A new 512-prompt benchmark claims to measure LLM over-refusal on scientific dual-use questions, but its design and labeling flaws undermine the claim.
Defense Priorities in the Open-Source AI Debate: A Preliminary Assessment
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
A spirited debate is taking place over the regulation of open foundation models: artificial intelligence models whose underlying architectures and parameters are made public and can be inspected, modified, and run by end users. Proposed limits on releasing open foundation models may have significant defense industrial impacts. If model training is a form of defense production, these impacts deserve further scrutiny. Preliminary evidence suggests that an open foundation model ecosystem could benefit the U.S. Department of Defense's supplier diversity, sustainment, cybersecurity, and innovation priorities. Follow-on analyses should quantify impacts on acquisition cost and supply chain security.
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
REJECT 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Forbidden Science: Dual-Use AI Challenge Benchmark and Scientific Refusal Tests
A new 512-prompt benchmark claims to measure LLM over-refusal on scientific dual-use questions, but its design and labeling flaws undermine the claim.