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 · pagettwl0228_4.png

Which company name is mentioned at the bottom?

OCR text of the page · 3,395 characters; the scanned image itself is not published
Unsettled Science
Knowing that weather forecasts are reliable for a
Moreover, computer models relied upon
few days at best, we should recognize the enor-
by climate scientists predict that lower atmos-
mous challenge facing scientists seeking to pre-
pheric temperatures will rise as fast as or faster
dict climate change and its impact over the next
than temperatures at the surface. However, only
century. In spite of everyone's desire for clear
within the last 20 years have reliable global
answers, it is not surprising that fundamental
measurements of temperatures in the lower at-
gaps in knowledge leave scientists unable to
mosphere been available through the use of
make reliable predictions about future changes.
satellite technology. These measurements show
A recent report from the National Re-
little if any warming.
search Council (NRC) raises important issues,
Even less is known about the potential
including these still-unanswered questions:
positive or negative impacts of climate change.
(1) Has human activity al-
In fact, many academic
ready begun to change
Sargasso Sea Temperature
studies and field experi-
temperature and the cli-
78-
ments have demonstrated
mate, and (2) How signifi-
77
Medieval
that increased levels of car-
cant will future change be?
varm period
76 -
Little
bon dioxide can
The NRC report con-
ice age
crop and forest growth.
firms that Earth's surface
So, while some argue
temperature has risen by
that the science debate is
about 1 degree Fahrenheit
73
settled and governments
over the past 150 years.
should focus only on near-
Some use this result to
72
71 .
OF
term policies-that is empty
claim that humans are
rhetoric. Inevitably, future
causing global warming,
70 -
scientific research will help
and they point to storms or
1000
500
0
- B.C. A.D. -
500 1000 1500 2000
us understand how human
floods to say that danger-
actions and natural climate
ous impacts are already
Source: Science (1996)
change may affect the world
under way. Yet scientists remain unable to con-
and will help determine what actions may be de-
firm either contention.
sirable to address the long-term.
Geological evidence indicates that climate
Science has given us enough information
and greenhouse gas levels experience significant
to know that climate changes may pose long-
natural variability for reasons having nothing to
term risks. Natural variability and human activity
do with human activity. Historical records and
may lead to climate change that could be signif-
current scientific evidence show that Europe and
North America experienced a medieval warm
cant and perhaps both positive and negative.
Consequently, people, companies and govern-
period one thousand years ago, followed cen-
ments should take responsible actions now to
turies later by a little ice age. The geological
address the issue.
record shows even larger changes throughout
One essential step is to encourage devel-
Earth's history. Against this backdrop of large,
opment of lower-emission technologies to meet
poorly understood natural variability, it is impos-
our future needs for energy. We'll next look at
sible for scientists to attribute the recent small
the promise of technology and what is being
surface temperature increase to human causes.
done today.
ExonMobil'
www.exxon.mobil.com
2000 Exxon Mobil Corporation
Source: https://www.industrydocuments.ucsf.edu/docs/ttw10228
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
Exxonmobil
Models
4 of 4 columns · click a model to add or remove it

Ours

Exact match

ExxonMobil

10 characters4 tokens

Aux 2015

Wrong

Unsettled Science

Knowing that weather forecasts are reliable for a Moreover, computer models relied upon few days at best, we should recognize the enor- by climate scientists predict that lower atmos- mospheric challenge facing scientists seeking to pre- dict climate change and its impact over the next cent

310 characters64 tokens

PiT-FT 2015

Wrong

<|<<> <|<<> <|<<> <|<<> <|<<> <|<<> <|<<> <|<<> <|<<> <|<<> <|<<

64 characters64 tokens

ChronoGPT 2015

Wrong

OCR text: Unsettled Science The NRC report has raised important questions about the future of climate change.

Question:

What is the most important question about climate change?

OCR text: OCR text: OCR text:

214 characters64 tokens