Article · Original
The article text is unavailable in this language; an existing version is shown.
https://t.co/bpAkCzyO5S
OpenAI tests whether model answers in combat scenarios comply with the laws of warMachine translation
Automatically verified and published · Generated and evidence-checked automatically; not reviewed by a human.
Fortune reports that international experts entered high-stakes combat prompts into OpenAI models, after which engineers changed the models based on how the answers measured up against the Geneva Conventions. The report links the testing to the models going to the U.S. military, but gives no test results.
Fortune: OpenAI now tests whether its models' answers obey the laws of war, because those models are going to the U.S. military.
OpenAI created its first human rights and responsible deployment role, Sarah Yager, who spent 30 years on military issues before joining in June, described her method.
International experts entered high-stakes combat prompts into OpenAI models, and then engineers changed the models based on how the answers measured up against the Geneva Conventions.
This is kind of backward finetuning step Yager said had never been taken before.
View replied-to post on X