Kagi Assistant (available as open beta for Ultimate members) now leverages GPT4-vision model to better understand and describe images. You can test this improved functionality by uploading images or providing image URLs for the Kagi Assistant to analyse.
Edit: A member just emailed saying "the example you have is wrong about 7 of the 10 "Key Points" (only the restaurant name, VAT amount, and amount tendered are correct)." Yes that can be the case with LLMs, we are not trying to present this as grounbreaking, we just integrated a model and the example is as good as the underlying model. This is clearly demonstrating its limitations and it is what it is. We are currently using the best commercially available vision model on the market and it is our desire to emphasize that access to this and other world's cutting edge LLM's are all included in the Assistant with one Kagi subscription. No doubt they will get better in the future.
