If you don’t use vi/vim, you might not find this post very practical, but maybe it’ll convince you to try it out!
I have been a vim fan for a long time, but a few years ago I watched this video from Tomnomnom and Stok. It taught me how to pipe stdin into a vim buffer:
echo "helly world" | vim -
It’s also where I learned that you could be inside vim, but manipulate the entire file as if you were piping the contents of the file into a command. The output of the command does in-line replacement of the entire file with those changes. That sounds confusing, but it just means you can be inside a vim file and do :%!grep test and it’ll remove all lines that don’t contain test, for example.
This post is a simple showcase of taking that concept, but throwing an llm into the mix to add more dynamic functionality.
Leveraging vim’s Visual Mode
For those unfamiliar, vim has a feature called ‘visual mode’. This allows you to select some text, which can then be manipulated in various ways. One such way is hitting the colon key, followed by an exclamation point. This allows you to run a shell command using the selected text as input from stdin. Then the output from the command is written directly into the file, replacing the selected text right before your eyes. Here’s a simple sort example.
Your browser does not support the video tag.
Using llm to sort lines in vim
Applying llm to this feature
This might sound complex, but it’s not too bad (and super useful). When combined with the capabilities of an LLM, it becomes a superpower.
Imagine, for instance, that you’ve written a piece of text with some typos in it. Rather than having to manually go through and correct each one, you could simply select the text in vim, run a command using an LLM, and have all of the errors automatically corrected.
Or perhaps you’ve written a piece of code, but forgot to add comments explaining what it does. Again, you could select the code in vim, run a command using an LLM, and have comments automatically generated and inserted into your code.
You could even use an LLM to help finish a sentence or paragraph that you’re struggling with live inside your text editor. Simply select the text you’ve written so far, run a command using an LLM, and have the rest of the sentence or paragraph automatically generated for you. Naturally with all of these, there is a bit of latency for the calls to the llm.
Your browser does not support the video tag.
Using llm to comment some code
Your browser does not support the video tag.
Using llm to finish a paragraph
Caveats
There are a few caveats. Vim doesn’t load .bashrc/.zshrc or .bash_profile/.zsh_profile for its command execution. It does however, load the env file. So you will need to put your functions in .zshenv.
You will also have to install the command line tool llm by Simon Willison if you don’t have it. Here is a link to the github repo: https://github.com/simonw/llm/
Also, llms don’t always reflect the entire input into the response (and they may include some preamble about their response) so your llm calls will need to account for that. To make it easy to understand and replicate, I’ve put my .zshenv functions below. Simple add these to your .zshenv and they’ll start working in vim:
comment(){
llm -s 'Add comments to this code. Respond with the code and comments. Do not alter the functional aspect of the code, but still return it. Be sure and include the code in the response. Do not respond in a markdown code block. Just respond with the code and comments. Do not preamble or say anything before or after the code. for example: If the user sent "print(1)nprint(2)", you would reply "# Prints 1nprint(1)n# Prints 2nprint(2)"' - o temperature .2
}
fix(){
llm -s 'Fix the syntax of this code. Respond with the code including any fixes. Do not alter the functional aspect of the code, but simply fix it and respond with all of it. Do not respond in a markdown code block. Just respond with the code. Do not preamble or say anything before or after the code. for example: If the user sent "print(1", you would simply reply "print(1)"' -o temperature .2
}
edit(){
llm -s 'Edit this text to remove unnecessary filler words such as "like", "you know", and unimportant adverbs. Respond with the edited text only. Do not alter the speaking style or primary content.' -o temperature .1
}
blog(){
llm -s 'Write a blog about the topic from the user as a wise and succinct writer such as Paul Graham or Tyler Cowen, but only use high school term paper vocabulary or lower.' -o temperature .4 -o presence_penalty .2 -m gpt-4
}
finish(){
message=`cat`
echo -n $message
echo -n $message | llm -s 'Finish this input. Respond with only the completion text. Do not respond with the input. Do not preamble or say anything before or after the completion. For example: If the user sent "The sky is", you would simply reply " blue." If the input is code, write quality code that is syntactically correct. If the input is text, respond as a wise, succinct writer such as Paul Graham or Tyler Cowen, but only use high school term paper vocabulary or lower.' -o temperature .4 -o presence_penalty .2 -m gpt-4
}
Obviously, there are many more ways to use this functionality. I just wanted to share how I’m using it. If you set this up, let me know what you come up with.
Peace
I really hope you enjoyed these tips. If you’re a vim and llm user (or willing to try them), I think it will really open your mind up to what is possible with these tools!
– rez0
For more of my thoughts on hacking and other stuff, follow me on twitter.