The Download: coding's future, the 'Steroid Olympics,' and AI-driven science (www.technologyreview.com)

At a recent developer event in London, Anthropic showcased its AI coding tool, Code with Claude, revealing that nearly half of the attendees had deployed code generated entirely by the AI without reviewing it. This reflects a growing trend among developers to increasingly trust AI for coding tasks, raising questions about the implications of such automation. While Anthropic aims to push the boundaries of coding automation, skepticism remains within parts of the developer community regarding the potential risks and quality of AI-generated code. In parallel, Google's recent I/O keynote highlighted a pivotal shift in AI-driven scientific research with the introduction of Gemini for Science, an advanced agentic system designed to potentially conduct research projects autonomously. This move signals a departure from specialized AI systems, emphasizing the growing ambition for AI to independently tackle complex scientific inquiries. Moreover, the concept of world models is gaining traction in AI research, championed by leading figures including those at Google DeepMind, suggesting a future where AI systems could better comprehend and interpret the physical environment, fundamentally transforming our interaction with AI.