Commercial Present Value, of Engineering Paradigms

Folks talking about AI models, seem funny, often enough.

You guys gotta stop throwing model names, and look at the fundamental difference between model structure. The main two aspects of computation are space and time : corresponding to data and operations on data.

2x2 matrix renders :

• - good operation, on good data
• - good operation, on bad data
• - bad operation, on good data
• - bad operation, on bad data

Consider for each model you have named, where it falls on such a matrix in the present SOTA, and where you would expect it to fall on a future SOTA ... now you have a cube of which models are good now, but eventually going to be replaced, and which are poor now, but eventually going to be dominant.