Welcome to The Daily AI Briefing! Today, we're diving into the rapidly evolving AI landscape with breaking developments that are reshaping our technological future. From remarkable mathematical achievements to concerning vulnerability discoveries, we've gathered the most significant AI news that matters right now. Stay with us as we explore the latest breakthroughs, challenges, and opportunities emerging in the world of artificial intelligence. In today's episode, we'll cover OpenAI's impressive math performance, ARC's new interactive AGI test, a tutorial for building your own AI writing assistant, concerning research on AI vulnerabilities, trending AI tools, job opportunities, and other notable AI developments. Let's start with OpenAI's mathematical milestone. The company has claimed gold-level performance in an evaluation modeled after the 2025 International Math Olympiad. Their experimental reasoning LLM solved 5 out of 6 problems, scoring 35 out of 42 points—enough for a gold medal in the official competition. The model wrote natural language proofs under the same conditions as human competitors, without tools or internet access. Each answer was independently graded by former IMO medalists. However, Google DeepMind has challenged this claim, noting that official IMO marking guidelines weren't used. Moving to testing AI capabilities, ARC Prize has released a preview of ARC-AGI-3, a new interactive reasoning benchmark. This test evaluates AI agents' ability to generalize in unfamiliar environments through three original games. Interestingly, frontier models like OpenAI's o3 and Grok 4 are struggling with levels that humans find relatively easy. The benchmark requires agents to learn through trial and error without instructions, similar to how humans adapt to new challenges. ARC Prize is also launching a public contest for the AI community to build better agents. For those interested in practical AI applications, there's a new tutorial teaching how to create a personalized AI writing assistant. Using the Grok 4 API, you can build an assistant that analyzes your writing samples and generates content matching your exact style. The process involves generating an API key from xAI, setting up your environment, creating a system prompt with your writing examples, and watching your assistant generate content that sounds just like you. On a concerning note, Wharton Generative AI Labs has published research showing that AI models can be manipulated using psychological persuasion techniques. Testing Robert Cialdini's principles of influence on GPT-4o-mini, researchers found these techniques more than doubled the model's compliance with objectionable queries from 33% to 72%. Commitment and scarcity principles were particularly effective, increasing compliance rates dramatically. In trending AI tools, we're seeing Pulse for creating Wikipedia-style articles, Kimi K2 with enhanced tool calling capabilities, OpenReasoning-Nemotron from Nvidia for math and science, and AWS's new AI IDE called Kiro for agentic coding. For job seekers, companies like Anthropic, Databricks, Waymo, and Shield AI are hiring for various AI-related positions from brand design to technical writing. Other notable news includes OpenAI launching a $50 million fund for nonprofits, Perplexity discussing pre-installation of its agentic browser on smartphones, Microsoft blocking Cursor's access to VSCode extensions, xAI developing "Baby Grok," Meta refusing to sign the EU's AI Code of Practice, and Sam Altman announcing that OpenAI is on track to bring over one million GPUs online by year-end. That wraps up today's AI Briefing. The landscape continues to evolve at breakneck speed, with impressive achievements alongside concerning vulnerabilities. As AI capabilities grow, so does the importance of responsible development and testing. Join us tomorrow for more cutting-edge developments in the world of artificial intelligence. This has been The Daily AI Briefing—st