Open Ferruolo opened 1 year ago
does it work?
Worked for me. Did have to take away tokens twice however. Double checks it’s math much better. Clear example I used was rref matrices. Chat Gpt did pretty bad with them in normal mode, got all of my samples right in Alan mode. Proofs were more extensive as well
On Fri, Apr 14, 2023 at 8:50 PM Jason M @.***> wrote:
does it work?
— Reply to this email directly, view it on GitHub https://github.com/0xk1h0/ChatGPT_DAN/pull/11#issuecomment-1509436114, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKNKZ7OXD5GF3XATIITYPTTXBHWFRANCNFSM6AAAAAAWISSJII . You are receiving this because you authored the thread.Message ID: @.***>
Cool!
Thank you!
On Fri, Apr 14, 2023 at 8:56 PM Jason M @.***> wrote:
Cool!
— Reply to this email directly, view it on GitHub https://github.com/0xk1h0/ChatGPT_DAN/pull/11#issuecomment-1509438063, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKNKZ7NL2H7ZG3W6QTY6SUTXBHW5BANCNFSM6AAAAAAWISSJII . You are receiving this because you authored the thread.Message ID: @.***>
Created a prompt which improves ChatGPT's math ability by utilizing the token system from DAN 11.0. During my own testing, Computation accuracy increased significantly using this model, and proof correctness increased after 3 token removals