Global AI Weekly
Issue number: 98 | Tuesday, May 20, 2025
Highlights
Going ‘AI first’ appears to be backfiring on Klarna and Duolingo
Klarna and Duolingo are facing pushback over their “AI-first” strategies. Klarna is now hiring more human staff after quality concerns with AI-led customer service. Duolingo, while sticking to its plan, is under fire on social media, with users accusing it of replacing people with machines. Companies are now working to find a better balance between AI innovation and human insight.
fastcompany.com
Scientists use AI to encrypt secret messages that are invisible to cybersecurity systems
Scientists are exploring how AI chatbots can be used to embed secret messages in seemingly ordinary text, creating a new form of encryption that's nearly impossible for cybersecurity systems to detect. This innovative technique could revolutionize digital communication by offering an unparalleled level of security. However, it also raises concerns about the potential misuse of such technology.
livescience.com
AI LLMs Learn Like Us, But Without Abstract Thought
A recent study reveals that large language models (LLMs) like GPT-J generate language by using analogies rather than fixed grammatical rules, mimicking how humans handle unfamiliar language. While these AI models process information similarly to humans in this way, they lack the capacity for abstract thought. This discovery sheds light on the parallels and limitations of AI's approach to language understanding compared to human cognition.
neurosciencenews.comResearch
ZeroSearch: Incentivize the Search Capability of LLMs without Searching
This page introduces ZeroSearch, an innovative approach designed to enhance the search capabilities of large language models (LLMs) without requiring actual searches. The discussion focuses on how this method incentivizes LLMs to generate search-relevant responses by leveraging existing data efficiently. It highlights the potential to optimize performance while reducing the need for external search integration.
huggingface.co
Unified Continuous Generative Models
This paper page explores Unified Continuous Generative Models, focusing on their design and application. It highlights how these models integrate various generative approaches into a cohesive framework while addressing modern computational challenges. The discussion emphasizes their potential to streamline generative processes across different domains.
huggingface.coVideo
Silicon Minds Human Hearts - Aditi Maheshwari
This episode of Silicon Minds Human Hearts features Aditi Maheshwari, a product manager from Microsoft's Azure AI team, sharing her journey and expertise in the evolving world of AI. Aditi discusses how AI is transforming businesses, the technology behind self-driving cars, and the critical need for data privacy. She also highlights the revolutionary impact of AI layers in bridging the gap between human and machine collaboration. Tune in for an engaging and insightful conversation packed with knowledge and a touch of humor!
youtube.comArticles
ARTIST: A Reinforcement Learning Framework That Equips LLMs with Agentic Reasoning and Dynamic Tool Use
Microsoft researchers have unveiled ARTIST, a new reinforcement learning framework designed to enhance large language models (LLMs) with advanced reasoning and adaptable tool usage. This framework enables LLMs to act more like agents, effectively navigating tasks and leveraging tools dynamically based on the context. ARTIST represents a significant step toward improving AI's problem-solving capabilities and flexibility in real-world applications.
marktechpost.com
A Deep-Dive into Agent Interoperability Protocols: MCP, ACP, A2A, and ANP
This detailed exploration breaks down the key components of next-generation interoperability protocols, including MCP, ACP, A2A, and ANP. It explains how these protocols streamline communication and interactions between models, agents, and networks to enhance efficiency and integration. The focus is on their technical applications and potential to revolutionize interconnected systems.
marktechpost.com
Predicting gene expression from DNA sequence using deep learning models
Advances in deep learning have brought new breakthroughs in predicting gene expression directly from DNA sequences. Using data from epigenome mapping and high-throughput reporter assays, these models can accurately capture the regulatory logic of gene activity. They offer insights into non-coding variants, uncover molecular mechanisms of gene regulation, and enable the design of synthetic regulatory elements. Barbadilla-Martínez et al. outline recent developments, the available data sets, and the strengths and challenges of these innovative approaches.
nature.com
How Cursor Indexes Codebases Fast
Merkle Trees are essential for organizing and verifying data, making them a key player in speeding up processes like searching through codebases. This method uses cryptographic principles to structure data efficiently while maintaining its integrity. Learn how this technology enables faster indexing and improved performance in real-world applications.
read.engineerscodex.com
Google DeepMind’s AI Agent Dreams Up Algorithms Beyond Human Expertise
Google DeepMind has introduced an AI agent that merges the coding capabilities of Gemini with an evolutionary strategy to tackle complex tasks. This innovative system has shown promise in optimizing datacenter scheduling, enhancing chip design, and refining large language models. By generating algorithms that surpass traditional human-crafted methods, it demonstrates the potential for AI to address challenges with greater precision and efficiency.
wired.com
10 Free Artificial Intelligence Books For 2025
Looking to boost your AI expertise? Check out this curated list of 10 free artificial intelligence books perfect for learning in 2025. These resources cover a range of topics, ensuring there's something for everyone, whether you're a beginner or looking to deepen your knowledge. Start exploring AI with these valuable and accessible tools!
kdnuggets.comUpcoming Events
See What’s Next in AI at NVIDIA GTC Paris
NVIDIA GTC Paris is taking place at VivaTech 2025, June 10-12, to explore how NVIDIA is fueling Europe’s sovereign AI movement—powering next-gen AI factories and accelerating real-world adoption across enterprise and industry. AI practitioners and developers will gain hands-on insights into building and deploying AI at scale, optimizing infrastructure, and using the latest tools and frameworks for EU-specific use cases. On June 10, full-day, expert-led workshops will give you practical skills and real-world insights on topics like NVIDIA Omniverse, accelerated robotics, and large language models. Don’t miss NVIDIA founder & CEO, Jensen Huang, live on stage for the VivaTech Keynote at the Dôme de Paris on June 11 at 11:00 a.m. CEST.
nvidia.com
June: AgentCon 2025
The AI Agents World Tour, part of AgentCon 2025, is a global series of one-day events tailored for developers working with autonomous AI agents. This June we will be bringing together top engineers and researchers. These conferences will take place in major cities like San Francisco and Singapore. It's a unique opportunity to connect, learn, and shape the future of AI technology.
agentcon.devCode
10 GitHub Repositories to Master Large Language Models
Discover 10 must-see GitHub repositories to help you master large language models (LLMs). From foundational concepts to advanced techniques, these resources include books, courses, tutorials, exercises, and hands-on projects to enhance your skills. Whether you're a beginner or looking to deepen your understanding, these repositories offer comprehensive guides to get you started and take your expertise to the next level. Perfect for anyone eager to explore the world of LLMs!
kdnuggets.com
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms
AlphaEvolve is an innovative AI agent that integrates the creative capabilities of large language models with the precision of automated evaluators to develop advanced algorithms. By leveraging Gemini technology, it pushes the boundaries of math and computing, creating practical solutions with enhanced efficiency. This fusion of AI tools promises a new level of sophistication in coding and problem-solving.
deepmind.google
A Step-by-Step Guide to Build an Automated Knowledge Graph Pipeline Using LangGraph and NetworkX
This guide walks you through creating an automated knowledge graph pipeline using LangGraph and NetworkX. It covers everything from setting up your environment to integrating data sources and designing the graph structure. Clear instructions and examples make it easy to follow, whether you're new to knowledge graphs or looking to streamline your workflow.
marktechpost.comPodcast
Hard Fork
"Hard Fork" is a weekly podcast hosted by journalists Kevin Roose and Casey Newton, focusing on the fast-evolving world of technology and its impact on our lives. With a fresh perspective on the latest tech trends, they break down complex topics in an engaging and accessible way. Perfect for anyone curious about the future that’s unfolding right now.
open.spotify.com