Refreshed just now
https://apolloresearch.ai/science/we-need-3rd-party-training-run-assessments
Why it’s here: Argues for third-party training-run assessments as a standard part of frontier AI safety, a policy-relevant AI safety governance proposal bearing on federal oversight and developer credibility.
Trace across the boardItem
4d agoApollo Researchhttps://apolloresearch.ai/science/we-need-3rd-party-training-run-assessmentsApollo Research argues that third-party Training-Run Assessments of frontier model post-training should become a standard part of AI safety, since final-checkpoint evaluations may miss scheming.1 outletAI6 linked
Connected
6 links- NewsApollo ResearchBoth involve Apollo Research
- Newshttps://apolloresearch.ai/blog/pilot-towards-embedded-evaluationsBoth involve Apollo Research
- NewsBlog – Apollo ResearchBoth involve Apollo Research
- Newshttps://apolloresearch.ai/components-nBoth involve Apollo Research
- NewsOpenAI will embed safety researchers to investigate AI subversionBoth involve Apollo Research
- Newshttps://apolloresearch.ai/blog-nBoth involve Apollo Research
About
- Apollo Research