First off, Leo AI is helpful and appreciated when using the Brave Browser.
BUT … some of the answers should be easy to improve/correct when they are blatant errors (math and data not matching). Here’s an example when reviewing IRS tax code: note the math (real number is “additional $1950/person, not $1550”)
My question is how can users help improve the answers that Leo AI generates???
Hello - You can click on the Thumbs down option at the bottom of the summarizer to help provide feedback.
There’s also an option to provide more detailed feedback when using Leo in the sidebar. If you click on the three dots in the top-right of Leo’s answer, you can Dislike the option and add more detailed feedback.
To help improve Leo AI’s answers in Brave, use the feedback button (thumbs up/down) to report inaccuracies, specifying the error and providing the correct information (e.g., “$1950, not $1550”). Alternatively, discuss issues on Brave’s community forum or contact their support team with detailed examples for better AI refinement.
LLMs are horrible at doing arithmetic operations and you are better off avoiding using them for that. They are above average on doing the math (not the arithmetic) for you, but once they do the math just do the arithmetic with a calculator yourself. I am certain that Anthropic and every single company that trains LLMs are already feeding them with a ton of arithmetic operations in their training sets to try and improve them on that front so I highly doubt they actually benefit from any feedback you can submit regarding failed arithmetic. It’s trivial for these companies (and even for users who can program) to generate an infinite stream of failed arithmetic operations so why bother submitting a couple more?
