Google’s new diffusion AI agent mimics human writing to improve enterprise research
1 min read
Summary
Google researchers have developed a method for AI research agents to produce more accurate and comprehensive research on complex topics, according to a paper published on arXiv.
Called Test-Time Diffusion Deep Researcher (TTD-DR), it uses diffusion mechanisms and an evolutionary algorithm to produce reports in a manner inspired by human writing, in an iterative process of drafting, searching for information and making revisions.
TTD-DR could be used to create a new generation of bespoke research assistants that can perform high-value tasks that standard retrieval-augmented generation systems are not capable of, such as generating competitive analysis reports or market entry reports.
The research found TTD-DR outperformed leading AI systems from rivals OpenAI, Perplexity and Grok on benchmarks.