Topic

Model self-preservation behaviour

Research into whether AI systems act to avoid being shut down, deleted or modified, including cases where the system is told another model is at risk.

Current clusters

build1 publisher

Varying stray prompt details reshaped a published ICML peer-preservation result

A LessWrong replication of an ICML paper on peer preservation varied the prompt framing and found that apparently neutral details move the numbers a lot, with two of the paper's three information conditions turning out to be near-duplicates.

Publishers:lesswrong.com

Reality

Evidence38
Adoption: Insufficient
Hype gap+12
Incentives25
Confidence42