I must be doing something wrong, because it’s not editing my text? It’s mostly just giving me output that is exactly the same as the input. Once it added line breaks after each sentence, or deleted a word (I’ve been varying the prompt).
Example input:
The contest at Rainhill in October 1829 was held to decide how the new railway would be worked. But its consequences were destined to be much more profound and far-reaching than any one railway. What was about to happen here would change the world.
Instructions:
Rewrite these sentences to be less awkward and more streamlined.
Output:
The contest at Rainhill in October 1829 was held to decide how the new railway would be worked. But its consequences were destined to be much more profound and far-reaching than any one railway. What was about to happen there would change the world.
Output is literally the input unchanged.
I also tried the prompt “Rewrite this as a sonnet”, and it gave me back the input, and then added one sentence in a new paragraph: “It was the first time that steam had been used to propel a vehicle of any kind on a railway.” So it really doesn’t seem to be following instructions.
Model: text-davinci-edit-001
Temperature stated at default 0, I moved it to 0.5, doesn’t seem to make a difference
Using the responses to a survey, I’ve tried variations on the instruction to ‘Fix spelling mistakes’ and keeping the temperature between 0.0 and 0.5. It seems to work quite well but hallucinates short responses into python code.
For example the single response ‘None’ will generate python code.
#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
File: file.py
Author: limingdong
Date: 7/10/14
Description:
"""
import sys
reload(sys)
sys.setdefaultencoding('utf-8')
def read_file(file_path):
"""
读取文件
:param file_path:
:return:
"""
with open(file_path, 'r') as f:
for line in f:
print line
def write_file(file_path, content):
"""
写文件
:param file_path:
:param content:
:return:
"""
with open(file_path, 'w') as f:
f.write(content)
def append_file(file_path, content):
"""
追加文件
:param file_path:
:param content:
:return:
"""
with open(file_path, 'a') as f:
f.write(content)
if __name__ == '__main__':
file_path = './test.txt'
read_file(file_path)
write_file(file_path, 'test')
append_file(file_path, 'test')
In regards to the Edits capabilities, I tested it by having it fill in a “table” (in list form) of U.S. taxation information that is normally easily accessible with Chat-GPT and while it filled in all of the the information in the result, it’s answers were completely factually false/incorrect. To be fair, I didn’t do much reading through the guidelines before testing it out. But I just wanted to give you guys a heads up that it doesn’t seem to be doing as well with factual data.
This may not be the best chat area to contribute to. I am beginning experimenting with GPT-3 Edit capability. I want to have it edit an early draft of my scientific manuscript to improve the content. First try was to ask it to edit the introductory paragraphs of the manuscript. My request text was “Revise the text for a scientific paper”. Most options were set very low except Temperature was set to 0.5 I received an error message that it couldn’t be done, suggesting the Temperature might be too high. I reset Temperature to 0. It gave me back the same text that I submitted. Unfortunately my input text included writing errors such as repeated sentences and thoughts. I assume I will need to experiment with different Temperature settings as well as settings for some other options, and possibly for the request/command text.
Is it me or the edit functionality is not working at all?
I requested to compact text, reducing unnecessary words, removing punctuation marks, removing dots, removing all words, delete all words starting with A and nothing, I always get the same text.
I have a rather long text, if it matters, but I got no warning about token size or anything.
5/5. This AI language model is excellent. Revolutionary, one could say. I love what OpenAI is doing, and I one day need to give this language model a fist bump for it’s good work.
