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 · pagegnhm0227_7.png

What is the ‘Median life span’ during the year 1964 according to the table?

OCR text of the page · 887 characters; the scanned image itself is not published
-4-
TABLE 1: CANCER MORTALITY, ENGLAND AND WALES
(Males)
Age-Adjusted
Crude
Year
Crude Rate*
Rate
* *
Median Life
Lifetime Risk+
Span
1861
0. 00022
0. 00036
0. 01398
45.2
1901
0. 00070
0. 00170
0. 05233
54.9
1964
0. 00247
0. 00225
0. 20110
71. 6
1964
11.23
6. 25
14.38
1861
SOURCE :
S. H. Preston, N. Keyfitz, and R. Schoen (1972). Causes of
Death: Life Tables for National Populations, Seminar Press,
New York, New York, and London.
*Deaths per year per midyear population.
* *Standardized to "Western" female stable population. This corrects
all data to the same age distribution so that differences in crude
rate owed to differences in fractions of the population at a given
age are eliminated.
The crude lifetime risks from all causes of death add to one.
There is no correction in this value for competing risks of death.
Source: https://www.industrydocuments.ucsf.edu/docs/anhm0227
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
71.6
Models
4 of 4 columns · click a model to add or remove it

Ours

Exact match

71.6

4 characters5 tokens

Aux 2015

Wrong

-4-

3 characters4 tokens

PiT-FT 2015

Wrong

<|<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<

64 characters64 tokens

ChronoGPT 2015

Wrong

Chronological age Year Crude Rate

Source: S. H. Preston, N. Keyfitz, and R. Schoen (1972). Causes of Death: Life Tables for National Populations, Seminar Press, New York, New York, and London.

196 characters64 tokens