides localized and timely precipitation predictions, helping communities better prepare for weather changes. It’s a step forward in combining advanced AI with accessibility to support day-to-day decision-making in areas where consistent weather updates are crucial.
blog.googleUpcoming Events
Global AI Bootcamp 2025
The Global AI Bootcamp 2025 is still in full swing and we can look back at an amazing first month of local chapters organizing bootcamps, bringing together developers and AI enthusiasts from around the world. Bootcamps feature workshops, sessions, and discussions to explore and learn about AI and are a fantastic opportunity to connect with others passionate about artificial intelligence. During the coming weeks, more bootcamps are happening, definitely check these events out.
globalai.communityCode
GitHub - microsoft/Power-CAT-Copilot-Studio-Kit
Microsoft's Power CAT Copilot Studio Kit is an open-source project available on GitHub. It offers tools and resources to enhance development experiences in creating powerful applications with Microsoft's Power Platform. Join the community, collaborate, and contribute to the ongoing development of this kit.
github.com
GitHub - microsoft/Obliviate-Unmemorization: The code for Obliviate: Efficient Unmemorization for Protecting Intellectual Property in Large Language Models
This repository contains the code for Obliviate, a method designed to efficiently unmemorize sensitive or proprietary data from large language models, helping protect intellectual property. It focuses on ensuring that critical information can be removed without compromising the model’s overall functionality. The project provides innovative solutions for maintaining data security and privacy in AI systems.
github.comPodcast
Leaders of AI Podcast
Leaders of AI is the podcast for decision-makers looking to stay ahead in the AI revolution. Co-hosts Julia McCoy, a pioneer in content marketing turned AI integration expert, and David Shapiro, a seasoned AI strategist with a background in cutting-edge automation, bring their expertise to the table. Together with special guests, they explore the intersection of AI, business, and humanity, offering practical insights and forward-thinking discussions.
open.spotify.com