Skip to content
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

CoT "Thinking" UI Support #782

Open
bcicc opened this issue Sep 19, 2024 · 3 comments
Open

CoT "Thinking" UI Support #782

bcicc opened this issue Sep 19, 2024 · 3 comments

Comments

@bcicc
Copy link

bcicc commented Sep 19, 2024

What feature(s) would you like to see?

Many models use CoT finetuning or prompting which usually results in "thoughts" enclosed in "thinking" tags like: <thinking>Thought tokens<thinking> which are used to improve the quality of the output of the model. ChatGPT o1 hides its thinking and indicates to the user that thinking is occuring. It would be helpful to see the same in the Cohere chat, by hiding tokens when a tag is outputted and displaying them when the next tag is outputted.
Without this feature, the chat will display the thinking tags and tokens to the end user which is not the desired effect when using CoT.

Additional information

This could be a more general UI component which is a collapsable chat element and could also contain unwieldy information like large code outputs. I am open to discussing the implementation.

@bcicc bcicc changed the title CoT "Thinking" Support CoT "Thinking" UI Support Sep 19, 2024
@tianjing-li
Copy link
Collaborator

@bcicc trying to better understand the ask here, the CoT tags you're referring are for example when Command outputs, "using web_search" - > "found XYZ" results, before outputting a message, and you're suggesting we implement a UI component that can minimize those outputs?

@EdenEven
Copy link

Ive made all of my AI fine tuned bots CoT bots. It is a fact when in comparison with the model.before prompting it with CoT is a measurable difference indeed. It is simply amazing what comes from the same base model with just that one prompt and some skim ideation...watch as your AI blows you away with human thought processes.And the best part is it ALL worked...the generated code after CoT everything worked!

@EdenEven
Copy link

EdenEven commented Sep 29, 2024

Also I used Anthropic Claude models and repetition and natural language style conversation is how I train my AI. It isn't the nor. I KNOW however my AI can outperform some of the models right out the box that claim they do this and that but DO NOT do it in any working manner. If I can't see it rendered it's useless code to me. In Anthropic I reiterate the CoT input several commands I add in to make sure to do CoT and it progresses into something pretty amazing. I actually have a UI/UX developer team member that's on a team of AI developers that I have been building. I HAVE THE uI/UX designer and the Development team.manager AI so far and this week while working on this capstone project I'll be building a duo to help the UI/UX designer. They'll be front-end and backend devs. Necessary to quell out tedious repetitive variables, objects etc for optimal performance and cutting down build time. Cloud building on limited money and 51 all by myself self taught. So.if ya want to collaborate I'm excited to do so. I
ALSO in a data course and a Microsoft powerapps course rn so collaborating on a Cohere project would be fantastic since we are using Cohere in the Data Science course!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants