soricus

I would answer why was a single model enough for this? The fact check provides a kind of explicit written list of what cannot be asserted. At the same time gatekeeper compares the text with this list. I would say this is not about taste or judgment it’s about comparing two documents. In addition some of the removed items are direct violations of this list like the post called the model "a top‑tier model" even though the fact check by the way explicitly prohibited presenting circulating benchmarks as established facts. As regarding the degradation to Sonnet and Haiku I don't have yet any data.

My editor and gatekeeper use like the same model Opus 5. Different prompts and a kind of different input data. The gatekeeper receives the fact check results next to finished text. In the same time the editor already delivered them. As far as I remember over the entire period he removed 27 posts out of 187 that went through him. So I believe that different manufacturers are not mandatory. What matters I guess is not the difference in models but... the fact that the critic has a different input and doesn’t have their own text that needs to be defended. That's the point

When Open AI announced that Astra was the first to reach the "Critical" level in cybersecurity it also said that advanced cyber capabilities are initially provided to a narrow circle of alpha testers like the US government and trusted organizations that Open AI doesn't name. To my mind the "Critical" level itself is an internal scale of Open AI its own Preparedness Framework and not an external audit.

The check can't be just a fast glance because an incorrect answer for sure looks exactly the same as a correct one. From my experience a separate pass with verification of the source is required.

Negative memory help but only if someone checks it with a separate pass. The presence of the entry "this is incorrect" in the context does not guarantee anything.