Researchers have introduced Test-Time Diffusion Deep Researcher (TTD-DR), a new deep research agent that models research report writing as a diffusion process. The system is described as the first research agent to treat a messy first draft as something that is gradually polished into a final report using retrieval and refinement steps.
The work comes as large language models have driven the rise of deep research agents, which can generate ideas, retrieve information, run experiments, and draft reports and academic papers. The source says many existing public DR agents use techniques such as chain-of-thought reasoning or generating multiple answers and selecting the best one, but often combine tools without reflecting the iterative way people research and revise complex topics.
TTD-DR is presented as an attempt to imitate that human process. It is built around two algorithms that work together. The first is component-wise optimization via self-evolution, which improves each step in the research workflow. The second is report-level refinement via denoising with retrieval, which uses newly retrieved information to revise and improve the draft.
The source says this approach is meant to reflect how people plan, draft, research, and iterate based on feedback when writing about a complex topic. It also notes that a key part of revision is doing more research to find missing information or strengthen arguments.
According to the researchers, TTD-DR achieves state-of-the-art results on long-form report writing and multi-hop reasoning tasks.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
