-
Notifications
You must be signed in to change notification settings - Fork 93
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add tool usage #291
Comments
Ah, nice!
I may wait a little to see how things develop.
Great! Let's use this issue to discuss approach when we get to it. |
Here's the announcement: https://www.reddit.com/r/emacs/comments/1hqyuvn/getting_llms_to_do_things_tool_use_with_gptel/. Karthink made a nice blog post on it as well: https://karthinks.com/software/tool-use-with-gptel-looking-for-testers/ The downside of tool use is that you have to tell the model about each tool you want it to be able to use so there will always be many functions missing. An alternative I thought about was simply generating code. That might be ok for some simple tasks but has the issues that LLM code is often buggy (especially when it's trying to do something uncommon) and it would be necessary to have the user review the code before actually running it. |
Yep. I'd be weary of freely allowing elisp execution. Btw, you can somewhat accomplish this by asking for elisp code and C-c C-c on it. On a somewhat related tangent, it may be interesting to experiment with attaching additional actions to code block responses (requiring user intervention).
s/'/` |
ps. If you compose on selected text and press i on a response code block, it attempts to paste pack to origin. |
Yeah, having to review the elisp before executing it would probably make it more trouble than it's worth. The i binding is nice. |
Karthink announced today that gptel has experimental support for tool use now.
Have you thought about adding it to chatgpt-shell? I would be interested in contributing.
The text was updated successfully, but these errors were encountered: