Budget forcing
Beginner
Making an AI think shorter or longer by cutting it off, or by adding “Wait” when it wants to stop.
Novice
Controlling how long a reasoning model thinks at answer time: cut the thinking off at a limit, or suppress its end-of-thinking marker and append “Wait” to make it keep going.
Expert
A decoding-time intervention on the end-of-thinking delimiter: forced termination caps tokens; appending “Wait” extends the chain and often triggers self-checking. Introduced with s1.
See also: Inference-time scaling.