1. The problem
Due to its thinking process… Deepseek R1 is amazing. It’s competing with Claude Sonnet and ChatGPT 4o. Only one tiny problem … the output contains its entire thought process in “THINK” tags.
2. Screenshot or task log of the problem
Me: how many Rs in "Strawberry"?
DeepSeek R1:
<think>
Alright, so I need to figure out how many times the letter 'R' appears in the word "Strawberry". Let me think about this step by step.
First, let's break down the word "Strawberry" and look at each letter one by one. The letters are S, T, R, A, W, B, E, R, R, Y. Okay, so counting them out loud might help:
S
T
R
A
W
B
E
R
R
Y
From this breakdown, I can see that the letter 'R' appears at positions 3, 8, and 9. So that makes three 'R's in total.
But wait a minute—is there any chance I missed an 'R'? Let me double-check by spelling out "Strawberry" again: S-T-R-A-W-B-E-R-R-Y. Yep, still only three 'R's. No additional 'R's are present beyond those three instances.
So after carefully examining each letter and counting them individually, it seems clear that the word "Strawberry" contains exactly three 'R's.
</think>
The word "Strawberry" contains exactly 3 Rs.
FYI … most LLMs get this wrong.
3. QUESTION:
Obviously, we do not want to include the thinking process in our articles. Is there a way SCM can: hide / filter out / ignore / block / suppress / not include … anything in between specified tags? Maybe somehow using Regex?