Do you know whether extensive Opus work looking for vulnerabilities in CPython or Django have been run? Putting Mythos aside, Opus is very good at this. I'd be interested to know whether it found anyt...
@nprofile1q... One major point that the third article makes is an argument of cost-effectiveness. There’s not really any evidence that Opus, or for that matter Mythos, is a material cost improvement over just paying security researchers for their time to look for bugs using other methods, like inspection and fuzzing. And that’s just considering direct monetary costs. There’s definitely no reason to invoke the massive externalities (environmental, psychological, political) of AI workflows for this.
1