4 ms·

I switched from VSCode to Doom Emacs a couple of months ago. I've been using ChatGPT (GPT-4) to help me write config (~1,000 lines of elisp code so far) and it

I switched from VSCode to Doom Emacs a couple of months ago. I've been using ChatGPT (GPT-4) to help me write config (~1,000 lines of elisp code so far) and it's been very interesting/fun. I'm not going back.

bachback 3y ago
very cool can you share more?
- Pr0ject217 3y ago
  My process is kind of organic. I'll start working on a project, and if I run into something that feels wrong about the Doom Emacs default config/keybindings, I'll just open up a conversation with ChatGPT (the GPT-4 model, it's much better than 3.5). I'll tell it that I'm using Doom Emacs, and then explain what I want, or what problem I'm having, and it will output code that I can try. It's not always correct on the first try, and sometimes I have to have a dialogue with it for a little while before it gets it right. Sometimes it struggles and begins to guess - and for those cases, I'll fall back to doing manual searches in Google, and if that fails, give up entirely and just write myself a TODO comment and keep marching forward. It's help me set up config quickly that would've taken me a very long time to figure out.
BeetleB 3y ago
That's interesting. I've asked Bing Chat and ChatGPT several times to write me elisp code, and not once did it work well. It often invents APIs.
- ParetoOptimal 3y ago
  Hm, I've had luck with openai gpt4. I've had it make fake APIs before, but usually when I say "fooFakeApi doesn't exist" it corrects itself.
  - pmoriarty 3y ago
    But how well does it do on elisp?
    - BeetleB 3y ago
      When I've informed it of fake APIs, it apologizes, and replaces it with another fake API.
      - ParetoOptimal 3y ago
        Yes. It gave me quite a lot of trouble when I asked it to replace "cut -d:" with pure elisp. Eventually after a few minutes I got it to work, but during real work I probably would have just written it myself if something so simple didn't work the first time I asked. Here's the log: ChatGPT> Replace the cut shell command with a pure elisp implementation in this code: (let* ((group-command "cut -d: -f1 /etc/group") ; command to list the groups (group-data (with-temp-buffer (shell-command group-command (current-buffer)) (sort (split-string (buffer-substring-no-properties (point-min) (point-max)) "\n") #'string<)))) (message "Selected group: %s" (completing-read "Choose a group: " group-data nil t))) <shell-maker-end-of-prompt> (let* ((group-command "cut -d: -f1 /etc/group") ; command to list the groups (group-data (sort (split-string (with-output-to-string (call-process-shell-command group-command nil standard-output)) "\n") #'string<))) (message "Selected group: %s" (completing-read "Choose a group: " group-data nil t))) ChatGPT> replace group-command with and call-process-shell-command with elisp that does the same <shell-maker-end-of-prompt> (setq group-command "cut -d: -f1 /etc/group") (setq group-data (sort (split-string (shell-command-to-string group-command) "\n") #'string<)) (message "Selected group: %s" (completing-read "Choose a group: " group-data nil t))) ChatGPT> What is the elisp equivalent of: cut -d: -f1 /etc/group <shell-maker-end-of-prompt> The equivalent Elisp code for `cut -d: -f1 /etc/group` is: (concat (nth 0 (split-string (with-temp-buffer (insert-file-contents "/etc/group") (buffer-string)) "\n")) ":") ChatGPT> Give me elisp I can evaluate to check that the elisp you gave is equal to the cut call <shell-maker-end-of-prompt> You can evaluate the following Elisp code to check that the Elisp I provided is equal to the `cut` call: (equal (concat (nth 0 (split-string (with-temp-buffer (insert-file-contents "/etc/group") (buffer-string)) "\n")) ":") (shell-command-to-string "cut -d: -f1 /etc/group")) ChatGPT> It's not equal. the cut shell command gave a list of results. The elisp you gave only gives one result <shell-maker-end-of-prompt> I apologize for the confusion. You're right, the Elisp code I provided only returns the first entry in `/etc/group`. Here's the updated code to get all group names from `/etc/group` file separated by a newline: (with-temp-buffer (insert-file-contents "/etc/group") (let ((groups (split-string (buffer-substring-no-properties (point-min) (point-max)) "\n"))) (mapconcat #'(lambda (group) (car (split-string group ":"))) groups "\n"))) When you evaluate this Elisp code, it will return all the group names separated by a newline character, as the `cut -d: -f1 /etc/group` command would. ChatGPT> Give me elisp to check the equality of it <shell-maker-end-of-prompt> You can use the following Elisp code to check the equality of the two commands: (equal (with-temp-buffer (insert-file-contents "/etc/group") (let ((groups (split-string (buffer-substring-no-properties (point-min) (point-max)) "\n"))) (mapconcat #'(lambda (group) (car (split-string group ":"))) groups "\n"))) (shell-command-to-string "cut -d: -f1 /etc/group"))
- newtwilly 3y ago
  I recently had GPT 3.5 write me a function for translating the number under point to an ISO timestamp. It handled whether the epoch number was milliseconds or seconds since epoch. I kinda had to iterate on it and ask it to do some things different, but it never wrote invalid code.

bachback 3y ago

very cool can you share more?

