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

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

http://github.com/semiosis/engines

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

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. 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.