Basis Lab

Many Models, One Job

Three Points of Accuracy for Twice the Cost

Last timeWhen It Will Not Stop

The four step version is right 81 times in 100 against 78 for one call, at twice the cost and 2.3 times the delay, and whether that is worth buying is arithmetic.

Both versions of this service now exist. One sends the whole job to a single

call, which gets 3,000 words of instruction, order record and policy and returns

  1. The other does it in four steps, sorting the request, finding the matching

order and policy, drafting the reply and checking the draft against the policy.

Somebody has graded the replies of both on the same hundred customer questions.

The single call version is right 78 times in 100. The four step version is right

81.

This is the only comparison that settles anything, and it is the one almost

nobody runs, because by the time the four step version exists there is nobody

left who wants the answer to be no.

The three numbers, in the units they are paid in

Both costs come from the same prices. Input is 3 units per million words and

output is 15. The single call sends 3,000 and returns 600, so it is

3000×3/106=0.0093000 \times 3 / 10^6 = 0.009 in and 600×15/106=0.009600 \times 15 / 10^6 = 0.009 out, which

is 0.018, and it takes 1.4 seconds. Each of the four steps sends 1,500 and

returns 300, so each is 0.0045 plus 0.0045, which is 0.009 and 0.8 seconds, and

four of them are 0.036 and 3.2 seconds.

FIG 1
versionright in 100cost a requestsecondscost a day at 10,000 requests
one call780.0181.4180
four steps810.0363.2360

Three points of accuracy, exactly twice the cost, and 3.2/1.4=2.33.2 / 1.4 = 2.3 times

the delay. Every argument from here is about those three differences, and the

first thing to establish is whether the first of them exists at all.

The lesson stops here

10 more paragraphs to go

You have read the opening. The rest of the argument, the problems that check whether it landed, and the lines worth keeping at the end all come with a plan.

The first lesson of every course in the library reads the whole way through, free, so you can see exactly what the rest of them are.

See the planThe contents

This is the reading half

Starting the course gives you your own copy of it. Every idea on every page has problems standing under it, marked with a reason rather than a tick, and any sentence you do not believe can be opened and argued with. None of that can happen on a page nobody owns.

The contents

The rest of this course

  1. 01What Splitting a Job Across Model Calls Buys, and What It Charges
  2. 02A Step Good 95 Times in 100 Is Not Good Enough to Be the Fourth of Fouropening only
  3. 03Chain, Fork, Loop and Crowd, Priced on One Jobopening only
  4. 04Who Decides the Next Step, and What Deciding Costsopening only
  5. 05What One Step Has to Tell the Next, and What It Must Notopening only
  6. 06What a Gate Catches, What It Destroys and What It Costsopening only
  7. 07The Loop That Will Not Settle, and the Loop That Settles on Something Wrongopening only
  8. 08Three Points of Accuracy for Twice the Costyou are here

Read alongside