A Genudi Pioneer approached me with the intent to teach the program Mathematics.

Although likely difficult, the idea is very interesting. Genudi’s algorithm is after all, a strong A.I product.

I must confess that for the sake of having a conversation with Genudi on this website, a few details of the algorithm’s implementation have been biased towards language processing.

The most obvious example is that the beginning of the user’s input is internally prefixed with a hash tag, making the word ‘#Hello’ different from a ‘hello’ encountered later in the query string.

That, of course, doesn’t play well with mathematics, were the number 1 equals 1, regardless of its position in an equation.

I thus came up with a syntax workaround, for the sake of getting Genudi high on home brewed formulas.

To shield the equation values from the language processing bias, I surrounded it in brackets, followed by a punctuation mark (mandatory for language processing). Within the brackets, I wrapped every token with white spaces, to have them treated like individual words.

I used two different patterns:

An assertion pattern: [ x = y + z ].

A querying pattern: [ x + z = ]?

i made a few assertions, asked a few questions, and when Genudi asked something in return, I responded with an equation instead of a plain value. I also explored a few [ x > y ] to make things easier.

By the 20th time I hit [ENTER], Genudi came up with a valid, rather complex assertion:

[ 10 – 1 = 100 – 9 * 10 – 1 * 10 + 9 ].

Looking at the transcript of the full dialog, it’s obvious that there is some luck involved in this result, but Genudi’s current implementation is mostly biased towards processing languages, and from the look of it, it might not take too long before it gives mathematics a real shot.