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 · pagejzbn0226_41.png

If priority is High what is the next step?

OCR text of the page · 672 characters; the scanned image itself is not published
Attachment C
PRIORITIZATION PROCESS STEPS
Public Comment
81-3 Batches
Progression of Information Gathering
1. Reasonably available information
2. Voluntary Call-ins
S. Section Es and 8d rules
* is information sufficient
4. Section & niles, orders, consent agreements
for priority decision?
If necessary; using thered approach
* Is information sufficient
for scoping?
is additional spormation
Soneeded to designate?
Yes Law...
Gather/develop
information
Initiate Priority Designation
Propose Priority Designation
Finalize Priority Designation
High Priority
Low Priority
Risk Evaluation
Subject to Judicial Review
Source: https://www.industrydocuments.ucsf.edu/docs/jzbn0226
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
RISK EVALUATIONrisk evaluation
Models
4 of 4 columns · click a model to add or remove it

Ours

Wrong

Propose Priority Designation

28 characters6 tokens

Aux 2015

Wrong

Attachment C PRIORITIZATION PROCESS STEPS Public Comment 81-3 Batches Progression of Information Gathering

  1. Reasonably available information
  2. Voluntary Call-ins S. Section Es and 8d rules
  • is information sufficient
  1. Section & niles, orders
246 characters64 tokens

PiT-FT 2015

Wrong

Empty response.

0 characters1 tokens

ChronoGPT 2015

Wrong

High Priority

Question:

If priority is High what is the next step?

OCR text: Attachment C PRIORITIZATION PROCESS STEPS Public Comment 81-3 Batches Progression of Information Gathering

  1. Reasonably available information
  2. Vol
232 characters64 tokens