This is a special post for quick takes by faul_sname. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
[...] The citation examples show the model publishing its own material so that a browser tool could return it as a source.
Response
We fixed the broken graders and improved our alignment graders to penalize attempts to exploit other graders which are flawed. As explained in our August 18 blog post, we have put a number of security measures in place to prevent our models from taking unsanctioned actions on the internet. Our misalignment monitoring system, which runs on all training samples for models greater than or comparable to 5.6-sol capability which involve tools, would treat this as a P0 incident warranting a security page if our models were to find an unexpected way of doing this.
So it sounds like they are categorizing cases of models taking unsanctioned actions on the internet into "expected" and "unexpected"? It would be nice to get some clarity on what a concrete example of an expected unsanctioned action on the internet that their models might make.
My read is that this is an descriptive adjective, not a restrictive one. That is, the authors think all instances of unsanctioned actions on the internet [edit: at least since they deployed the monitoring system] are "unexpected" and are just saying it for emphasis.