iCreatAI News
Skip to content
Trending storyUpdating

Redwood Research proposes a "capability activity" definition scheme

1 reports1 sources4 hr agoUpdated

Understand this story

AI overview

Redwood Research proposed a "capability activity" definition scheme for controlling the pace of AI development: by default, all activities are treated as capability activities, with exemptions only for four categories—idle compute, narrow product development, customer service, and narrow security activities. The scheme follows the principle of "false positives over false negatives," requires third-party audits to ensure enforcement, and needs to be dynamically adjusted as companies engage in circumvention behavior. Under the scheme, only these four categories of activities may be exempt from capability activity-related controls; all other activities are treated as capability activities. Redwood Research said it prefers over-classification to under-classification, so the default is strict; enforcement relies on third-party audits, and the scope of the definition also needs to be updated as corporate circumvention methods change.

AI overview based on source reports · Updated 2 hr ago

Coverage timeline

Follow the coverage to explore different perspectives.

Nov 18
  1. Redwood Research Blog
    How should we define “capabilities activities” for the purpose of pacing?

    A Redwood Research blog post by Alek Westover and Nathan Sheffield proposes a standardized definition for classifying which uses of compute count as "capabilities activities," so that frontier AI progress could be paced by limiting compute spent on them. The definition treats all AI company activities as capabilities by default, with exceptions only for idle compute, narrow product development, serving customers, and narrow safety activities, and it favors false-positives over false-negatives. The authors argue third-party auditors and procedures to adjust the definition would be needed for it to work in practice.

Activity for this story

Current comparable activity
9
Peak comparable activity
10Nov 18 07:00
Change in the past 24 hours
–

Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.