English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

This report examines whether advanced AIs that perform well in training willbe doing so in order to gain power later -- a behavior I call "scheming" (alsosometimes called "deceptive alignment"). I conclude that scheming is adisturbingly plausible outcome of using baseline machine learning methods totrain goal-directed AIs sophisticated enough to scheme (my subjectiveprobability on such an outcome, given these conditions, is roughly 25%). Inparticular: if performing well in training is a good strategy for gaining power(as I think it might well be), then a very wide variety of goals would motivatescheming -- and hence, good training performance. This makes it plausible thattraining might either land on such a goal naturally and then reinforce it, oractively push a model's motivations towards such a goal as an easy way ofimproving performance. What's more, because schemers pretend to be aligned ontests designed to reveal their motivations, it may be quite difficult to tellwhether this has occurred. However, I also think there are reasons for comfort.In particular: scheming may not actually be such a good strategy for gainingpower; various selection pressures in training might work against schemer-likegoals (for example, relative to non-schemers, schemers need to engage in extrainstrumental reasoning, which might harm their training performance); and wemay be able to increase such pressures intentionally. The report discussesthese and a wide variety of other considerations in detail, and it suggests anarray of empirical research directions for probing the topic further.

Scheming AIs: Will AIs fake alignment during training in order to get power?