The Daily AI Briefing - 20/03/2025

Welcome to The Daily AI Briefing, here are today's headlines! Today we're looking at groundbreaking research showing AI capabilities follow a "Moore's Law" pattern, Hollywood's pushback against AI copyright proposals, techniques for improving non-reasoning AI responses, Nvidia's new open-source reasoning models, and a roundup of the latest AI tools making waves. These developments highlight the accelerating pace of AI advancement alongside growing tensions over its implementation. **AI Capabilities Following "Moore's Law" Pattern** Researchers at METR have made a fascinating discovery about AI development trajectories. Their study reveals that the length of tasks AI agents can complete autonomously has been doubling approximately every 7 months since 2019, effectively establishing a "Moore's Law" for AI capabilities. The research team tracked human and AI performance across 170 software tasks ranging from quick decisions to complex engineering challenges. Current top-tier models like 3.7 Sonnet demonstrate a "time horizon" of 59 minutes, meaning they can complete tasks that would take skilled humans about an hour with at least 50% reliability. Meanwhile, older models like GPT-4 handle tasks requiring 8-15 minutes of human time, while 2019 systems struggle with anything beyond a few seconds. If this exponential trend continues, we could see AI systems capable of completing month-long human-equivalent projects with reasonable reliability by 2030. This predictable growth pattern provides an important forecasting tool for the industry and could significantly impact how organizations plan for AI integration in the coming years. **Hollywood Creatives Push Back Against AI Copyright Proposals** More than 400 Hollywood creatives, including stars like Ben Stiller, Mark Ruffalo, Cate Blanchett, Paul McCartney, and Aubrey Plaza, have signed an open letter urging the Trump administration to reject proposals from OpenAI and Google that would expand AI training on copyrighted works. The letter directly responds to submissions in the AI Action Plan where tech giants argued for expanded fair use protections. OpenAI even framed AI copyright exemptions as a "matter of national security," while Google maintained that current fair use frameworks already support AI innovation. The creative community strongly disagrees, arguing these proposals would allow AI companies to "freely exploit" creative industries. Their position is straightforward: AI companies should simply "negotiate appropriate licenses with copyright holders – just as every other industry does." This confrontation highlights the growing tension between technology companies pushing AI advancement and creative professionals concerned about the devaluation of their work. **Improving Non-Reasoning AI Responses Through Structured Approaches** A new tutorial is making waves by demonstrating how to dramatically enhance the intelligence of non-reasoning AI models. The approach implements structured reasoning with XML tags, forcing models to think step-by-step before providing answers. The method involves carefully structuring prompts with XML tags to separate the reasoning process from the final output. By providing specific context and task details, including examples, and explicitly instructing the model to "think" first, then answer, the quality of AI-generated content improves significantly. This technique proves especially valuable when asking AI to match specific writing styles or analyze complex information before generating content. Comparison tests show dramatic improvements when using this reasoning framework versus standard prompting techniques, offering a practical approach for anyone looking to get more sophisticated responses from existing AI systems. **Nvidia Releases Open-Source Reasoning Models** Nvidia has launched its Llama Nemotron family of open-source reasoning models, designed to accelerate enterprise adoption of agentic AI capable of complex problem-solv