Commit
•
29e5c25
1
Parent(s):
f77f672
Update README.md
Browse files
README.md
CHANGED
|
@@ -10,12 +10,31 @@ tags:
|
|
| 10 |
---
|
| 11 |
# relive-qa
|
| 12 |
|
| 13 |
-
Re-hosting questions from
|
| 14 |
while we discuss ways to continue posting articles + Q&A using a semi-automated format.
|
| 15 |
|
| 16 |
Suggestion: en.WikiNews.org
|
| 17 |
|
| 18 |
-
##
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
Citation:
|
| 21 |
|
|
|
|
| 10 |
---
|
| 11 |
# relive-qa
|
| 12 |
|
| 13 |
+
Re-hosting questions from RealtimeQA
|
| 14 |
while we discuss ways to continue posting articles + Q&A using a semi-automated format.
|
| 15 |
|
| 16 |
Suggestion: en.WikiNews.org
|
| 17 |
|
| 18 |
+
## Scraper version
|
| 19 |
+
|
| 20 |
+
Prerequisites: `pip install openai lxml cssselector requests xmltodict` and OpenAI API key
|
| 21 |
+
|
| 22 |
+
I've added two scripts:
|
| 23 |
+
- scrape.py : base script to load plain text from the latest WikiNews articles
|
| 24 |
+
- scrape_with_openai.py : pass scraped text to OpenAI's GPT-4o to generate questions and answers for each article
|
| 25 |
+
|
| 26 |
+
An LLM evaluated on this Q&A could read the WikiNews summary, the links collected from the story, or do its own web searches.
|
| 27 |
+
|
| 28 |
+
### Issues
|
| 29 |
+
|
| 30 |
+
- Prompt needs review to avoid making answer choices which are obvious or more specific than alternative choices.
|
| 31 |
+
- Questions and answers should be based on new information in the article, and not general knowledge.
|
| 32 |
+
- Links and the title may give away the answer as the subject of the article, rather than using reading comprehension.
|
| 33 |
+
- WikiNews articles may be niche or local stories, where facts are not known to an LLM unless it reads the specific article
|
| 34 |
+
|
| 35 |
+
## Original paper and dataset
|
| 36 |
+
|
| 37 |
+
https://github.com/realtimeqa/realtimeqa_public
|
| 38 |
|
| 39 |
Citation:
|
| 40 |
|