google has biased my navigation so I see tons of daily articles news like
>((((NLU))))
>(((transformer models)))
Just to find they just do stupid shit like , sentiment analysis, text classification, etc. Using giant and slow models where only the happy path works correctly.
Why is this people forcing their shit, when even an LSTM RNN is not able to learn correctly a simple substitution task like:
>a->x
>b->w
>c->z
And then you find articles like "How I made a chattbot in 100 lines of code using transformers" Just to find that even a kid could reach the same level of performance using if then.
The proof is the Loebner price belongs to a rule based system (mitsuku).
>((((NLU))))
>(((transformer models)))
Just to find they just do stupid shit like , sentiment analysis, text classification, etc. Using giant and slow models where only the happy path works correctly.
Why is this people forcing their shit, when even an LSTM RNN is not able to learn correctly a simple substitution task like:
>a->x
>b->w
>c->z
And then you find articles like "How I made a chattbot in 100 lines of code using transformers" Just to find that even a kid could reach the same level of performance using if then.
The proof is the Loebner price belongs to a rule based system (mitsuku).
