◇ Could this help me?
Yes. The central claim is correct and the resources named are real and checkable, which is more than most content in this genre offers, and the argument that the category is being judged by its largest commercial examples is a fair one. The two honest qualifications worth carrying alongside it: running a model locally avoids the inference cost of that request but does not undo the training that already happened, and a laptop running a model for an hour is not free, so local is lower impact rather than no impact. Licensed-data certification also covers a small slice of available models today, so it narrows the options considerably rather than being a drop-in filter. General practice, and the useful action is checking the two named lists before picking a model rather than assuming either the worst or the best.
I want to choose AI tools on evidence rather than on vibes. Help me evaluate the specific models and tools I currently use against three questions: what is known about the licensing and consent status of their training data, what published energy or efficiency figures exist for them, and which of my actual tasks could be done by a model running locally on my own hardware. Point me at the published certification lists and energy leaderboards rather than giving me your impression. Then be honest about the limits: tell me which of my questions simply cannot be answered from public information, and where running locally would cost me more in quality or time than it saves in impact.