Comparison of LLMs with tools (agents) to the DEEP Research mode by OPENai and GooGLE.

The authors assessed several state-of-the-art (SOTA) LLMs, originally lacking native Internet access, by augmenting them with an external search engine and link reader to enable the web-retrieval capabilities essential for completing our evaluation tasks. "In the evaluation of base models, we integrated search and link-reading tools using each model's native function call interface."

Additional info by the authors: "Specifically, we used SerpAPI for Google Search access and Firecrawl for retrieving web pages in Markdown format."

All rights w/ authors:
ReportBench: Evaluating Deep Research Agents via Academic Survey Tasks
Minghao Li, Ying Zeng, Zhihao Cheng, Cong Ma, Kai Jia
from
ByteDance BandAI

#aireasoning
#aiagents
#scienceexplained #trust