The guardrails i.e. the classifier would also fire on Fable's output and even Fable's internals (i.e. prior to the shutdown people would see it firing when Fable got angry)
Yes, limiting the full scope of capabilities, which is why I differentiated between Mythos (unrestricted) from guardrailed.
Main point being, we have no idea the measurable capabilities of this, it could be as great or better as unrestricted Mythos, or on par with guardrailed Fable.. or just OpenAI hype that measures up to neither.
The distinction is important because if it truly is a Mythos level (or even guardrailed Fable) it in theory would require that 30 day US government validation before release as well as oversight to the approved partners allowed to use it.
Op was drawing a parallel as to why we should be outraged at the double standard, I was drawing a better parallel by which to compare.
