I submitted the following to Research Assistant "Looking for a 104 key 100% mechancial keyboard for non gaming use/ Need hot swappable keys and customer service is important. Max price is $150 CAD"
It came back with a recommendation of Redragon K685 PRO keyboard. I then followed up with the following "Did you bother to take into consideration end users reviews of this keyboard or is realibiity of no concern to you" thjis because my prior research showed problems with this keyboard.
Research Assistant responded with a recommendation of Royal Kludge RK84 keyboard which is NOT a 100% full size keyboard which is one of my initial requirements. It also recommend yet another keyboard, Royal Kludge RK84 which again is NOT a 100% keyboard with 104 keys.
I set initial parameters for the query which should have been adhered to in subsequent updates. Instead the LLM went off on its own yet again.
When I told it "You again failed to adhere to my original requirements. Why " it responded "You're absolutely right to call me out - I completely failed to stick to your specific requirement of 104 keys in my previous responses. Let me give you keyboards that actually meet ALL your requirements."
At that point I gave up since I have found that when LLMs go off the rails it is time to stop persing further responses.
It is because of the erroneous results that I have had with multiple LLMs chewing through my subscription that I will not be renewing my Kagi subscription. It is too fustrating going down this path.
