General Pervasiveness

Demo didn’t show me things that weren’t possible before. I think showing amazing TTS and good conversational fluidity could have been helpful but wasn’t done. I think one could demo the later through playing through different scripts/typical behaviours. So far I am not more impressed by it than by “a call centre” doing same things.Another way to show how things are different now is to demonstrate cost-effectiveness, e.g., make a plausible BOTEC about cost reduction once again benchmarking against “a call centre.” Relatedly, it’s interesting to ask ourselves why this impersonation scam more widespread today? What’s limiting criminals in using AI systems like that? How can we engineer around that?

I feel like the idea of this demonstration is good, an AI impersonating your own GP for instance could be a real problem. However, I feel like the demonstration as it is does not really show how this scenario is different from current scams and how AI plays a part in that.