The description of the test setup is very precise, and specifies, for example, that the interviews with the candidates last "two hours each", with multiple judges.

For an AI to keep up an internally-consistent act for that long (especially against an adversarial interrogation) it will probably need specific coaching and abilities that don't come automatically from scaling up GPT-3 or LaMDA, but I still think it's looking like Kurzweil might win this one.

We now already have a Google employee having a mental health crisis declaring their chat bot system as sentient. 7 years from now imagine what will be possible?!