Document/Pasted content retrieval is very bad compared to Claude
under review
l
lindi
I'm trying to give Merlin some HTML code and extract all the URLs from it from a product list. Merlin struggles to find more than 2-3 product urls (using Claude 3.5 sonnet), while on Claude Pro (using 3.5 sonnet) it gets the job done first try.
I feel like there is something very wrong with how Merlin uses pasted content and I've had this experience before even with text only content.
I assume Merlin summarizes the pasted content to cut down tokens but this is really bad when it comes to stuff like this, especially coding as you can't just ignore chunks of the code...
Please do something about it. I was really considering going all in on Merlin and cancel my Claude Pro subscription but I can't commit to that if Merlin can't do stuff like this...
S
Siddhartha
under review
S
Siddhartha
Hey lindi, we have just launched long context mode to solve for this. When you will past snippets now, it will guide you to do the same. Basically just select the long-context models and things will be passed upto 100K tokens.