Reproduce “Deep Research of Deep Research: From Transformer to Agent, From AI to AI for Science” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yipeng Yu
This paper establishes a unified conceptual framework for 'deep research' and 'AI for Science' (AI4S), defining deep research as LLM-based AI systems that interact multimodally with environments to aid human scientific discovery. It traces AI's evolution from foundational generative models to agents, assesses current agent capabilities and human-AI interaction paradigms, and identifies critical challenges for advancing AI in scientific inquiry.
Key insights
- Deep research is formally defined as LLM-based AI systems that interact multimodally and iteratively with external environments via tools, aiming to assist or surpass human scientific capabilities.
- Large Language Models and Stable Diffusion are identified as the foundational 'twin pillars' of generative AI, evolving into agents that operate on a 'Thought→Action→Observation' loop with tools and memory.
- AI for Science applications primarily function at a 'Collaborative-Level' (L3) of automation, indicating that full autonomy in scientific discovery remains largely unachieved, and agents often reconfigure human effort rather than eliminating it.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.