Definition
Plain language
One of the autonomous research-agent systems used as a comparison in citation-integrity audits.
As stated in the literature
An autonomous research-agent baseline evaluated in Chain-of-Evidence audits, where it exhibited a high rate of hallucinated bibliography references despite having a citation-verification tool available.
Why it matters: It illustrates that giving an agent a citation-checking tool isn't enough if it doesn't reliably use it, underscoring the need to verify automated research.
For example, it can run a research workflow and produce a paper, but in audits it filled its bibliography with references that did not actually exist.
Heard on the show
“So one of the systems — DeepScientist — produces papers with a twenty-point-nine percent hallucinated reference rate.”Episode 089 — When AI-Written Papers Read Well But the Evidence Underneath Is Broken