Anonymous
6/12/2025, 6:37:41 PM
No.105572823
>>105564850
Gemini 2.5 Pro is actually pretty good. When prompted with
>Write a Python function that takes an array of model values, an array of data values, and a covariance matrix as inputs and calculates the chi-squared cost function value.
it produces code that does the calculation in a numerically stable way without being explicitly told to do so (solving a system of linear equations instead of matrix inversion).
I don't think they publicized the parameter count but since the model is pretty fast on the UI I'd assume it's not that large.
Makes me hopeful about the state of local models given enough time.
Gemini 2.5 Pro is actually pretty good. When prompted with
>Write a Python function that takes an array of model values, an array of data values, and a covariance matrix as inputs and calculates the chi-squared cost function value.
it produces code that does the calculation in a numerically stable way without being explicitly told to do so (solving a system of linear equations instead of matrix inversion).
I don't think they publicized the parameter count but since the model is pretty fast on the UI I'd assume it's not that large.
Makes me hopeful about the state of local models given enough time.