They know how it works. They don't know every value for every variable in a neural layer but that's the point of it. Like you do have to actually code this shit. You have to normalize the data sets, you need to have clear inputs and expected outputs, you have to assemble very large data sets, you have to use the right loss function, you have to reserve a test set ensure the model is actually making predictions.
These guys are tying to fluff up the fact that their chat bot is far more capable than expected and that certain tasks are actually easier to accomplish than expected... like translation because you have clear inputs and outputs and there are large datasets available in the form of the entire internet.
They know how it works. They don't know every value for every variable in a neural layer but that's the point of it. Like you do have to actually code this shit. You have to normalize the data sets, you need to have clear inputs and expected outputs, you have to assemble very large data sets, you have to use the right loss function, you have to reserve a test set ensure the model is actually making predictions.
These guys are tying to fluff up the fact that their chat bot is far more capable than expected and that certain tasks are actually easier to accomplish than expected... like translation because you have clear inputs and outputs and there are large datasets available in the form of the entire internet.
(post is archived)