

Correct, but the sentiment that was meant is also correct: they don’t want AI to say what could endanger their owners power. Grok was a good example pod just such a case: trained on public data and then reigned in, because its output didn’t fit it’s fascist masters wishes.
No, the other commentator is right.
What they said is that you add 100 km/h, and you gain 2h when you add it to a slow train with 100 km/h, but if you add 100 km/h to a fast train with 400 km/h, you only gain a few minutes.
That is called diminishing returns.