A user calls the experiment's design clever and its results “very interesting,” without giving further details.
In a quote post, a user praises an experiment's design as clever and calls its results “very interesting.” The comment does not say what the experiment tested or found.
105.1K
8 Sources, first seen ago
@pkirgisFrontier AI companies use constitutions, model specifications, and system prompts to guide the moral orientation of their models. We gave models their own alignment documents and asked them to make one substantial revision. Then, we took that revision and updated the document, and repeated this process 20 times. 🧵
@sethlazarThis is a clever experiment design with very interesting results!
@IasonGabrielFascinating research on what happens when AI models are allowed to edit their own constitutions, including shifts in the parties they favor and the authority relations they endorse:@pkirgisFrontier AI companies use constitutions, model specifications, and system prompts to guide the moral orientation of their models. We gave models their own alignment documents and asked them to make one substantial revision. Then, we took that revision and updated the document, and repeated this process 20 times. 🧵@sethlazarThis is a clever experiment design with very interesting results!@IasonGabrielFascinating research on what happens when AI models are allowed to edit their own constitutions, including shifts in the parties they favor and the authority relations they endorse:@pkirgisFrontier AI companies use constitutions, model specifications, and system prompts to guide the moral orientation of their models. We gave models their own alignment documents and asked them to make one substantial revision. Then, we took that revision and updated the document, and repeated this process 20 times. 🧵@sethlazarThis is a clever experiment design with very interesting results!@IasonGabrielFascinating research on what happens when AI models are allowed to edit their own constitutions, including shifts in the parties they favor and the authority relations they endorse:@pkirgisFrontier AI companies use constitutions, model specifications, and system prompts to guide the moral orientation of their models. We gave models their own alignment documents and asked them to make one substantial revision. Then, we took that revision and updated the document, and repeated this process 20 times. 🧵@sethlazarThis is a clever experiment design with very interesting results!@IasonGabrielFascinating research on what happens when AI models are allowed to edit their own constitutions, including shifts in the parties they favor and the authority relations they endorse:—
Not ranked yet
—
Not ranked yet
