Grokking Agentic Design Patterns
Vote

0% completed

​

Exploration: Multiple Candidates, Scoring, and Pruning

Suppose one attempt at a task succeeds 60 percent of the time, attempts are independent, and one attempt is a 10-step run, which sends 141,000 tokens under this course's standard inputs. Running several attempts and keeping a good one gives this table.

AttemptsChance at least one is goodCalculationTokens sentMultiple of one attempt
160 percent1 - 0.4141,0001
284 percent1 - 0.4^2282,0002
393.6 percent1 - 0.4^3423,0003
5About 99 percent1 - 0.4^5705,0005

.....

.....

.....

Like the course? Get enrolled and start learning!

Reading Progress

0%

Vote for new content