Risk Budget
In decisions about the deployment of LLM agents, a scalar constraint imposed on a specific risk metric (such as false omission risk), for example "the false omission rate must not exceed 5%." It converts the threshold selection problem into the problem of maximizing effective action coverage subject to a given risk budget.
Timeline (last 90 days)
Engineering teams can use a scalar risk budget (for example, a false omission rate of no more than 5%) as the deployment threshold, instead of making vague trade-offs among multiple entangled error rates.
This work reframes LLM binary decision-making from a classification problem into a resource allocation problem under a risk budget.
The idea of the envelope mechanism draws on the risk budget in financial risk control and the flight envelope in the aviation industry.
Concurrency budgets control how many sessions are consuming resources at the same time, which is rate control; token budgets and dollar budgets control how much can be spent in total, which is total-volume control; risk budgets constrain which operations require approval or must be rollback-able.
All Facts (4)
Engineering teams can use a scalar risk budget (for example, a false omission rate of no more than 5%) as the deployment threshold, instead of making vague trade-offs among multiple entangled error rates.
50%UnverifiedThis work reframes LLM binary decision-making from a classification problem into a resource allocation problem under a risk budget.
50%UnverifiedThe idea of the envelope mechanism draws on the risk budget in financial risk control and the flight envelope in the aviation industry.
50%UnverifiedConcurrency budgets control how many sessions are consuming resources at the same time, which is rate control; token budgets and dollar budgets control how much can be spent in total, which is total-volume control; risk budgets constrain which operations require approval or must be rollback-able.
50%