Pr0ject217 3y ago
My process is kind of organic. I'll start working on a project, and if I run into something that feels wrong about the Doom Emacs default config/keybindings, I'll just open up a conversation with ChatGPT (the GPT-4 model, it's much better than 3.5). I'll tell it that I'm using Doom Emacs, and then explain what I want, or what problem I'm having, and it will output code that I can try. It's not always correct on the first try, and sometimes I have to have a dialogue with it for a little while before it gets it right. Sometimes it struggles and begins to guess - and for those cases, I'll fall back to doing manual searches in Google, and if that fails, give up entirely and just write myself a TODO comment and keep marching forward. It's help me set up config quickly that would've taken me a very long time to figure out.

BeetleB 3y ago

That's interesting. I've asked Bing Chat and ChatGPT several times to write me elisp code, and not once did it work well. It often invents APIs.

ParetoOptimal 3y ago
Hm, I've had luck with openai gpt4. I've had it make fake APIs before, but usually when I say "fooFakeApi doesn't exist" it corrects itself.
- pmoriarty 3y ago
  But how well does it do on elisp?
  - BeetleB 3y ago
    When I've informed it of fake APIs, it apologizes, and replaces it with another fake API.
    - ParetoOptimal 3y ago
      Yes. It gave me quite a lot of trouble when I asked it to replace "cut -d:" with pure elisp. Eventually after a few minutes I got it to work, but during real work I probably would have just written it myself if something so simple didn't work the first time I asked. Here's the log: ChatGPT> Replace the cut shell command with a pure elisp implementation in this code: (let* ((group-command "cut -d: -f1 /etc/group") ; command to list the groups (group-data (with-temp-buffer (shell-command group-command (current-buffer)) (sort (split-string (buffer-substring-no-properties (point-min) (point-max)) "\n") #'string<)))) (message "Selected group: %s" (completing-read "Choose a group: " group-data nil t))) <shell-maker-end-of-prompt> (let* ((group-command "cut -d: -f1 /etc/group") ; command to list the groups (group-data (sort (split-string (with-output-to-string (call-process-shell-command group-command nil standard-output)) "\n") #'string<))) (message "Selected group: %s" (completing-read "Choose a group: " group-data nil t))) ChatGPT> replace group-command with and call-process-shell-command with elisp that does the same <shell-maker-end-of-prompt> (setq group-command "cut -d: -f1 /etc/group") (setq group-data (sort (split-string (shell-command-to-string group-command) "\n") #'string<)) (message "Selected group: %s" (completing-read "Choose a group: " group-data nil t))) ChatGPT> What is the elisp equivalent of: cut -d: -f1 /etc/group <shell-maker-end-of-prompt> The equivalent Elisp code for `cut -d: -f1 /etc/group` is: (concat (nth 0 (split-string (with-temp-buffer (insert-file-contents "/etc/group") (buffer-string)) "\n")) ":") ChatGPT> Give me elisp I can evaluate to check that the elisp you gave is equal to the cut call <shell-maker-end-of-prompt> You can evaluate the following Elisp code to check that the Elisp I provided is equal to the `cut` call: (equal (concat (nth 0 (split-string (with-temp-buffer (insert-file-contents "/etc/group") (buffer-string)) "\n")) ":") (shell-command-to-string "cut -d: -f1 /etc/group")) ChatGPT> It's not equal. the cut shell command gave a list of results. The elisp you gave only gives one result <shell-maker-end-of-prompt> I apologize for the confusion. You're right, the Elisp code I provided only returns the first entry in `/etc/group`. Here's the updated code to get all group names from `/etc/group` file separated by a newline: (with-temp-buffer (insert-file-contents "/etc/group") (let ((groups (split-string (buffer-substring-no-properties (point-min) (point-max)) "\n"))) (mapconcat #'(lambda (group) (car (split-string group ":"))) groups "\n"))) When you evaluate this Elisp code, it will return all the group names separated by a newline character, as the `cut -d: -f1 /etc/group` command would. ChatGPT> Give me elisp to check the equality of it <shell-maker-end-of-prompt> You can use the following Elisp code to check the equality of the two commands: (equal (with-temp-buffer (insert-file-contents "/etc/group") (let ((groups (split-string (buffer-substring-no-properties (point-min) (point-max)) "\n"))) (mapconcat #'(lambda (group) (car (split-string group ":"))) groups "\n"))) (shell-command-to-string "cut -d: -f1 /etc/group"))
newtwilly 3y ago
I recently had GPT 3.5 write me a function for translating the number under point to an ISO timestamp. It handled whether the epoch number was milliseconds or seconds since epoch. I kinda had to iterate on it and ask it to do some things different, but it never wrote invalid code.