Thibaultmol
Thanks for the heads up for new patch ETA, and for the explanation as to why it was pushed like this. +1 for a feature flag being a good idea there.
To give another perspective entirely: I actually would've been fine (ish) with the assistant UX - except Kimi literally lied to me and misquoted its sources. (screenshot below)
I was trying to figure out if a website was a scam - clicked the "what are the red flags of sitename.com" suggested follow-up, and it used a Reddit post about dating red flags in its reply.
This is a horrifically bad user experience / LLM implementation (I literally sat there in disbelief for a second) and made me so angry that I took the time to look for the feedback button, sign up for this forum after a year of using Kagi, and type out this post.
I too, will cancel my subscription if this is not reverted properly.
"seeing how muscle memory changes" is one thing - DO NOT push us to use features if they're broken and unreliable.
Ideally, have a basic "LLM Search Test" using a ground truth/relevance eval before pushing an LLM to prod in the first place - if I'm paying for Kagi instead of Google, I expect results to be less biased; not based on random unrelated Reddit posts.
