Summary

I demonstrate generating perl one-liners in emacs using a prompt function and explain parts of the prompt.

<:pp>perl -

This enables you to specify the first part of the output. <:pp> is a placed where you want to include the rest of the prompt in the beginning of the output.

split-patterns

README
http://github.com/semiosis/prompts/

split-patterns is a field for describing patterns to separate multiple prompt results from one output if the LM multiplexed the results intentionally or unintentionally.

engine: OpenAI Davinci

README
http://github.com/semiosis/engines

This specifies the engine which specifies both the model and the shell command that calls the API.

Demo

Prompt

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

title: Generate perl command
prompt-version: 2
prompt: |
  perl

  Sort by line length
  perl -e 'print sort { length($b) <=> length($a) } <>'
  ###
  Increment all numbers by one
  perl -pe 's/(\d+)/ 1 + $1 /ge'
  ###
  Reverses paragraphs
  perl -00 -e 'print reverse <>'
  ###
  Look for duplicated words in a line
  perl -0777 -ne 'print "$.: doubled $_\n" while /\b(\w+)\b\s+\b\1\b/gi'
  ###
  <1>
  <:pp>perl -  
engine: OpenAI Davinci
temperature: 0.3
max-tokens: 60
top-p: 1
stop-sequences:
- "###"
vars:
- description
n-collate: 2
split-patterns:
- "\n"
examples:
- Make all occurrences of Steve lowercase

Example outputs

These outputs appear to not all be complete

Extract URLs.

Close, but not perfect.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

perl -0777 -e 'print "$.: $_" while /<a href="\S+">\S+<\/a>/gi'
# perl -0777 -ne 'print "$.: $_" while /\b(http|https|ftp|gopher):\/\/[-a-z0-9+&@#\/%?=~_|!:,.;]*[-a-z0-9+&@#\/
# perl -0777 -ne 'print "$.: $_" while /\b(https?|ftp|file):\/\/[-a-z0-9+&@#\/%?=~_|!:,.;]*[-a-z0-9+&@#\/%=~
# perl -0777 -ne 'print "$.: $_" while /\b(https?|ftp|file)://[-a-z0-9+&@#\/%?=~_|!:,.;]*[-a-z0-9+&@#\/%=~
perl -0777 -ne 'print "$.: $_" while /<a href="(.+?)">/gi'
perl -0777 -ne 'print "$.: $_" while /<a href="(.*?)">(.*?)<\/a>/gi'
# perl -0777 -ne 'print "$.: $_" while /<a href="(.*?)">(.*?)<\/a
perl -0777 -ne 'print "$.: $_" while /\b(http|https|ftp|gopher|file|news|nntp):\b/gi'
perl -0777 -ne 'print "$.: $_\n" while /<a href="(.*?)">(.*?)<\/a>/gi'
# perl -0777 -ne 'print "$.: $_\n" while /<a href="(.*?)">(.*
perl -0777 -ne 'print "$.: $_" while /<a href="(.*?)"/gi'
perl -0777 -e 'print "$.: $_" while /<a href="(.*?)">(.*?)<\/a>/gi'

increment the first number

I will need to also have a validation filter for prompt outputs.

1
2
3
4
5
6
7

perl -pe 's/(\d+)/ 1 + $1 /ge'
perl -0777 -ne 'print "$.: doubled $_\n" while /\b(\d+)\b/gi'
perl -pe '$_ = $. + 1'
perl -0777 -ne 'print "$.: doubled $_\n" while /\b(\d+)\b\s+\b\1\b/gi'
# perl -0777 -ne 'print "$.: doubled $_\n" while /\b(\d+)\b
perl -e '$_=reverse'
perl -e '$_=reverse; s/\d+/1+$_/ge'

Much better:

However, none of the one-liners did the desired thing.

There is at least one URL I can see at the top which should be found.

https://openaiapi-site.azureedge.net/public-assets/d/19ff159b15/favicon.svg

Codex, however, could accomplish this with mere descriptions. It’s still very important to design, encode, share and curate these ‘descriptions’. GPT-3 is still very useful practice.