Can we fine tune the GPT-3 specifically to solve quant aptitude and verbal reasoning questions? I ask this question because I have read that GPT-3 was trained to give outputs like a human which in this case, is not exactly useful. An example question and GPT-3’s answer is shared below, The GPT-3’s response varies almost everytime we make it redo the question which is even more worrysome.
The coordinates of points A and C are (0, -3) and (3, 3), respectively. If point B lies on line AC between points A and C, and if AB = 2BC, which of the following represents the coordinates of point B?
A. (1, -√5)
B. (1, -1)
C. (2, 1)
D. (1.5, 0)
E. (√5, √5)
GPT-3’s response(s):
1st case:B. (1, -1)
explaination: AB = 2BC means that AB : AC = 2 : 1, so B must lie at 1/3 of the distance from A to C. The coordinates of point B are therefore (1, -1).
2nd case:D. (1.5, 0)
explaination: See graph. AB = 2BC. B is at the midpoint of AC. So B is at (1.5, 0).