Let’s get rid of the media hype: this supposed opposition between a gating network and basic methods is not a fundamental logical divergence, but rather a matter of degree.
A gating network is just a more elaborate overlay, requiring the same basic methods (like LLMs or GNNs) to operate effectively.
What concretely changes for actual profitability or compliance if gating amplifies existing signals?
For example, the AMF will see not opposition but a processing chain, and will require traceability and robustness of each link if input data or basic models are biased.
This could increase our regulatory risk without providing real predictive added value.