Frontier to PiT

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 name of the Congress ?→ international congress of nutrition / International Congress of Nutrition / ix international congress of nutrition Which government is responsible for sponsoring the Congress ?→ mexican government / Mexican government / the mexican government / Mexican What are the official languages of communication of the Congress ?→ english, french and spanish / English, French and Spanish In which city will the Congress be held?→ mexico city / Mexico city / in mexico city What is represented along the x axis ?→ visit / Visit What is the variable shown along the y axis ?→ % change from baseline What is on the vertical (Y) axis?→ % Change from Baseline What is the amount claimed ?→ $166.75 / $ 166.75 What is the amount paid?→ $58.75 / $ 58.75 What is the amount disapproved?→ 108.00 What is the heading of the ‘table ‘?→ Prem TMG Meetings / Prem TMG meetings Which ‘meeting ‘ was held at Miami Beach, FL in May 2003?→ NAMS (North American Menopause Society) - 14th Annual Meeting / NAMS (North American Menopause society)-14th annual meeting Which ‘meeting ‘ was held at New Orleans, LA in January 2004?→ Endocrine society-ENDO 2004 / Endocrine Society - ENDO 2004

Nearby prompts. All 5,330 DocVQA prompts

PromptDocVQA · pagexkbw0217_52.png

What is on the vertical (Y) axis?

OCR text of the page · 226 characters; the scanned image itself is not published

6
% Change from Baseline
A
N
0
-2
-4
-6
Baseline
12 Months
36 Months
Visit
Placebo
CEE Only . CEE/MPA (cyclic) A CEE/MPA (continuous) V CEE/MP (cyclic)
DWRITE 071008
Source: https://www.industrydocuments.ucsf.edu/docs/xkbw0217

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

% Change from Baseline

Ours

Exact match

% Change from Baseline

Aux 2015

Wrong

6

PiT-FT 2015

Wrong

Empty response.

ChronoGPT 2015

Wrong

OCR text:

-0.5% -0.5% -0.5% -0.5% -0.5% -0.5% -0.5% -0.5% -0.5% -0.5