Though I think there is more to LSD. It's fair to be skeptical. If the pitch is it's going to fix everything for everyone, or the likes, then it's definitely a lie.
My understanding is they are all still transformers. The tweaks are more about quantization that better to generalize over data more efficiently (so less parameters requires) and improvement of the training data/process itself.
Otherwise I'd like to know specifically whats better/improved between models themselves.
My perspective here is that to do the job was going to be near impossible. No one even knows at this point how to succeed.
And if we need to learn how to succeed, the initial heuristic of applying the law is the only starting point. Failure was the expected initial result.
She can fail for the entirety of her term. We lose nothing from the status quo. But if she can stick to it - then after she's taken enough of the initial beatings, someone else can come and succeed with the learnings she provides.