Llama Impact Hackathon

Start

Playing with recursively self-improving AI while avoiding existential risk. We dance with basilisks.

FOOM and the Basilisks

lumpenspace

cynthialuo

V McCoy

Palinor addresses the alignment problem in AI by providing a model steering solution that is fluid and adaptive. It enables users to adjust the model's ethical boundaries dynamically based on context, offering reversible modifications that allow temporary behavioral shifts without permanently altering the model. This solution targets AI safety researchers, enterprise AI teams, and LLM developers, providing them with stackable controls, a simple Python API, and agile experimentation capabilities. Unique features include granular control over multiple control vectors and the ability to test alignments in seconds with immediate feedback, accelerating research cycles and democratizing alignment research.

Palinor offers adaptable ethical stances and reversible modifications for fluid AI alignment.

"FOOM and the Basilisks team on Llama Impact Hackathon Hackathon"

Team Idea

Submission