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

What is the total number of respondents?→ 506 What is the total number of respondents?→ 506 How many men have responded?→ 249 How many women have responded?→ 257 What is the calories in pat of butter?→ 50 calories / 50 What % of women don't know the calories in a teaspoon of sugar?→ 18 What % of the respondents said No ?→ 86 What % of the men said No ?→ 90 What % of the women said No ?→ 81 What does the symbol * indicate?→ LESS THAN 0.5% / Less than 0.5%Of the total respondents what percentage use sugar with foods other than beverages?→ 63% / 63 Of the total male respondents what percentage use sugar with foods other than beverages?→ 53% / 53 What percentage of female respondents use sugar with foods other than beverages?→ 72% / 72

Nearby prompts. All 5,330 DocVQA prompts

PromptDocVQA · pagelpdl0226_15.png

What % of the respondents said No ?

OCR text of the page · 269 characters; the scanned image itself is not published

Question 4: Do you make a conscious effort to count the
calories in a day's time?
Total
Men
Women
Total Respondents
(506-100%)
(249=1.00%%) (257=100% )
Yes
9%
19%
NO
86
90
81
No. answer
*
1
* Less than 0.5%
Source: https:7/www.industrydocuments. ucsf. edu/docs/Ipd10226

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

Ours

Partial · ANLS 0.50

Aux 2015

Wrong

PiT-FT 2015

Wrong

Empty response.

ChronoGPT 2015

Wrong

Question: What % of the respondents said No ?

OCR text: Question 4: Do you make a conscious effort to count the calories in a day's time? Total Men Women Total Respondents (506-100%) (249-100%) (257-100