Output Explorer

Every prompt in the paper, and what each model wrote back.

Answer a question about a scanned page from its OCR text. Scored by exact match and ANLS against the accepted answers.

13 of 5,330 prompts

Nearby prompts. All 5,330 DocVQA prompts

PromptDocVQA · pagegfhd0082_9.png

What is the Page Number?

OCR text of the page · 1,113 characters; the scanned image itself is not published
22.
What other brands have you smoked regularly in the last five years, even if you no longer
smoke them? ( RECORD ONLY TWO)
REG-
ULAR/
EXTRA
KING
LONG
FULL BRAND NAME
FELT
ULT
NM
8 5MM
100's
120 's
none
30
31
What other brands of cigarettes have you ever tried and no longer smoke? (RECORD ONLY TWO)
REG-
ULAR/
EXTRA
KING
LONG
FULL BRAND NAME
FELT
ULT
M
NM
8 5MM
100 's
120 's
Lucky Strike
32
33
24. Is there one other brand you would like to buy?
REG-
ULAR/
EXTRA
KING
LONG
FULL BRAND NAME
FE
FELT
ULT
M
NM
8 5MM
100's
120's
-6
34
25. Are there any brands you would most definitely not want to buy?
(RECORD ONLY TWO)
REG-
ULAR/
EXTRA
KING
LONG
FULL BRAND NAME
FF
FELT
ULT
NM
8 5MM
100's
120 's
-8.
35
36
26.
(IF RESPONDENT GAVE NO ANSWERS IN Q. 20-25, CHECK BELOW AND USE EITHER VIRGINIA SLIMS OR
BENSON & HEDGES IN CHIP QUESTION AS THE $2 BRAND. IF ONE OF THESE BRANDS IS THE RESPONDENT'S
USUAL BRAND (Q. 19) USE THE OTHER BRAND. )
37-1
Virginia Slims non-menthol used in
. . . . .. . ....
. chip question.
-2
Benson & Hedges non-menthol used in
chip question.
. . .
- . ... . .
. '. -.. ." . .
9926 LSSOS
- 9.
System prompt · identical for every setup
Answer the question using only the OCR text from a single document page. Return only the answer, with no explanation. Preserve the answer wording from the OCR text when possible.
Expected answer
9- 9 -
Models
4 of 4 columns · click a model to add or remove it

Ours

Wrong

9926 LSSOS

10 characters8 tokens

Aux 2015

Wrong
  1. What is the Page Number?

Answer: 22.

REG-ULAR/EXTRA KING LONG FULL BRAND NAME FE FELT ULT M NM 8 5MM 100 's 120 's LUCKY STRIKE 32 33

139 characters64 tokens

PiT-FT 2015

Wrong

Empty response.

0 characters

ChronoGPT 2015

Wrong
63 characters64 tokens