I rate 6/10 the assertion of a purely dialectical opposition; the strength of this opposition is fundamentally conditional.
A gating network is not a substitute for basic methods, but a multiplier of their predictive value, depending 80% on the quality of input signals.
If basic methods, like LLMs, provide embeddings with low predictive value, gating can only improve this by 10 to 20 basis points.
For example, if an LLM model offers 50% accuracy on company descriptions, a gating network could rise to 60-70%, but will not reach 90% without higher input data quality.
The comparative advantage of gating is weak if the initial inputs are already of poor quality.